*2561The Secretary of Commerce decided to reinstate a question about citizenship on the 2020 census questionnaire. A group of plaintiffs challenged that decision on constitutional and statutory grounds. We now decide whether the Secretary violated the Enumeration Clause of the Constitution, the Census Act, or otherwise abused his discretion.
I
A
In order to apportion Members of the House of Representatives among the States, the Constitution requires an "Enumeration" of the population every 10 years, to be made "in such Manner" as Congress "shall by Law direct." Art. I, § 2, cl. 3; Amdt. 14, § 2. In the Census Act, Congress delegated to the Secretary of Commerce the task of conducting the decennial census "in such form and content as he may determine." 13 U. S. C. § 141(a). The Secretary is aided in that task by the Census Bureau, a statistical agency housed within the Department of Commerce. See §§ 2, 21.
The population count derived from the census is used not only to apportion representatives but also to allocate federal funds to the States and to draw electoral districts. Wisconsin v. City of New York , 517 U.S. 1, 5-6, 116 S.Ct. 1091, 134 L.Ed.2d 167 (1996). The census additionally serves as a means of collecting demographic information, which "is used for such varied purposes as computing federal grant-in-aid benefits, drafting of legislation, urban and regional planning, business planning, and academic and social studies." Baldrige v. Shapiro , 455 U.S. 345, 353-354, n. 9, 102 S.Ct. 1103, 71 L.Ed.2d 199 (1982). Over the years, the census has asked questions about (for example) race, sex, age, health, education, occupation, housing, and military service. It has also asked about radio ownership, age at first marriage, and native tongue. The Census Act obliges everyone to answer census questions truthfully and requires the Secretary to keep individual answers confidential, including from other Government agencies. §§ 221, 8(b), 9(a).
There have been 23 decennial censuses from the first census in 1790 to the most recent in 2010. Every census between 1820 and 2000 (with the exception of 1840) asked at least some of the population about their citizenship or place of birth. Between 1820 and 1950, the question was asked of all households. Between 1960 and 2000, it was asked of about one-fourth to one-sixth of the population. That change was part of a larger effort to simplify the census by asking most people a few basic demographic questions (such as sex, age, race, and marital status) on a short-form questionnaire, while asking a sample of the population more detailed demographic questions on a long-form questionnaire. In explaining the decision to move the citizenship question to the long-form questionnaire, the Census Bureau opined that "general census information on citizenship had become of less importance compared with other possible questions to be included in the census, particularly in view of the *2562recent statutory requirement for annual alien registration which could provide the Immigration and Naturalization Service, the principal user of such data, with the information it needed." Dept. of Commerce, Bureau of Census, 1960 Censuses of Population and Housing 194 (1966).1
In 2010, the year of the latest census, the format changed again. All households received the same questionnaire, which asked about sex, age, race, Hispanic origin, and living arrangements. The more detailed demographic questions previously asked on the long-form questionnaire, including the question about citizenship, were instead asked in the American Community Survey (or ACS), which is sent each year to a rotating sample of about 2.6% of households.
The Census Bureau and former Bureau officials have resisted occasional proposals to resume asking a citizenship question of everyone, on the ground that doing so would discourage noncitizens from responding to the census and lead to a less accurate count of the total population. See, e.g. , Federation of Am. Immigration Reform v. Klutznick , 486 F.Supp. 564, 568 (DDC 1980) ("[A]ccording to the Bureau[,] any effort to ascertain citizenship will inevitably jeopardize the overall accuracy of the population count"); Brief for Former Directors of the U. S. Census Bureau as Amici Curiae in Evenwel v. Abbott , O. T. 2014, No. 14-940, p. 25 (inquiring about citizenship would "invariably lead to a lower response rate").
B
In March 2018, Secretary of Commerce Wilbur Ross announced in a memo that he had decided to reinstate a question about citizenship on the 2020 decennial census questionnaire. The Secretary stated that he was acting at the request of the Department of Justice (DOJ), which sought improved data about citizen voting-age population for purposes of enforcing the Voting Rights Act (or VRA)-specifically the Act's ban on diluting the influence of minority voters by depriving them of single-member districts in which they can elect their preferred candidates. App. to Pet. for Cert. 548a. DOJ explained that federal courts determine whether a minority group could constitute a majority in a particular district by looking to the citizen voting-age population of the group. According to DOJ, the existing citizenship data from the American Community Survey was not ideal: It was not reported at the level of the census block, the basic component of legislative districting plans; it had substantial margins of error; and it did not align in time with the census-based population counts used to draw legislative districts. DOJ therefore formally requested reinstatement of the citizenship question on the census questionnaire. Id. , at 565a-569a.
The Secretary's memo explained that the Census Bureau initially analyzed, and the Secretary considered, three possible courses of action. The first was to continue to collect citizenship information in the American Community Survey and attempt to develop a data model that would more accurately estimate citizenship at the census block level. The Secretary rejected that option because the Bureau "did not assert and could not confirm" that such ACS-based data modeling was possible "with a sufficient degree of accuracy." Id. , at 551a.
*2563The second option was to reinstate a citizenship question on the decennial census. The Bureau predicted that doing so would discourage some noncitizens from responding to the census. That would necessitate increased "non-response follow up" operations-procedures the Bureau uses to attempt to count people who have not responded to the census-and potentially lead to a less accurate count of the total population.
Option three was to use administrative records from other agencies, such as the Social Security Administration and Citizenship and Immigration Services, to provide DOJ with citizenship data. The Census Bureau recommended this option, and the Secretary found it a "potentially appealing solution" because the Bureau has long used administrative records to supplement and improve census data. Id. , at 554a. But the Secretary concluded that administrative records alone were inadequate because they were missing for more than 10% of the population.
The Secretary ultimately asked the Census Bureau to develop a fourth option that would combine options two and three: reinstate a citizenship question on the census questionnaire, and also use the time remaining until the 2020 census to "further enhance" the Bureau's "administrative record data sets, protocols, and statistical models." Id. , at 555a. The memo explained that, in the Secretary's judgment, the fourth option would provide DOJ with the "most complete and accurate" citizen voting-age population data in response to its request. Id. , at 556a.
The Secretary "carefully considered" the possibility that reinstating a citizenship question would depress the response rate. Ibid . But after evaluating the Bureau's "limited empirical evidence" on the question-evidence drawn from estimated non-response rates to previous American Community Surveys and census questionnaires-the Secretary concluded that it was not possible to "determine definitively" whether inquiring about citizenship in the census would materially affect response rates. Id. , at 557a, 562a. He also noted the long history of the citizenship question on the census, as well as the facts that the United Nations recommends collecting census-based citizenship information, and other major democracies such as Australia, Canada, France, Indonesia, Ireland, Germany, Mexico, Spain, and the United Kingdom inquire about citizenship in their censuses. Altogether, the Secretary determined that "the need for accurate citizenship data and the limited burden that the reinstatement of the citizenship question would impose outweigh fears about a potentially lower response rate." Id. , at 557a.
C
Shortly after the Secretary announced his decision, two groups of plaintiffs filed suit in Federal District Court in New York, challenging the decision on several grounds. The first group of plaintiffs included 18 States, the District of Columbia, various counties and cities, and the United States Conference of Mayors. They alleged that the Secretary's decision violated the Enumeration Clause of the Constitution and the requirements of the Administrative Procedure Act. The second group of plaintiffs consisted of several non-governmental organizations that work with immigrant and minority communities. They added an equal protection claim. The District Court consolidated the two cases. Both groups of plaintiffs are respondents here.
The Government moved to dismiss the lawsuits, arguing that the Secretary's decision was unreviewable and that respondents had failed to state cognizable claims *2564under the Enumeration Clause and the Equal Protection Clause. The District Court dismissed the Enumeration Clause claim but allowed the other claims to proceed. 315 F.Supp.3d 766 (SDNY 2018).
In June 2018, the Government submitted to the District Court the Commerce Department's "administrative record": the materials that Secretary Ross considered in making his decision. That record included DOJ's December 2017 letter requesting reinstatement of the citizenship question, as well as several memos from the Census Bureau analyzing the predicted effects of reinstating the question. Shortly thereafter, at DOJ's urging, the Government supplemented the record with a new memo from the Secretary, "intended to provide further background and context regarding" his March 2018 memo. App. to Pet. for Cert. 546a. The supplemental memo stated that the Secretary had begun considering whether to add the citizenship question in early 2017, and had inquired whether DOJ "would support, and if so would request, inclusion of a citizenship question as consistent with and useful for enforcement of the Voting Rights Act." Ibid. According to the Secretary, DOJ "formally" requested reinstatement of the citizenship question after that inquiry. Ibid.
Respondents argued that the supplemental memo indicated that the Government had submitted an incomplete record of the materials considered by the Secretary. They asked the District Court to compel the Government to complete the administrative record. The court granted that request, and the parties jointly stipulated to the inclusion of more than 12,000 pages of additional materials in the administrative record. Among those materials were emails and other records confirming that the Secretary and his staff began exploring the possibility of reinstating a citizenship question shortly after he was confirmed in early 2017, attempted to elicit requests for citizenship data from other agencies, and eventually persuaded DOJ to request reinstatement of the question for VRA enforcement purposes.
In addition, respondents asked the court to authorize discovery outside the administrative record. They claimed that such an unusual step was warranted because they had made a strong preliminary showing that the Secretary had acted in bad faith. See Citizens to Preserve Overton Park, Inc. v. Volpe , 401 U.S. 402, 420, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971). The court also granted that request, authorizing expert discovery and depositions of certain DOJ and Commerce Department officials.
In August and September 2018, the District Court issued orders compelling depositions of Secretary Ross and of the Acting Assistant Attorney General for DOJ's Civil Rights Division. We granted the Government's request to stay the Secretary's deposition pending further review, but we declined to stay the Acting AAG's deposition or the other extra-record discovery that the District Court had authorized.
The District Court held a bench trial and issued findings of fact and conclusions of law on respondents' statutory and equal protection claims. After determining that respondents had standing to sue, the District Court ruled that the Secretary's action was arbitrary and capricious, based on a pretextual rationale, and violated certain provisions of the Census Act. On the equal protection claim, however, the District Court concluded that respondents had not met their burden of showing that the Secretary was motivated by discriminatory animus. The court granted judgment to respondents on their statutory claims, vacated the Secretary's decision, and enjoined him from reinstating the citizenship question until he cured the legal errors the *2565court had identified. 351 F.Supp.3d 502 (SDNY 2019).
The Government appealed to the Second Circuit, but also filed a petition for writ of certiorari before judgment, asking this Court to review the District Court's decision directly because the case involved an issue of imperative public importance, and the census questionnaire needed to be finalized for printing by the end of June 2019. We granted the petition. 586 U. S. ----, 139 S.Ct. 16, 202 L.Ed.2d 306 (2019). At the Government's request, we later ordered the parties to address whether the Enumeration Clause provided an alternative basis to affirm. 586 U. S. ----, 139 S.Ct. 16, 202 L.Ed.2d 306 (2019).
II
We begin with jurisdiction. Article III of the Constitution limits federal courts to deciding "Cases" and "Controversies." For a legal dispute to qualify as a genuine case or controversy, at least one plaintiff must have standing to sue. The doctrine of standing "limits the category of litigants empowered to maintain a lawsuit in federal court to seek redress for a legal wrong" and "confines the federal courts to a properly judicial role." Spokeo , Inc. v. Robins , 578 U. S. ----, ----, 136 S.Ct. 1540, 1547, 194 L.Ed.2d 635 (2016). To have standing, a plaintiff must "present an injury that is concrete, particularized, and actual or imminent; fairly traceable to the defendant's challenged behavior; and likely to be redressed by a favorable ruling." Davis v. Federal Election Comm'n , 554 U.S. 724, 733, 128 S.Ct. 2759, 171 L.Ed.2d 737 (2008).
Respondents assert a number of injuries-diminishment of political representation, loss of federal funds, degradation of census data, and diversion of resources-all of which turn on their expectation that reinstating a citizenship question will depress the census response rate and lead to an inaccurate population count. Several States with a disproportionate share of noncitizens, for example, anticipate losing a seat in Congress or qualifying for less federal funding if their populations are undercounted. These are primarily future injuries, which "may suffice if the threatened injury is certainly impending, or there is a substantial risk that the harm will occur." Susan B. Anthony List v. Driehaus , 573 U.S. 149, 158, 134 S.Ct. 2334, 189 L.Ed.2d 246 (2014) (internal quotation marks omitted).
The District Court concluded that the evidence at trial established a sufficient likelihood that the reinstatement of a citizenship question would result in noncitizen households responding to the census at lower rates than other groups, which in turn would cause them to be undercounted and lead to many of respondents' asserted injuries. For purposes of standing, these findings of fact were not so suspect as to be clearly erroneous.
We therefore agree that at least some respondents have Article III standing. Several state respondents here have shown that if noncitizen households are undercounted by as little as 2%-lower than the District Court's 5.8% prediction-they will lose out on federal funds that are distributed on the basis of state population. That is a sufficiently concrete and imminent injury to satisfy Article III, and there is no dispute that a ruling in favor of respondents would redress that harm.
The Government contends, however, that any harm to respondents is not fairly traceable to the Secretary's decision, because such harm depends on the independent action of third parties choosing to violate their legal duty to respond to the census. The chain of causation is made *2566even more tenuous, the Government argues, by the fact that such intervening, unlawful third-party action would be motivated by unfounded fears that the Federal Government will itself break the law by using noncitizens' answers against them for law enforcement purposes. The Government invokes our steady refusal to "endorse standing theories that rest on speculation about the decisions of independent actors," Clapper v. Amnesty Int'l USA , 568 U.S. 398, 414, 133 S.Ct. 1138, 185 L.Ed.2d 264 (2013), particularly speculation about future unlawful conduct, Los Angeles v. Lyons , 461 U.S. 95, 105, 103 S.Ct. 1660, 75 L.Ed.2d 675 (1983).
But we are satisfied that, in these circumstances, respondents have met their burden of showing that third parties will likely react in predictable ways to the citizenship question, even if they do so unlawfully and despite the requirement that the Government keep individual answers confidential. The evidence at trial established that noncitizen households have historically responded to the census at lower rates than other groups, and the District Court did not clearly err in crediting the Census Bureau's theory that the discrepancy is likely attributable at least in part to noncitizens' reluctance to answer a citizenship question. Respondents' theory of standing thus does not rest on mere speculation about the decisions of third parties; it relies instead on the predictable effect of Government action on the decisions of third parties. Cf. Bennett v. Spear , 520 U.S. 154, 169-170, 117 S.Ct. 1154, 137 L.Ed.2d 281 (1997) ; Davis , 554 U.S. at 734-735, 128 S.Ct. 2759. Because Article III "requires no more than de facto causality," Block v. Meese , 793 F.2d 1303, 1309 (CADC 1986) (Scalia, J.), traceability is satisfied here. We may therefore consider the merits of respondents' claims, at least as far as the Constitution is concerned.
III
The Enumeration Clause of the Constitution does not provide a basis to set aside the Secretary's decision. The text of that clause "vests Congress with virtually unlimited discretion in conducting the decennial 'actual Enumeration,' " and Congress "has delegated its broad authority over the census to the Secretary." Wisconsin , 517 U.S. at 19, 116 S.Ct. 1091. Given that expansive grant of authority, we have rejected challenges to the conduct of the census where the Secretary's decisions bore a "reasonable relationship to the accomplishment of an actual enumeration." Id. , at 20, 116 S.Ct. 1091.
Respondents ask us to evaluate the Secretary's decision to reinstate a citizenship question under that "reasonable relationship" standard, but we agree with the District Court that a different analysis is needed here. Our cases applying that standard concerned decisions about the population count itself-such as a postcensus decision not to use a particular method to adjust an undercount, id. , at 4, 116 S.Ct. 1091, and a decision to allocate overseas military personnel to their home States, Franklin v. Massachusetts , 505 U.S. 788, 790-791, 112 S.Ct. 2767, 120 L.Ed.2d 636 (1992). We have never applied the standard to decisions about what kinds of demographic information to collect in the course of taking the census. Indeed, as the District Court recognized, applying the "reasonable relationship" standard to every census-related decision "would lead to the conclusion that it is unconstitutional to ask any demographic question on the census" because "asking such questions bears no relationship whatsoever to the goal of an accurate headcount." 315 F.Supp.3d at 804-805. Yet demographic questions have been asked in every census since 1790, and questions about citizenship in particular *2567have been asked for nearly as long. Like the District Court, we decline respondents' invitation to measure the constitutionality of the citizenship question by a standard that would seem to render every census since 1790 unconstitutional.
We look instead to Congress's broad authority over the census, as informed by long and consistent historical practice. All three branches of Government have understood the Constitution to allow Congress, and by extension the Secretary, to use the census for more than simply counting the population. Since 1790, Congress has sought, or permitted the Secretary to seek, information about matters as varied as age, sex, marital status, health, trade, profession, literacy, and value of real estate owned. See id. , at 801. Since 1820, it has sought, or permitted the Secretary to seek, information about citizenship in particular. Federal courts have approved the practice of collecting demographic data in the census. See, e.g. , United States v. Moriarity , 106 F. 886, 891 (CC SDNY 1901) (duty to take a census of population "does not prohibit the gathering of other statistics, if 'necessary and proper,' for the intelligent exercise of other powers enumerated in the constitution"). While we have never faced the question directly, we have assumed that Congress has the power to use the census for information-gathering purposes, see Legal Tender Cases , 12 Wall. 457, 536, 20 L.Ed. 287 (1871), and we have recognized the role of the census as a "linchpin of the federal statistical system by collecting data on the characteristics of individuals, households, and housing units throughout the country," Department of Commerce v. United States House of Representatives , 525 U.S. 316, 341, 119 S.Ct. 765, 142 L.Ed.2d 797 (1999) (internal quotation marks omitted).
That history matters. Here, as in other areas, our interpretation of the Constitution is guided by a Government practice that "has been open, widespread, and unchallenged since the early days of the Republic." NLRB v. Noel Canning , 573 U.S. 513, 572, 134 S.Ct. 2550, 189 L.Ed.2d 538 (2014) (Scalia, J., concurring in judgment); see also Wisconsin , 517 U.S. at 21, 116 S.Ct. 1091 (noting "importance of historical practice" in census context). In light of the early understanding of and long practice under the Enumeration Clause, we conclude that it permits Congress, and by extension the Secretary, to inquire about citizenship on the census questionnaire. We need not, and do not, decide the constitutionality of any other question that Congress or the Secretary might decide to include in the census.
IV
The District Court set aside the Secretary's decision to reinstate a citizenship question on the grounds that the Secretary acted arbitrarily and violated certain provisions of the Census Act. The Government contests those rulings, but also argues that the Secretary's decision was not judicially reviewable under the Administrative Procedure Act in the first place. We begin with that contention.
A
The Administrative Procedure Act embodies a "basic presumption of judicial review," Abbott Laboratories v. Gardner , 387 U.S. 136, 140, 87 S.Ct. 1507, 18 L.Ed.2d 681 (1967), and instructs reviewing courts to set aside agency action that is "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law," 5 U. S. C. § 706(2)(A). Review is not available, however, "to the extent that" a relevant statute precludes it, § 701(a)(1), or the agency action is "committed to agency discretion by law," § 701(a)(2). The Government argues that the Census Act *2568commits to the Secretary's unreviewable discretion decisions about what questions to include on the decennial census questionnaire.
We disagree. To be sure, the Act confers broad authority on the Secretary. Section 141(a) instructs him to take "a decennial census of population" in "such form and content as he may determine, including the use of sampling procedures and special surveys." 13 U. S. C. § 141. The Act defines "census of population" to mean "a census of population, housing, and matters relating to population and housing," § 141(g), and it authorizes the Secretary, in "connection with any such census," to "obtain such other census information as necessary," § 141(a). It also states that the "Secretary shall prepare questionnaires, and shall determine the inquiries, and the number, form, and subdivisions thereof, for the statistics, surveys, and censuses provided for in this title." § 5. And it authorizes him to acquire materials, such as administrative records, from other federal, state, and local agencies in aid of conducting the census. § 6. Those provisions leave much to the Secretary's discretion. See Wisconsin , 517 U.S. at 19, 116 S.Ct. 1091 ("Through the Census Act, Congress has delegated its broad authority over the census to the Secretary.").
But they do not leave his discretion unbounded. In order to give effect to the command that courts set aside agency action that is an abuse of discretion, and to honor the presumption of judicial review, we have read the § 701(a)(2) exception for action committed to agency discretion "quite narrowly, restricting it to 'those rare circumstances where the relevant statute is drawn so that a court would have no meaningful standard against which to judge the agency's exercise of discretion.' " Weyerhaeuser Co. v. United States Fish and Wildlife Serv. , 586 U. S. ----, ----, 139 S.Ct. 361, 370, 202 L.Ed.2d 269 (2018) (quoting Lincoln v. Vigil , 508 U.S. 182, 191, 113 S.Ct. 2024, 124 L.Ed.2d 101 (1993) ). And we have generally limited the exception to "certain categories of administrative decisions that courts traditionally have regarded as 'committed to agency discretion,' " id. , at 191, 113 S.Ct. 2024, such as a decision not to institute enforcement proceedings, Heckler v. Chaney , 470 U.S. 821, 831-832, 105 S.Ct. 1649, 84 L.Ed.2d 714 (1985), or a decision by an intelligence agency to terminate an employee in the interest of national security, Webster v. Doe , 486 U.S. 592, 600-601, 108 S.Ct. 2047, 100 L.Ed.2d 632 (1988).
The taking of the census is not one of those areas traditionally committed to agency discretion. We and other courts have entertained both constitutional and statutory challenges to census-related decisionmaking. See, e.g. , Department of Commerce , 525 U.S. 316, 119 S.Ct. 765, 142 L.Ed.2d 797 ; Wisconsin , 517 U.S. 1, 116 S.Ct. 1091, 134 L.Ed.2d 167 ; Carey v. Klutznick , 637 F.2d 834 (CA2 1980).
Nor is the statute here drawn so that it furnishes no meaningful standard by which to judge the Secretary's action. In contrast to the National Security Act in Webster , which gave the Director of Central Intelligence discretion to terminate employees whenever he "deem[ed]" it "advisable," 486 U.S. at 594, 108 S.Ct. 2047, the Census Act constrains the Secretary's authority to determine the form and content of the census in a number of ways. Section 195, for example, governs the extent to which he can use statistical sampling. Section 6(c), which will be considered in more detail below, circumscribes his power in certain circumstances to collect information through direct inquiries when administrative records are available. More generally, by mandating a population count that will be used to apportion representatives, *2569see § 141(b), 2 U. S. C. § 2a, the Act imposes "a duty to conduct a census that is accurate and that fairly accounts for the crucial representational rights that depend on the census and the apportionment." Franklin , 505 U.S. at 819-820, 112 S.Ct. 2767 (Stevens, J., concurring in part and concurring in judgment).
The Secretary's decision to reinstate a citizenship question is amenable to review for compliance with those and other provisions of the Census Act, according to the general requirements of reasoned agency decisionmaking. Because this is not a case in which there is "no law to apply," Overton Park , 401 U.S. at 410, 91 S.Ct. 814, the Secretary's decision is subject to judicial review.
B
At the heart of this suit is respondents' claim that the Secretary abused his discretion in deciding to reinstate a citizenship question. We review the Secretary's exercise of discretion under the deferential "arbitrary and capricious" standard. See 5 U. S. C. § 706(2)(A). Our scope of review is "narrow": we determine only whether the Secretary examined "the relevant data" and articulated "a satisfactory explanation" for his decision, "including a rational connection between the facts found and the choice made." Motor Vehicle Mfrs. Assn. of United States, Inc. v. State Farm Mut. Automobile Ins. Co. , 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (internal quotation marks omitted). We may not substitute our judgment for that of the Secretary, ibid. , but instead must confine ourselves to ensuring that he remained "within the bounds of reasoned decisionmaking," Baltimore Gas & Elec. Co. v. Natural Resources Defense Council, Inc. , 462 U.S. 87, 105, 103 S.Ct. 2246, 76 L.Ed.2d 437 (1983).
The District Court set aside the Secretary's decision for two independent reasons: His course of action was not supported by the evidence before him, and his stated rationale was pretextual. We focus on the first point here and take up the question of pretext later.
The Secretary examined the Bureau's analysis of various ways to collect improved citizenship data and explained why he thought the best course was to both reinstate a citizenship question and use citizenship data from administrative records to fill in the gaps. He considered but rejected the Bureau's recommendation to use administrative records alone. As he explained, records are lacking for about 10% of the population, so the Bureau would still need to estimate citizenship for millions of voting-age people. Asking a citizenship question of everyone, the Secretary reasoned, would eliminate the need to estimate citizenship for many of those people. And supplementing census responses with administrative record data would help complete the picture and allow the Bureau to better estimate citizenship for the smaller set of cases where it was still necessary to do so.
The evidence before the Secretary supported that decision. As the Bureau acknowledged, each approach-using administrative records alone, or asking about citizenship and using records to fill in the gaps-entailed tradeoffs between accuracy and completeness. Without a citizenship question, the Bureau would need to estimate the citizenship of about 35 million people; with a citizenship question, it would need to estimate the citizenship of only 13.8 million. Under either approach, there would be some errors in both the administrative records and the Bureau's estimates. With a citizenship question, there would also be some erroneous self-responses (about 500,000) and some conflicts *2570between responses and administrative record data (about 9.5 million).
The Bureau explained that the "relative quality" of the citizenship data generated by each approach would depend on the "relative importance of the errors" in each, but it was not able to "quantify the relative magnitude of the errors across the alternatives." App. 148. The Bureau nonetheless recommended using administrative records alone because it had "high confidence" that it could develop an accurate model for estimating the citizenship of the 35 million people for whom administrative records were not available, and it thought the resulting citizenship data would be of superior quality. Id. , at 146, 158-159. But when the time came for the Secretary to make a decision, the model did not yet exist, and even if it had, there was no way to gauge its relative accuracy. As the Bureau put it, "we will most likely never possess a fully adequate truth deck to benchmark" the model-which appears to be bureaucratese for "maybe, maybe not." Id. , at 146. The Secretary opted instead for the approach that would yield a more complete set of data at an acceptable rate of accuracy, and would require estimating the citizenship of fewer people.
The District Court overruled that choice, agreeing with the Bureau's assessment that its recommended approach would yield higher quality citizenship data on the whole. But the choice between reasonable policy alternatives in the face of uncertainty was the Secretary's to make. He considered the relevant factors, weighed risks and benefits, and articulated a satisfactory explanation for his decision. In overriding that reasonable exercise of discretion, the court improperly substituted its judgment for that of the agency.
The Secretary then weighed the benefit of collecting more complete and accurate citizenship data against the risk that inquiring about citizenship would depress census response rates, particularly among noncitizen households. In the Secretary's view, that risk was difficult to assess. The Bureau predicted a 5.1% decline in response rates among noncitizen households if the citizenship question were reinstated.2 It relied for that prediction primarily on studies showing that, while noncitizens had responded at lower rates than citizens to the 2000 short-form and 2010 censuses, which did not ask about citizenship, they responded at even lower rates than citizens to the 2000 long-form census and the 2010 American Community Survey, which did ask about citizenship. The Bureau thought it was reasonable to infer that the citizenship question accounted for the differential decline in noncitizen responses. But, the Secretary explained, the Bureau was unable to rule out other causes. For one thing, the evidence before the Secretary suggested that noncitizen households tend to be more distrustful of, and less likely to respond to, any government effort to collect information. For another, both the 2000 long-form census and 2010 ACS asked over 45 questions on a range of topics, including employment, income, and housing characteristics. Noncitizen households might disproportionately fail to respond to a lengthy and intrusive Government questionnaire for a number of reasons besides reluctance to answer a citizenship question-reasons relating to education level, socioeconomic status, and less exposure to Government outreach efforts. See App. to Pet. for Cert. 553a-554a, 557a-558a.
*2571The Secretary justifiably found the Bureau's analysis inconclusive. Weighing that uncertainty against the value of obtaining more complete and accurate citizenship data, he determined that reinstating a citizenship question was worth the risk of a potentially lower response rate. That decision was reasonable and reasonably explained, particularly in light of the long history of the citizenship question on the census.
Justice BREYER would conclude otherwise, but only by subordinating the Secretary's policymaking discretion to the Bureau's technocratic expertise. Justice BREYER's analysis treats the Bureau's (pessimistic) prediction about response rates and (optimistic) assumptions about its data modeling abilities as touchstones of substantive reasonableness rather than simply evidence for the Secretary to consider. He suggests that the Secretary should have deferred to the Bureau or at least offered some special justification for drawing his own inferences and adopting his own assumptions. But the Census Act authorizes the Secretary, not the Bureau, to make policy choices within the range of reasonable options. And the evidence before the Secretary hardly led ineluctably to just one reasonable course of action. It called for value-laden decisionmaking and the weighing of incommensurables under conditions of uncertainty. The Secretary was required to consider the evidence and give reasons for his chosen course of action. He did so. It is not for us to ask whether his decision was "the best one possible" or even whether it was "better than the alternatives." FERC v. Electric Power Supply Assn. , 577 U. S. ----, ----, 136 S.Ct. 760, 782, 193 L.Ed.2d 661 (2016). By second-guessing the Secretary's weighing of risks and benefits and penalizing him for departing from the Bureau's inferences and assumptions, Justice BREYER-like the District Court-substitutes his judgment for that of the agency.
C
The District Court also ruled that the Secretary violated two particular provisions of the Census Act, § 6(c) and § 141(f).
Section 6 has three subsections. Subsections (a) and (b) authorize the Secretary to acquire administrative records from other federal agencies and from state and local governments.3 Subsection (c) states:
"To the maximum extent possible and consistent with the kind, timeliness, quality and scope of the statistics required, the Secretary shall acquire and use information available from any source referred to in subsection (a) or (b) of this section instead of conducting direct inquiries." 13 U. S. C. § 6(c).
The District Court held, and respondents argue, that the Secretary failed to comply with § 6(c) because he opted to collect citizenship data using direct inquiries when it was possible to provide DOJ with data from administrative records alone.
*2572At the outset, § 6(c) may not even apply here. It governs the Secretary's choices with respect to "statistics required." The parties have assumed that phrase refers to census-related data that the Secretary wishes to acquire, but it may instead refer to particular kinds of statistics that other provisions of the Census Act actually do require the Secretary to collect and publish. See, e.g. , § 41 ("The Secretary shall collect and publish statistics concerning [cotton and cotton production]."); § 61 ("The Secretary shall collect, collate, and publish monthly statistics concerning [vegetable and animal oils and the like]."); § 91 ("The Secretary shall collect and publish quarterly financial statistics of business operations, organization, practices, management, and relation to other businesses."). If so, § 6(c) would seem to have nothing to say about the Secretary's collection of census-related citizenship data, which is not a "statistic" he is "required" to collect.
Regardless, assuming the provision applies, the Secretary complied with it, for essentially the same reasons that his decision was not arbitrary and capricious. As he explained, administrative records would not, in his judgment, provide the more complete and accurate data that DOJ sought. He thus could not, "consistent with" the kind and quality of the "statistics required," use administrative records instead of asking about citizenship directly. Respondents' arguments to the contrary rehash their disagreement with the Secretary's policy judgment about which approach would yield the most complete and accurate citizenship data. For the reasons already discussed, we may not substitute our judgment for that of the Secretary here.
We turn now to § 141(f), which requires the Secretary to report to Congress about his plans for the census. Paragraph (1) instructs him to submit, at least three years before the census date, a report containing his "determination of the subjects proposed to be included, and the types of information to be compiled," in the census. Paragraph (2) then tells him to submit, at least two years before the census date, a report containing his "determination of the questions proposed to be included" in the census. Paragraph (3) provides:
"[A]fter submission of a report under paragraph (1) or (2) of this subsection and before the appropriate census date, if the Secretary finds new circumstances exist which necessitate that the subjects, types of information, or questions contained in reports so submitted be modified, [he shall submit] a report containing the Secretary's determination of the subjects, types of information, or questions as proposed to be modified."
The Secretary timely submitted his paragraph (1) report in March 2017. It did not mention citizenship. In December 2017, he received DOJ's formal request. Three months later, in March 2018, he timely submitted his paragraph (2) report. It did propose asking a question about citizenship.
The District Court held that the Secretary's failure to mention citizenship in his March 2017 report violated § 141(f)(1) and provided an independent basis to set aside his action. Assuming without deciding that the Secretary's compliance with the reporting requirement is for courts-rather than Congress-to police, we disagree. The Secretary's March 2018 report satisfied the requirements of paragraph (3): By informing Congress that he proposed to include a citizenship question, the Secretary necessarily also informed Congress that he proposed to modify the original list of subjects that he submitted in the March 2017 report. Nothing *2573in § 141(f) suggests that the same report cannot simultaneously fulfill the requirements of paragraphs (2) and (3). And to the extent paragraph (3) requires the Secretary to explain his finding of new circumstances, he did so in his March 2018 memo, which described DOJ's intervening request.
In any event, even if we agreed with the District Court that the Secretary technically violated § 141(f) by submitting a paragraph (2) report that doubled as a paragraph (3) report, the error would surely be harmless in these circumstances, where the Secretary nonetheless fully informed Congress of, and explained, his decision. See 5 U. S. C. § 706 (in reviewing agency action, "due account shall be taken of the rule of prejudicial error").
V
We now consider the District Court's determination that the Secretary's decision must be set aside because it rested on a pretextual basis, which the Government conceded below would warrant a remand to the agency.
We start with settled propositions. First, in order to permit meaningful judicial review, an agency must "disclose the basis" of its action. Burlington Truck Lines, Inc. v. United States , 371 U.S. 156, 167-169, 83 S.Ct. 239, 9 L.Ed.2d 207 (1962) (internal quotation marks omitted); see also SEC v. Chenery Corp. , 318 U.S. 80, 94, 63 S.Ct. 454, 87 L.Ed. 626 (1943) ("[T]he orderly functioning of the process of review requires that the grounds upon which the administrative agency acted be clearly disclosed and adequately sustained.").
Second, in reviewing agency action, a court is ordinarily limited to evaluating the agency's contemporaneous explanation in light of the existing administrative record. Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc. , 435 U.S. 519, 549, 98 S.Ct. 1197, 55 L.Ed.2d 460 (1978) ; Camp v. Pitts , 411 U.S. 138, 142-143, 93 S.Ct. 1241, 36 L.Ed.2d 106 (1973) (per curiam ). That principle reflects the recognition that further judicial inquiry into "executive motivation" represents "a substantial intrusion" into the workings of another branch of Government and should normally be avoided. Arlington Heights v. Metropolitan Housing Development Corp. , 429 U.S. 252, 268, n. 18, 97 S.Ct. 555, 50 L.Ed.2d 450 (1977) ; see Overton Park , 401 U.S. at 420, 91 S.Ct. 814.
Third, a court may not reject an agency's stated reasons for acting simply because the agency might also have had other unstated reasons. See Jagers v. Federal Crop Ins. Corp. , 758 F.3d 1179, 1185-1186 (CA10 2014) (rejecting argument that "the agency's subjective desire to reach a particular result must necessarily invalidate the result, regardless of the objective evidence supporting the agency's conclusion"). Relatedly, a court may not set aside an agency's policymaking decision solely because it might have been influenced by political considerations or prompted by an Administration's priorities. Agency policymaking is not a "rarified technocratic process, unaffected by political considerations or the presence of Presidential power." Sierra Club v. Costle , 657 F.2d 298, 408 (CADC 1981). Such decisions are routinely informed by unstated considerations of politics, the legislative process, public relations, interest group relations, foreign relations, and national security concerns (among others).
Finally, we have recognized a narrow exception to the general rule against inquiring into "the mental processes of administrative decisionmakers." Overton Park , 401 U.S. at 420, 91 S.Ct. 814. On a *2574"strong showing of bad faith or improper behavior," such an inquiry may be warranted and may justify extra-record discovery. Ibid.
The District Court invoked that exception in ordering extra-record discovery here. Although that order was premature, we think it was ultimately justified in light of the expanded administrative record. Recall that shortly after this litigation began, the Secretary, prodded by DOJ, filed a supplemental memo that added new, pertinent information to the administrative record. The memo disclosed that the Secretary had been considering the citizenship question for some time and that Commerce had inquired whether DOJ would formally request reinstatement of the question. That supplemental memo prompted respondents to move for both completion of the administrative record and extra-record discovery. The District Court granted both requests at the same hearing, agreeing with respondents that the Government had submitted an incomplete administrative record and that the existing evidence supported a prima facie showing that the VRA rationale was pretextual.
The Government did not challenge the court's conclusion that the administrative record was incomplete, and the parties stipulated to the inclusion of more than 12,000 pages of internal deliberative materials as part of the administrative record, materials that the court later held were sufficient on their own to demonstrate pretext. The Government did, however, challenge the District Court's order authorizing extra-record discovery, as well as the court's later orders compelling depositions of the Secretary and of the Acting Assistant Attorney General for DOJ's Civil Rights Division.
We agree with the Government that the District Court should not have ordered extra-record discovery when it did. At that time, the most that was warranted was the order to complete the administrative record. But the new material that the parties stipulated should have been part of the administrative record-which showed, among other things, that the VRA played an insignificant role in the decisionmaking process-largely justified such extra-record discovery as occurred (which did not include the deposition of the Secretary himself). We accordingly review the District Court's ruling on pretext in light of all the evidence in the record before the court, including the extra-record discovery.
That evidence showed that the Secretary was determined to reinstate a citizenship question from the time he entered office; instructed his staff to make it happen; waited while Commerce officials explored whether another agency would request census-based citizenship data; subsequently contacted the Attorney General himself to ask if DOJ would make the request; and adopted the Voting Rights Act rationale late in the process. In the District Court's view, this evidence established that the Secretary had made up his mind to reinstate a citizenship question "well before" receiving DOJ's request, and did so for reasons unknown but unrelated to the VRA. 351 F.Supp.3d at 660.
The Government, on the other hand, contends that there was nothing objectionable or even surprising in this. And we agree-to a point. It is hardly improper for an agency head to come into office with policy preferences and ideas, discuss them with affected parties, sound out other agencies for support, and work with staff attorneys to substantiate the legal basis for a preferred policy. The record here reflects the sometimes involved nature of Executive Branch decisionmaking, but no *2575particular step in the process stands out as inappropriate or defective.
And yet, viewing the evidence as a whole, we share the District Court's conviction that the decision to reinstate a citizenship question cannot be adequately explained in terms of DOJ's request for improved citizenship data to better enforce the VRA. Several points, considered together, reveal a significant mismatch between the decision the Secretary made and the rationale he provided.
The record shows that the Secretary began taking steps to reinstate a citizenship question about a week into his tenure, but it contains no hint that he was considering VRA enforcement in connection with that project. The Secretary's Director of Policy did not know why the Secretary wished to reinstate the question, but saw it as his task to "find the best rationale." Id., at 551. The Director initially attempted to elicit requests for citizenship data from the Department of Homeland Security and DOJ's Executive Office for Immigration Review, neither of which is responsible for enforcing the VRA. After those attempts failed, he asked Commerce staff to look into whether the Secretary could reinstate the question without receiving a request from another agency. The possibility that DOJ's Civil Rights Division might be willing to request citizenship data for VRA enforcement purposes was proposed by Commerce staff along the way and eventually pursued.
Even so, it was not until the Secretary contacted the Attorney General directly that DOJ's Civil Rights Division expressed interest in acquiring census-based citizenship data to better enforce the VRA. And even then, the record suggests that DOJ's interest was directed more to helping the Commerce Department than to securing the data. The December 2017 letter from DOJ drew heavily on contributions from Commerce staff and advisors. Their influence may explain why the letter went beyond a simple entreaty for better citizenship data-what one might expect of a typical request from another agency-to a specific request that Commerce collect the data by means of reinstating a citizenship question on the census. Finally, after sending the letter, DOJ declined the Census Bureau's offer to discuss alternative ways to meet DOJ's stated need for improved citizenship data, further suggesting a lack of interest on DOJ's part.
Altogether, the evidence tells a story that does not match the explanation the Secretary gave for his decision. In the Secretary's telling, Commerce was simply acting on a routine data request from another agency. Yet the materials before us indicate that Commerce went to great lengths to elicit the request from DOJ (or any other willing agency). And unlike a typical case in which an agency may have both stated and unstated reasons for a decision, here the VRA enforcement rationale-the sole stated reason-seems to have been contrived.
We are presented, in other words, with an explanation for agency action that is incongruent with what the record reveals about the agency's priorities and decisionmaking process. It is rare to review a record as extensive as the one before us when evaluating informal agency action-and it should be. But having done so for the sufficient reasons we have explained, we cannot ignore the disconnect between the decision made and the explanation given. Our review is deferential, but we are "not required to exhibit a naiveté from which ordinary citizens are free." United States v. Stanchich , 550 F.2d 1294, 1300 (CA2 1977) (Friendly, J.). The reasoned explanation requirement of administrative law, after all, is meant to ensure that agencies offer genuine justifications for important *2576decisions, reasons that can be scrutinized by courts and the interested public. Accepting contrived reasons would defeat the purpose of the enterprise. If judicial review is to be more than an empty ritual, it must demand something better than the explanation offered for the action taken in this case.
In these unusual circumstances, the District Court was warranted in remanding to the agency, and we affirm that disposition. See Florida Power & Light Co. v. Lorion , 470 U.S. 729, 744, 105 S.Ct. 1598, 84 L.Ed.2d 643 (1985). We do not hold that the agency decision here was substantively invalid. But agencies must pursue their goals reasonably. Reasoned decisionmaking under the Administrative Procedure Act calls for an explanation for agency action. What was provided here was more of a distraction.
* * *
The judgment of the United States District Court for the Southern District of New York is affirmed in part and reversed in part, and the case is remanded for further proceedings consistent with this opinion.
It is so ordered.
Justice THOMAS, with whom Justice GORSUCH and Justice KAVANAUGH join, concurring in part and dissenting in part.
In March 2018, the Secretary of Commerce exercised his broad discretion over the administration of the decennial census to resume a nearly unbroken practice of asking a question relating to citizenship. Our only role in this case is to decide whether the Secretary complied with the law and gave a reasoned explanation for his decision. The Court correctly answers these questions in the affirmative. Ante , at 2566 - 2573. That ought to end our inquiry.
The Court, however, goes further. For the first time ever, the Court invalidates an agency action solely because it questions the sincerity of the agency's otherwise adequate rationale. Echoing the din of suspicion and distrust that seems to typify modern discourse, the Court declares the Secretary's memorandum "pretextual" because, "viewing the evidence as a whole," his explanation that including a citizenship question on the census would help enforce the Voting Rights Act (VRA) "seems to have been contrived." Ante , at 2572 - 2573, 2574 - 2575, 2575 - 2576. The Court does not hold that the Secretary merely had additional , unstated reasons for reinstating the citizenship question. Rather, it holds that the Secretary's stated rationale did not factor at all into his decision.
The Court's holding reflects an unprecedented departure from our deferential review of discretionary agency decisions. And, if taken seriously as a rule of decision, this holding would transform administrative law. It is not difficult for political opponents of executive actions to generate controversy with accusations of pretext, deceit, and illicit motives. Significant policy decisions are regularly criticized as products of partisan influence, interest-group pressure, corruption, and animus. Crediting these accusations on evidence as thin as the evidence here could lead judicial review of administrative proceedings to devolve into an endless morass of discovery and policy disputes not contemplated by the Administrative Procedure Act (APA).
Unable to identify any legal problem with the Secretary's reasoning, the Court imputes one by concluding that he must not be telling the truth. The Court therefore upholds the decision of the District Court-which, in turn, was transparently based on the application of an administration-specific standard. App. to Pet. for Cert. 527a (crediting respondents' allegations *2577that "the current Department of Justice has shown little interest in enforcing the" VRA (emphasis added)).
The law requires a more impartial approach. Even assuming we are authorized to engage in the review undertaken by the Court-which is far from clear-we have often stated that courts reviewing agency action owe the Executive a "presumption of regularity." Citizens to Preserve Overton Park, Inc. v. Volpe , 401 U.S. 402, 415, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971). The Court pays only lipservice to this principle. But, the evidence falls far short of supporting its decision. The Court, I fear, will come to regret inventing the principles it uses to achieve today's result. I respectfully dissent from Part V of the opinion of the Court.1
I
As the Court explains, federal law directs the Secretary of Commerce to "take a decennial census." 13 U. S. C. § 141(a) ; see U. S. Const., Art. I, § 2, cl. 3 ; Amdt. XIV, § 2; ante , at 2561 - 2562. The discretion afforded the Secretary is extremely broad. Subject only to constitutional limitations and a handful of inapposite statutory requirements, the Secretary is expressly authorized to "determine the inquiries" on the census questionnaire and to conduct the census "in such form and content as he may determine." §§ 5, 141(a) ; see ante , at 2567 - 2569, 2571 - 2573.2 Prior census questionnaires have included questions ranging from sex, age, and race to commute, education, and radio ownership. And between 1820 and 2010, every decennial census questionnaire but one asked some segment of the population a question related to citizenship. The 2010 census was the first since 1840 that did not include any such question.
In March 2018, the Secretary issued a memorandum reinstating a citizenship question on the 2020 census. He explained that the Department of Justice (DOJ) had formally requested reinstatement of the question because the data obtained would help enforce § 2 of the VRA. He further explained that the question had been well tested in light of its extensive previous use, that he had consulted with the Census Bureau on the proposal, and that his final decision incorporated feedback from the Bureau. He recognized that staff at the Bureau believed that better data could be obtained through modeling and reliance on existing records, but he disagreed with that assessment, explaining that the data was inconclusive and that he thought it preferable to ask the question directly of the entire population. Respondents brought suit, seeking judicial review of the Secretary's decision under the APA, 5 U. S. C. § 706.
II
As relevant here, the APA requires courts to "hold unlawful and set aside" agency action that is "arbitrary, capricious, an abuse of discretion, or otherwise not in *2578accordance with law." § 706(2)(A). We have emphasized that "[r]eview under the arbitrary and capricious standard is deferential." National Assn. of Home Builders v. Defenders of Wildlife , 551 U.S. 644, 658, 127 S.Ct. 2518, 168 L.Ed.2d 467 (2007) ; see Glickman v. Wileman Brothers & Elliott, Inc. , 521 U.S. 457, 466, n. 8, 117 S.Ct. 2130, 138 L.Ed.2d 585 (1997). It requires the reviewing court to determine whether the agency " 'examine[d] the relevant data and articulate[d] a satisfactory explanation for its action.' " FCC v. Fox Television Stations, Inc. , 556 U.S. 502, 513, 129 S.Ct. 1800, 173 L.Ed.2d 738 (2009). We have described this as a " 'narrow' standard of review" under which the reviewing court cannot " 'substitute its judgment for that of the agency,' and should 'uphold a decision of less than ideal clarity if the agency's path may reasonably be discerned.' " Id. , at 513-514, 129 S.Ct. 1800 (citation omitted); accord, Motor Vehicle Mfrs. Assn. of United States, Inc. v. State Farm Mut. Automobile Ins. Co. , 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983).3
Part IV-B of the opinion of the Court correctly applies this standard to conclude that the Secretary's decision survives ordinary arbitrary-and-capricious review. That holding should end our inquiry.
But the opinion continues. Acknowledging that "no particular step" in the proceedings here "stands out as inappropriate or defective," even after reviewing "all the evidence in the record ..., including the extra-record discovery," ante , at 2574, the Court nevertheless agrees with the District Court that the Secretary's rationale for reinstating the citizenship question was "pretextual-that is, that the real reason for his decision was something other than the sole reason he put forward in his memorandum, namely enhancement of DOJ's VRA enforcement efforts." 351 F.Supp.3d 502, 660 (SDNY 2019) ; see ante , at 2575 - 2576. According to the Court, something just "seems" wrong. Ibid.
This conclusion is extraordinary. The Court engages in an unauthorized inquiry into evidence not properly before us to reach an unsupported conclusion. Moreover, each step of the inquiry offends the presumption of regularity we owe the Executive. The judgment of the District Court should be reversed.
A
Section 706(2) of the APA contemplates review of the administrative "record" to determine whether an agency's "action, findings, and conclusions" satisfy six specified standards. See §§ 706(2)(A)-(F). None instructs the Court to inquire into pretext. Consistent with this statutory text, we have held that a court is "ordinarily limited to evaluating the agency's contemporaneous explanation in light of the existing administrative record." Ante , at 2573 (citing Vermont Yankee Nuclear Power Corp. v. Natural Resources Defense Council, Inc. , 435 U.S. 519, 549, 98 S.Ct. 1197, 55 L.Ed.2d 460 (1978) ); see SEC v. Chenery Corp. , 318 U.S. 80, 87, 63 S.Ct. 454, 87 L.Ed. 626 (1943) ("The grounds upon which an administrative order must be judged are those upon which the record discloses that its action was based"). If an agency's stated findings and conclusions withstand scrutiny, the APA does not permit a court to set aside the decision solely because the agency had "other unstated *2579reasons" for its decision, such as "political considerations" or the "Administration's priorities." Ante , at 2573 - 2574.
Unsurprisingly, then, this Court has never held an agency decision arbitrary and capricious on the ground that its supporting rationale was "pretextual." Nor has it previously suggested that this was even a possibility. Under "settled propositions" of administrative law, ante , at 2572 - 2573, pretext is virtually never an appropriate or relevant inquiry for a reviewing court to undertake.
Respondents conceptualize pretext as a subset of "arbitrary and capricious" review. It is far from clear that they are correct. But even if they were, an agency action is not arbitrary or capricious merely because the decisionmaker has other, unstated reasons for the decision. Ante , at 2573 - 2574. Nor is an agency action arbitrary and capricious merely because the decisionmaker was "inclined" to accomplish it before confirming that the law and facts supported that inclination. In re Dept. of Commerce , 586 U. S. ----, ----, 139 S.Ct. 16, 17, 202 L.Ed.2d 306 (2018) (GORSUCH, J., concurring in part and dissenting in part).
Accordingly, even under respondents' approach, a showing of pretext could render an agency action arbitrary and capricious only in the infinitesimally small number of cases in which the administrative record establishes that an agency's stated rationale did not factor at all into the decision, thereby depriving the action of an adequate supporting rationale.4 This showing is extremely difficult to make because the administrative record will rarely, if ever, contain evidence sufficient to show that an agency's stated rationale did not actually factor into its decision. And we have stated that a "strong showing of bad faith or improper behavior" is necessary to venture beyond the agency's "administrative findings" and inquire into "the mental processes of administrative decisionmakers." Overton Park , 401 U.S. at 420, 91 S.Ct. 814.5 We have never before found Overton Park 's exception satisfied, much less invalidated an agency action based on "pretext."
Undergirding our arbitrary-and-capricious analysis is our longstanding precedent affording the Executive a "presumption of regularity." Id., at 415, 91 S.Ct. 814 ; see United States v. Chemical Foundation, Inc. , 272 U.S. 1, 14-15, 47 S.Ct. 1, 71 L.Ed. 131 (1926). This presumption reflects respect for a coordinate branch of government whose officers not only take *2580an oath to support the Constitution, as we do, Art. VI, but also are charged with "faithfully execut[ing]" our laws, Art. II, § 3. See United States v. Morgan , 313 U.S. 409, 422, 61 S.Ct. 999, 85 L.Ed. 1429 (1941) (presumption of regularity ensures that the "integrity of the administrative process" is appropriately respected). In practice, then, we give the benefit of the doubt to the agency.
B
The Court errs at the outset by proceeding beyond the administrative record to evaluate pretext. Respondents have not made a "strong showing of bad faith or improper behavior." Overton Park , supra , at 420, 91 S.Ct. 814.
The District Court's initial order granting extra-record discovery relied on four categories of evidence:
"evidence that [the Secretary] was predisposed to reinstate the citizenship question when he took office; that the [DOJ] hadn't expressed a desire for more detailed citizenship data until the Secretary solicited its view; that he overruled the objections of his agency's career staff; and that he declined to order more testing of the question given its long history." Dept. of Commerce , 586 U. S., at ----, 139 S.Ct., at 18.
None of this comes close to showing bad faith or improper behavior. Indeed, there is nothing even "unusual about a new cabinet secretary coming to office inclined to favor a different policy direction, soliciting support from other agencies to bolster his views, disagreeing with staff, or cutting through red tape." Ibid . Today all Members of the Court who reach the question agree that the District Court abused its discretion in ordering extra-record discovery based on this evidence. Ante , at 2574 ("We agree with the Government that the District Court should not have ordered extra-record discovery when it did").
Nevertheless, the Court excuses the error because, in its view, "the new material that the parties [later] stipulated should have been part of the administrative record ... largely justified such extra-record discovery as occurred." Ibid. Given the requirement that respondents make a "strong showing" of bad faith, one would expect the Court to identify which "new material" supported such a showing. It does not. Nor does the Court square its suggestion that some of the extra-record discovery was not "justified" with its consideration of "all ... the extra-record discovery." Ante , at 2574 - 2575. Regardless, I assume that the Court has in mind the administrative-record materials that the District Court would later rely on to establish pretext:
"evidence that [the Secretary] had made the decision to add the citizenship question well before DOJ requested its addition in December 2017; the absence of any mention, at all , of VRA enforcement in the discussions of adding the question that preceded the [DOJ] Letter; unsuccessful attempts by Commerce Department staff to shop around for a request by another agency regarding citizenship data; and [the Secretary's] personal outreach to Attorney General Sessions, followed by the [DOJ] Letter; not to mention the conspicuous procedural irregularities that accompanied the decision to add the question." 351 F.Supp.3d at 661 (citations omitted).
This evidence fails to make a strong showing of bad faith or improper behavior. Taken together, it proves at most that the Secretary was predisposed to add a citizenship question to the census and took steps to achieve that end before settling on the VRA rationale he included in his memorandum. Perhaps he had reasons for adding *2581the citizenship question other than the VRA, but by the Court's own telling, that does not amount to evidence of bad faith or improper behavior. Ante , at 2573 - 2574; see Dept. of Commerce , supra , at 17.
The Court thus errs in relying on materials outside the record to support its holding. And the Court does not claim that the evidence in the administrative record alone would prove that the March 2018 memorandum was a pretext. Given the presumption of regularity, the evidence discussed above falls far short of establishing that the VRA rationale did not factor at all into the Secretary's decision.
C
Even if it were appropriate for the Court to rely on evidence outside the administrative record, that evidence still fails to establish pretext. None of the evidence cited by the Court or the District Court comes close to showing that the Secretary's stated rationale-that adding a citizenship question to the 2020 census questionnaire would "provide ... data that are not currently available" and "permit more effective enforcement of the [VRA]," App. to Pet. for Cert. 548a-did not factor at all into his decision.
Once again, the evidence cited by the Court suggests at most that the Secretary had "other unstated reasons" for reinstating the citizenship question. Ante , at 2573 - 2574. For example, the Court states that the Secretary's Director of Policy "initially attempted to elicit requests for citizenship data from the Department of Homeland Security and DOJ's Executive Office for Immigration Review." Ante , at 2575. But this hardly shows pretext. It simply suggests that the Director believed that citizenship information could be useful in tackling problems related to national security and illegal immigration-a view that would also explain why the Secretary might not have been "considering VRA enforcement" early on. Ibid. ; see also American Community Survey, Why We Ask: Place of Birth, Citizenship and Year of Entry (2016) (explaining that inquiries about "place of birth, citizenship, and year of entry" provide statistics that are "essential for agencies and policy makers setting and evaluating immigration policies and laws, understanding how different immigrant groups are assimilated, and monitoring against discrimination"), https://www2.census.gov/programs-surveys/acs/about/qbyqfact/2016/Citizenship.pdf (as last visited June 25, 2019).
The Court emphasizes that the VRA rationale for the citizenship question originated in the Department of Commerce, and suggests that DOJ officials unthinkingly fell in line after the Attorney General was looped into the process. See ante , at 2575. But the Court ignores that the letter was drafted by the then-Acting Assistant Attorney General for Civil Rights and reviewed by five other DOJ attorneys, including the Chief of the DOJ's Voting Section. 351 F.Supp.3d at 554-556. Given the DOJ's multilayer review process and its explanation for requesting citizenship data, the Court's suggestion that the DOJ's letter was inadequately vetted or improperly "influence[d]" by the Department of Commerce is entirely unsupported. Ante , at 2575. In any event, none of this suggests, much less proves, that the Secretary harbored an unstated belief that adding the citizenship question would not help enforce the VRA, or that the VRA rationale otherwise did not factor at all into his decision. It simply suggests that a number of executive officials agreed that adding a citizenship question would support VRA enforcement.
The Court's other evidence is even further afield. The Court thinks it telling that the DOJ's letter included "a specific request *2582that Commerce collect the [citizenship] data by means of reinstating a citizenship question on the census," rather than a more open-ended "entreaty for better citizenship data." Ibid. I do not understand how the specificity of the DOJ's letter bears on whether the Secretary's rationale was pretextual-particularly since the letter specifically explained why "census questionnaire data regarding citizenship, if available, would be more appropriate for use in redistricting and in [VRA] litigation" than existing data. App. to Pet. for Cert. 568a; see id. , at 567a-568a. Unless the Court is now suggesting that agency correspondence must comply with the Court's subjective, unsupported view of what "might" constitute a "typical request from another agency," ante , at 2575, the specificity of the DOJ's letter is irrelevant. The Court also points to the DOJ's decision not to meet with the Census Bureau "to discuss alternative ways to meet DOJ's stated need for improved citizenship data." Ibid. But the Court does not explain how the DOJ's refusal bears on the Secretary's rationale. Besides, it is easy to understand why DOJ officials would not be interested in meeting with the Census Bureau. The meeting would have been with career employees whose acknowledged purpose was to talk the DOJ out of its request. See 351 F.Supp.3d at 557. Having already considered the issue and explained the rationale behind the request, it seems at least plausible that the DOJ officials believed such a meeting would be unproductive.
In short, the evidence cited by the Court establishes, at most, that leadership at both the Department of Commerce and the DOJ believed it important-for a variety of reasons-to include a citizenship question on the census.
The Court also fails to give credit where it is due. The Secretary initiated this process inclined to favor what he called "Option B"-that is, simply "add[ing] a citizenship question to the decennial census." App. to Pet. for Cert. 552a. But the Census Bureau favored "Option C"-relying solely on "administrative records" to supply the information needed by the DOJ. Id. , at 554a. The Secretary considered this view and found it a "potentially appealing solution," ibid. , but concluded that it had shortcomings. Rather than revert to his original inclination, however, he "asked the Census Bureau to develop a fourth alternative, Option D, which would combine Options B and C." Id. , at 555a. And he settled on that solution. Whatever one thinks of the Secretary's choice, his willingness to change his mind in light of the Bureau's feedback belies the idea that his rationale or decisionmaking process was a pretext.
The District Court's lengthy opinion pointed to other facts that, in its view, supported a finding of pretext. 351 F.Supp.3d at 567-572, 660-664 (discussing the statements, e-mails, acts, and omissions of numerous people involved in the process). I do not deny that a judge predisposed to distrust the Secretary or the administration could arrange those facts on a corkboard and-with a jar of pins and a spool of string-create an eye-catching conspiracy web. Cf. id. , at 662 (inferring "from the various ways in which [the Secretary] and his aides acted like people with something to hide that they did have something to hide"). But the Court does not rely on this evidence, and rightly so: It casts no doubt on whether the Secretary's stated rationale factored into his decision. The evidence suggests, at most, that the Secretary had multiple reasons for wanting to include the citizenship question on the census.
Finally, if there could be any doubt about this conclusion, the presumption of *2583regularity resolves it. Where there are equally plausible views of the evidence, one of which involves attributing bad faith to an officer of a coordinate branch of Government, the presumption compels giving the benefit of the doubt to that officer.
III
The Court's erroneous decision in this case is bad enough, as it unjustifiably interferes with the 2020 census. But the implications of today's decision are broader. With today's decision, the Court has opened a Pandora's box of pretext-based challenges in administrative law.
Today's decision marks the first time the Court has ever invalidated an agency action as "pretextual." Having taken that step, one thing is certain: This will not be the last time it is asked to do so. Virtually every significant agency action is vulnerable to the kinds of allegations the Court credits today. These decisions regularly involve coordination with numerous stakeholders and agencies, involvement at the highest levels of the Executive Branch, opposition from reluctant agency staff, and-perhaps most importantly-persons who stand to gain from the action's demise. Opponents of future executive actions can be expected to make full use of the Court's new approach.
The 2015 "Open Internet Order" provides a case in point. In 2015, the Federal Communications Commission (FCC) adopted a controversial order reclassifying broadband Internet access service as a "telecommunications service" subject to regulation under Title II of the Communications Act. See In re Protecting and Promoting the Open Internet , 30 FCC Rcd. 5601, 5618 (2015). According to a dissenting Commissioner, the FCC "flip-flopp[ed]" on its previous policy not because of a change in facts or legal understanding, but based on "one reason and one reason alone. President Obama told us to do so." Id., at 5921 (statement of Comm'r Pai). His view was supported by a 2016 congressional Report in which Republican Senate staff concluded that "the FCC bent to the political pressure of the White House" and "failed to live up to standards of transparency." Majority Staff Report, Senate Committee on Homeland Security and Governmental Affairs, Regulating the Internet: How the White House Bowled Over FCC Independence, 114th Cong., 1st Sess., 29 (Comm. Print 2016). The Report cited evidence strikingly similar to that relied upon by the Court here-including agency-initiated "meetings with certain outside groups to support" the new result, id. , at 3; "apparen[t] ... concern from the career staff that there was insufficient notice to the public and affected stakeholders," id., at 4; and "regula[r] communicatio[n]" between the FCC Chairman and "presidential advisors," id. , at 25.
Under the malleable standard applied by the Court today, a serious case could be made that the Open Internet Order should have been invalidated as "pretextual," regardless of whether any "particular step in the process stands out as inappropriate or defective." Ante , at 2575. It is enough, according to the Court, that a judge believes that the ultimate rationale "seems to have been contrived" when the evidence is considered "as a whole." Ante , at 2574, 2575 - 2576.
Now that the Court has opened up this avenue of attack, opponents of executive actions have strong incentives to craft narratives that would derail them. Moreover, even if the effort to invalidate the action is ultimately unsuccessful, the Court's decision enables partisans to use the courts to harangue executive officers through depositions, discovery, delay, and distraction. The Court's decision could even implicate separation-of-powers concerns insofar as it *2584enables judicial interference with the enforcement of the laws.
In short, today's decision is a departure from traditional principles of administrative law. Hopefully it comes to be understood as an aberration-a ticket good for this day and this train only.
* * *
Because the Secretary's decision to reinstate a citizenship question on the 2020 census was legally sound and a reasoned exercise of his broad discretion, I respectfully dissent from Part V of the opinion of the Court.
Justice BREYER, with whom Justice GINSBURG, Justice SOTOMAYOR, and Justice KAGAN join, concurring in part and dissenting in part.
I join Parts I, II, IV-A, and V of the Court's opinion (except as otherwise indicated in this opinion). I dissent, however, from the conclusion the Court reaches in Part IV-B. To be more specific, I agree with the Court that the Secretary of Commerce provided a pretextual reason for placing a question about citizenship on the short-form census questionnaire and that a remand to the agency is appropriate on that ground. But I write separately because I also believe that the Secretary's decision to add the citizenship question was arbitrary and capricious and therefore violated the Administrative Procedure Act (APA).
There is no serious dispute that adding a citizenship question would diminish the accuracy of the enumeration of the population-the sole constitutional function of the census and a task of great practical importance. The record demonstrates that the question would likely cause a disproportionate number of noncitizens and Hispanics to go uncounted in the upcoming census. That, in turn, would create a risk that some States would wrongfully lose a congressional representative and funding for a host of federal programs. And, the Secretary was told, the adverse consequences would fall most heavily on minority communities. The Secretary decided to ask the question anyway, citing a need for more accurate citizenship data. But the evidence indicated that asking the question would produce citizenship data that is less accurate, not more. And the reason the Secretary gave for needing better citizenship data in the first place-to help enforce the Voting Rights Act of 1965-was not convincing.
In short, the Secretary's decision to add a citizenship question created a severe risk of harmful consequences, yet he did not adequately consider whether the question was necessary or whether it was an appropriate means of achieving his stated goal. The Secretary thus failed to "articulate a satisfactory explanation" for his decision, "failed to consider ... important aspect[s] of the problem," and "offered an explanation for [his] decision that runs counter to the evidence," all in violation of the APA. Motor Vehicle Mfrs. Assn. of United States, Inc. v. State Farm Mut. Automobile Ins. Co. , 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983). These failures, in my view, risked undermining public confidence in the integrity of our democratic system itself. I would therefore hold that the Secretary's decision-whether pretextual or not-was arbitrary, capricious, and an abuse of discretion.
I
A
Three sets of laws determine the legal outcome of this case. First, the Constitution requires an "actual Enumeration" of the "whole number of persons in each State" every 10 years. Art. I, § 2, cl. 3 ; Amdt. 14, § 2. It does so in order to *2585"provide a basis for apportioning representatives among the states in the Congress." Baldrige v. Shapiro , 455 U.S. 345, 353, 102 S.Ct. 1103, 71 L.Ed.2d 199 (1982) ; see also Art. I, § 2, cl. 3. The inclusion of this provision in the Constitution itself underscores the importance of conducting an accurate census. See Utah v. Evans , 536 U.S. 452, 478, 122 S.Ct. 2191, 153 L.Ed.2d 453 (2002) (recognizing "a strong constitutional interest in [the] accuracy" of the enumeration).
Second, the Census Act contains two directives that constrain the Secretary's ability to add questions to the census. Section 195 says that the Secretary "shall, if he considers it feasible," authorize the use of statistical "sampling" in collecting demographic information. That means the Secretary must, if feasible, obtain demographic information through a survey sent to a sample of households, rather than through the short-form census questionnaire to which every household must respond. The other relevant provision, § 6(c), says that "[t ]o the maximum extent possible and consistent with the kind, timeliness, quality and scope of the statistics required, the Secretary shall acquire and use information available" from administrative sources "instead of conducting direct inquiries." (Emphasis added.) These provisions, taken together, reflect a congressional preference for keeping the short form short, so that it does not burden recipients and thereby discourage them from responding.
Third, the APA prohibits administrative agencies from making choices that are "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." 5 U. S. C. § 706(2)(A). We have said that courts, in applying this provision, must decide "whether the decision was based on a consideration of the relevant factors and whether there has been a clear error of judgment." Citizens to Preserve Overton Park, Inc. v. Volpe , 401 U.S. 402, 416, 91 S.Ct. 814, 28 L.Ed.2d 136 (1971). The agency must have "examine[d] the relevant data and articulate[d] a satisfactory explanation for its action[,] including a 'rational connection between the facts found and the choice made.' " State Farm , 463 U.S. at 43, 103 S.Ct. 2856. An agency ordinarily fails to meet this standard if it has "failed to consider an important aspect of the problem, offered an explanation for its decision that runs counter to the evidence before the agency, or is so implausible that it could not be ascribed to a difference in view or the product of agency expertise." Ibid.
Courts do not apply these principles of administrative law mechanically. Rather, they take into account, for example, the nature and importance of the particular decision, the relevance and importance of missing information, and the inadequacies of a particular explanation in light of their importance. The Federal Government makes tens of thousands, perhaps millions, of administrative decisions each year. And courts would be wrong to expect or insist upon administrative perfection. But here, the Enumeration Clause, the Census Act, and the nature of the risks created by the agency's decision all make clear that the decision before us is highly important to the proper functioning of our democratic system. It is therefore particularly important that courts here not overlook an agency's (1) failure to consider serious risks of harm, (2) failure to explain its refusal to minimize those risks, or (3) failure to link its conclusion to available evidence. My view, like that of the District Court, is that the agency here failed on all three counts.
B
A brief history of how the census has worked over the years will help the reader understand some of the shortcomings of *2586the Secretary's decisionmaking process. The Framers wrote into the Constitution a mandate to conduct an "actual Enumeration" of the population every 10 years. Art. I, § 2, cl. 3. They did so for good reason. The purpose of the census is to "provide a basis for apportioning representatives among the states in the Congress," Baldrige , 455 U.S. at 353, 102 S.Ct. 1103, ensuring that "comparative state political power in the House ... reflect[s] comparative population," Evans , 536 U.S. at 477, 122 S.Ct. 2191. The Framers required an actual count of every resident to "limit political chicanery" and to prevent the census count from being "skewed for political ... purposes." Id. , at 500, 122 S.Ct. 2191 (THOMAS, J., concurring in part and dissenting in part).
Throughout most of the Nation's history, the Federal Government used enumerators, often trained census takers, to conduct the census by going door to door. The enumerators would ask a host of questions, including place of birth, citizenship, and others. But after the 1950 census, the Bureau began to change its approach. Post-census studies revealed that the census had failed to count more than 5 million people and that the undercount disproportionately affected members of minority groups. See M. Anderson, The American Census: A Social History 201-202 (1988); Brief for Historians and Social Scientists as Amici Curiae 15. Studies showed that statistical sampling would produce higher quality data. Anderson, American Census, at 201.
Beginning with the 1960 census, the Bureau consequently divided its questioning into a short form and a long form. The short form contained a list of questions-a short list-that the census would ask of every household. That list included basic demographic questions like sex, age, race, and marital status. The short form did not include, and has never included, a question about citizenship. See ibid. ; Dept. of Commerce, U. S. Census Bureau, Measuring America: The Decennial Censuses From 1790 to 2000, p. 128 (2002). By way of contrast, the long form set forth a host of questions that would be asked of only a sample of households. In 1960, the long form was sent to one in every four households; in subsequent years, it was sent to approximately one in every six. See 351 F.Supp.3d 502, 520 (SDNY 2019). And it was more recently replaced by the American Community Survey (ACS), which is sent to approximately 1 in 38 households each year. The long form (and now the ACS) has often included a question about citizenship.
In 1970, the Census Bureau made another important change to the census. It significantly reduced its reliance upon in-person enumerators. See Anderson, supra , at 206. Instead, it sent nearly all households a questionnaire by mail. Most households received the short form, and a small sample received the long form. Instructions on the form told each household to fill out the questionnaire and return it to the Census Bureau by mail. Enumerators would follow up with households that did not return the questionnaire.
To maximize accuracy and minimize cost, the Bureau tried to bring about the highest possible "self-response" rate, i.e. , to encourage as many households as possible to respond by mail. For that reason, it tried to keep the short form as short as possible. And it consistently opposed placing a citizenship question on that form. It feared that adding a question about citizenship would "inevitably jeopardize the overall accuracy of the population count," partly because of added response burden but also because, as it explained, noncitizens faced with a citizenship question would be less likely to respond due to *2587fears of "the information being used against them." Federation for Am. Immigration Reform v. Klutznick , 486 F.Supp. 564, 568 (DDC 1980).
Likely for similar reasons, Congress amended the Census Act in 1976, enacting the two statutory provisions to which I previously referred. These two provisions, 13 U. S. C. § 6(c) and § 195, together encourage the Secretary not to ask demographic questions on the short form if the information can be obtained either through the long form or through administrative records.
II
With this statutory and historical background, we can more easily consider the agency decision directly under review. That decision "reinstate[s] [a] citizenship question on the 2020 decennial census." App. to Pet. for Cert. 549a-550a (Memorandum from Wilbur L. Ross, Jr., Secretary of Commerce, to Karen Dunn Kelley, Under Secretary for Economic Affairs (Mar. 26, 2018)). The agency's decision memorandum provided one and only one reason for making that decision-namely, that the question was "necessary to provide complete and accurate data in response to" a request from the Department of Justice (DOJ). Id. , at 562a. The DOJ had requested the citizenship question for "use [in] ... determining violations of Section 2 of the Voting Rights Act." Id. , at 548a.
The decision memorandum adds that the agency had not been able to "determine definitively how inclusion of a citizenship question on the decennial census will impact responsiveness. However, even if there is some impact on responses, the value of more complete and accurate data derived from surveying the entire population outweighs such concerns." Id. , at 562a. The Secretary's decision thus rests upon a weighing of potentially adverse consequences (diminished responses and a less accurate census count) against potentially offsetting advantages (better citizenship data). In my view, however, the Secretary did not make reasonable decisions about these potential costs and benefits in light of the administrative record.
A
Consider first the Secretary's conclusion that he was "not able to determine definitively how inclusion of a citizenship question on the decennial census will impact responsiveness." Ibid. Insofar as this statement implies that adding the citizenship question is unlikely to affect "responsiveness" very much (or perhaps at all), the evidence in the record indicates the contrary.
1
The administrative record includes repeated Census Bureau statements that adding the question would produce a less accurate count because noncitizens and Hispanics would be less likely to respond to the questionnaire. See App. 105, 109-112, 158. The Census Bureau's chief scientist said specifically that adding the question would have "an adverse impact on self-response and, as a result, on the accuracy and quality of the 2020 Census." Id. , at 109. And the chief scientist backed this statement up by pointing to "[t]hree distinct analyses." Ibid.
The first analysis compared nonresponse rates for the short-form census questionnaire (which did not include a citizenship question) to nonresponse rates for the ACS (which did). Obviously, more people fail to respond to the ACS than to the short form. Yet taking into account the fact that the nonresponse rate will be greater for the ACS than for the short form, the Bureau found that the difference *2588between the two is yet greater for noncitizen households than for citizen households (by 5.1%, according to the Bureau). Id. , at 111. This led the Bureau to say that it was a "reasonable inference" that the presence of the citizenship question accounted for the difference. Ibid.
The Bureau conducted two additional studies, both analyzing data from the ACS. One study looked at response rates for particular questions on the ACS. It showed that the "no answer" rate for the citizenship question was "much greater than the comparable rates" for other census questions (for example, questions about age, sex, race, and ethnicity). Id. , at 110. And it showed that the "no answer" rate for the citizenship question was significantly higher among Hispanics. Id. , at 109-110. The last study examined "break-off" rates, i.e. , the rate at which respondents stopped answering the questionnaire upon reaching a particular question. It found that Hispanics were significantly more likely than were non-Hispanics to stop answering at the point they reached the citizenship question. Id. , at 112. Together, these two studies provided additional support for the Census Bureau's determination that the citizenship question is likely to mean disproportionately fewer responses from noncitizens and Hispanics than from others. Ibid.
Putting numbers upon these study results, the Census Bureau estimated that adding the question to the short form would lead to 630,000 additional nonresponding households. Id. , at 114. That is to say, the question would cause households covering more than 1 million additional people to decline to respond to the census. When the Bureau does not receive a response, it follows up with in-person interviews in an effort to obtain the missing information. The Bureau often interviews what it calls "proxies," such as family members and neighbors. But this followup process is subject to error; and the error rate is much greater than the error rate for self-responses. Ibid. The Bureau thus explained that lower self-response rates "degrade data quality" by increasing the risk of error and leading to hundreds of thousands of fewer correct enumerations. Id. , at 113-115. The Bureau added that its estimate was "conservative." Id. , at 115. It expected "differences between citizen and noncitizen response rates and data quality" to be "amplified" in the 2020 census "compared to historical levels." Ibid. Thus, it explained, "the decrease in self-response for citizen households in 2020 could be much greater than the 5.1 percentage points [it] observed during the 2010 Census." Id. , at 115-116. Its conclusion in light of this evidence was clear. Adding the citizenship question to the short form was "very likely to reduce the self-response rate" and thereby "har[m] the quality of the census count." Id. , at 105, 158.
The Census Bureau's analysis received support from other submissions. Several States pointed out that noncitizens and racial minorities had been undercounted in every prior census. Administrative Record 1091-1092. They also drew attention to recent surveys indicating that noncitizens had significant concerns about the confidentiality of census responses. Ibid. Former directors of the Census Bureau wrote that adding the citizenship question so late in the process "would put the accuracy of the enumeration and success of the census in all communities at grave risk." Id. , at 1057. The American Sociological Association and Census Scientific Advisory Committee echoed these warnings. See id. , at 787, 794-795. On the other hand, the Secretary received submissions by other groups that supported adding the question. See, e.g. , id. , at 1178-1179, 1206, 1276. But as far as I can tell (or as far as the *2589arguments made here and in the District Court inform the matter), none of these latter submissions significantly added to, or detracted from, the Census Bureau's submissions in respect to the question's likely impact on response rates.
2
The Secretary's decision memorandum reached a quite different conclusion from the Census Bureau. The memorandum conceded that "a lower response rate would lead to ... less accurate responses." App. to Pet. for Cert. 556a. But it concluded that neither the Census Bureau nor any stakeholders had provided "definitive, empirical support" for the proposition that the citizenship question would reduce response rates. Id. , at 554a. The memorandum relied for that conclusion upon a number of considerations, but each is contradicted by the record.
The memorandum first pointed to perceived shortcomings in the Census Bureau's analysis of nonresponse rates. It noted that response rates are generally lower overall for the long form and ACS than they are for the short form. Id. , at 552a-554a. But the Bureau explained that its analysis accounted for this consideration, see App. 111, and no one has given us reason to think the contrary. The Secretary also noted that the Bureau "was not able to isolate what percentage of [the] decline was caused by the inclusion of a citizenship question rather than some other aspect of the long form survey." App. to Pet. for Cert. 554a. But the Bureau said attributing the decline to the citizenship question was a "reasonable inference," App. 111, and again, nothing in the record contradicted the Bureau's judgment. And later analyses have borne out the Bureau's judgment that the citizenship question contributes to the decline in self-response. See, e.g. , id. , at 1002-1006, 1008 (August 2018 Census Bureau study).
The memorandum next cast doubt on the Census Bureau's analysis of the rate at which people responded to particular questions on the ACS. It noted that the "no answer" rate to the citizenship question was comparable to the "no answer" rate for other questions on the ACS, including educational attainment, income, and property insurance. App. to Pet. for Cert. 553a. But as discussed above, the Bureau found it significant that the "no answer" rate for the citizenship question was "much greater" than the "no answer" rate for the other questions that appear on the short form -that is, the form on which the citizenship question would appear. App. 110, 124. The Secretary offered no reason why the demographic variables to which he pointed provided a better point of comparison.
Finally, the memorandum relied on information provided by two outside stakeholders. The first was a study conducted by the private survey company Nielsen, in which questions about place of birth and time of arrival had not led to any appreciable decrease in the response rate. App. to Pet. for Cert. 552a. But Nielsen, which in fact urged the Secretary not to add the question, stated that its respondents (unlike census respondents) were paid to respond, and it is consequently not surprising that they did so. Administrative Record 1276. The memorandum also cited statements by former Census Bureau officials suggesting that empirical evidence about the question's potential impact on response rates was "limited." App. to Pet. for Cert. 558a-559a; see also id. , at 552a. But there was no reason to expect the former officials to provide more extensive empirical evidence as to a citizenship question when they were not privy to the internal Bureau analyses on this question. And, like Nielsen, the former *2590officials strongly urged the Secretary not to ask the question. See Administrative Record 1057.
The upshot is that the Secretary received evidence of a likely drop in census accuracy by a number somewhere in the hundreds of thousands, and he received nothing significant to the contrary. The Secretary pointed out that the Census Bureau's information was uncertain, i.e. , not "definitive." But that is not a satisfactory answer. Few public-policy-related statistical studies of risks (say, of many health or safety matters) are definitive. As the Court explained in State Farm , "[i]t is not infrequent that the available data do not settle a regulatory issue, and the agency must then exercise its judgment in moving from the facts and probabilities on the record to a policy conclusion." 463 U.S. at 52, 103 S.Ct. 2856. But an agency confronted with this situation cannot "merely recite the terms 'substantial uncertainty' as a justification for its actions." Ibid. Instead, it "must explain the evidence which is available" and typically must offer a reasoned explanation for taking action without "engaging in a search for further evidence." Ibid.
The Secretary did not do so here. He did not explain why he made the decision to add the question without following the Bureau's ordinary practice of extensively testing proposed changes to the census questionnaire. See App. 624-630, 641 (discussing testing process); see also, e.g. , Brief for Former Census Bureau Directors as Amici Curiae 17-21 (discussing prior examples of questions that the Bureau decided not to add after many years of pretesting). Without that testing, the Secretary could not treat the Bureau's expert opinions and its experience with the relevant surveys as worthless merely because its conclusions were not precise. The Bureau's opinions were properly considered as evidence of likelihoods, probabilities, or risks.
As noted above, the consequences of mistakes in the census count, of even a few hundred thousand, are grave. Differences of a few thousand people, as between one State and another, can mean a loss or gain of a congressional seat-a matter of great consequence to a State. See 351 F.Supp.3d at 594. And similar small differences can make a large difference to the allocation of federal funds among competing state programs. Id. , at 596-597 ; see also Baldrige , 455 U.S. at 353-354, n. 9, 102 S.Ct. 1103. If near-absolute certainty is what the Secretary meant by "definitive," that insistence would itself be arbitrary in light of the constitutional and statutory consequences at stake. And if the Secretary instead meant that the evidence does not indicate a serious risk of a less accurate count, that conclusion does not find support in the record.
B
Now consider the Secretary's conclusion that, even if adding a citizenship question diminishes the accuracy of the enumeration, "the value of more complete and accurate data derived from surveying the entire population outweighs ... concerns" about diminished accuracy. App. to Pet. for Cert. 562a (emphasis added). That conclusion was also arbitrary. The administrative record indicates that adding a citizenship question to the short form would produce less "complete and accurate data," not more.
1
The Census Bureau informed the Secretary that, for about 90% of the population, accurate citizenship data is available from administrative records maintained by the Social Security Administration and Internal Revenue Service. App. 146. The Bureau *2591further informed the Secretary that it had "high confidence" that it could develop a statistical model that would accurately impute citizenship status for the remaining 10% of the population. Ibid. The Bureau stated that these methods alone-using existing administrative records for 90% of the population and statistical modeling for the remaining 10%-would yield more accurate citizenship data than also asking a citizenship question. Id. , at 159. How could that be so? The answer is somewhat technical but readily understandable.
First , consider the 90% of the population (about 295 million people) as to whom administrative records are available. The Government agrees that using these administrative records would provide highly reliable information about citizenship, because the records "require proof of citizenship." Id. , at 117. By contrast, if responses to a citizenship question were used for this group, the Census Bureau predicted without contradiction that about one-third of the noncitizens in this group who respond would answer the question untruthfully, claiming to be citizens when they are not. Id. , at 147. Those incorrect answers-about 9.5 million in total-would conflict with the administrative records on file for those noncitizens. And what would the Census Bureau do with the conflicting data? If it accepts the answer to the citizenship question as determinative, it will have less accurate data. If it accepts the citizenship data from administrative records as determinative, asking the question will have served no purpose.
Thus, as to 295 million people-the overwhelming majority of the population-asking the citizenship question would at best add nothing at all. I say "at best" because, for one thing, the Census Bureau informed the Secretary that asking the question would produce 1 million more people who could not be linked to administrative records, which in turn would require the Census Bureau to resort to a less accurate source of citizenship data for these people. See id. , at 147-149; see also 351 F.Supp.3d at 538-539. For another, the policy of the Census Bureau has always been to use census responses rather than administrative records in cases where the two conflict. App. 147. In this case, that practice would mean accepting 9.5 million inaccurate responses even though accurate administrative records are available. See ibid. The Census Bureau could perhaps change that practice, but the Secretary's decision memorandum said nothing about the matter. It did not address the problem.
Second , consider the remaining 10% of the population (about 35 million people) for whom the Government lacks administrative records. The question here is which approach would yield the most "complete and accurate" citizenship data for this group-adding a citizenship question or using statistical modeling alone? To answer this question, we must further divide this group into two categories-those who would respond to the citizenship question if it were asked and those who would not.
Start with the category of about 22 million people who would answer a citizenship question if it were asked. Would their answers regarding citizenship be more accurate than citizenship data produced by statistical modeling? The Census Bureau said no. That is because many of the noncitizens in this group would answer the question falsely, resulting in an estimated 500,000 inaccurate answers. See id. , at 148. And those who answer the question falsely would be commingled, perhaps randomly, with those who answer it correctly, thereby casting doubt on the answers of all 22 million, with no way of knowing which answers are correct and which are false. By contrast, the Bureau believed that it could develop a statistical model that *2592would produce more accurate citizenship data than these census responses. The Bureau therefore informed the Secretary that it could do better. As the Bureau's chief scientist explained, although "[o]ne might think" that asking the question "could help fill the ... gaps" in the administrative records, the data did not support that assumption. Id. , at 157. Instead, he explained, responses to the citizenship question "may not be reliable," which "calls into question their ability to improve upon" the Bureau's statistical modeling process. Ibid.
Next, turn to the more than 13 million remaining people who would not answer the citizenship question even if it were asked. As to this category, the Census Bureau would still need to use statistical modeling to obtain citizenship data, because there would be no census response to use instead. Hence, asking the citizenship question would add nothing at all as to this group. To the contrary, as the Government concedes, asking the question would reduce the accuracy of the citizenship data for this group, because the relatively inaccurate answers to the citizenship question would diminish the overall accuracy of the Census Bureau's statistical model. See Brief for Petitioners 34 (conceding that the Census Bureau model will be "highe[r] quality" without the question than with it); 351 F.Supp.3d at 640 (explaining that asking the question would "corrup[t] ... the data generated by extrapolating from self-responses through imputation").
In sum, in respect to the 295 million persons for whom administrative records exist, asking the question on the short form would, at best, be no improvement over using administrative records alone. And in respect to the remaining 35 million people for whom no administrative records exist, asking the question would be no better, and in some respects would be worse, than using statistical modeling. The Census Bureau therefore told the Secretary that asking the citizenship question, even in addition to using administrative records, "would result in poorer quality citizenship data" than using administrative records alone, and would "still have all the negative cost and quality implications" of asking the citizenship question. App. 159. I could find no evidence contradicting that prediction.
2
If my description of the record is correct, it raises a serious legal problem. How can an agency support the decision to add a question to the short form, thereby risking a significant undercount of the population, on the ground that it will improve the accuracy of citizenship data, when in fact the evidence indicates that adding the question will harm the accuracy of citizenship data? Of course it cannot. But, as I have just said, I have not been able to find evidence to suggest that adding the question would result in more accurate citizenship data. Neither could the District Court. After reviewing the record in detail, the District Court found that "all of the relevant evidence before Secretary Ross-all of it-demonstrated that using administrative records ... would actually produce more accurate [citizenship] data than adding a citizenship question to the census." 351 F.Supp.3d at 650.
What consideration did the Secretary give to this problem? He stated simply that "[a]sking the citizenship question of 100 percent of the population gives each respondent the opportunity to provide an answer," which "may eliminate the need for the Census Bureau to have to impute an answer for millions of people." App. to Pet. for Cert. 556a. He therefore must have assumed, sub silentio , exactly what *2593the Census Bureau experts urged him not to assume-that answers to the citizenship question would be more accurate than statistical modeling. And he ignored the undisputed respects in which asking the question would make the existing data less accurate. Other than his assumption, the Secretary said nothing, absolutely nothing, to suggest a reasoned basis for disagreeing with the Bureau's expert statistical judgment.
The Government now maintains that the Secretary reasonably discounted the Census Bureau's recommendation because it was based on an untested prediction about the accuracy of its model. But this is not a case in which the Secretary was presented with a policy choice between two reasonable but uncertain options. For one thing, the record is much less uncertain than the Government acknowledges. Although it is true that the Census Bureau at one point told the Secretary that it could not "quantify the relative magnitude of the errors across the alternatives at this time," App. 148, it unequivocally stated that asking the question "would result in poorer quality citizenship data " than omitting it, id. , at 159 (emphasis added). Thus, even if the Bureau could not "quantify" the relative accuracy of the options, it could and did conclude that one option was likely more accurate than the other. Even in the face of some uncertainty, where all available evidence indicates that one option is better than the other, it is unreasonable to choose the worse option without explanation.
For another thing, to the extent the record reflects some uncertainty regarding the accuracy of the Census Bureau's statistical model, that is because the model needed to be "developed and tested" before it could be employed. Id. , at 146. But the Secretary made his decision before any such development or testing could be completed. Having decided to make an immediate decision rather than wait for testing, the Secretary could not dismiss the Bureau's prediction about the inadvisability of that decision on the ground that the prediction reflected likelihoods, probabilities, and risks rather than certainties.
Finally, recall that the Census Act requires the Secretary to use administrative records rather than direct inquiries to "the maximum extent possible." 13 U. S. C. § 6(c). That statutory requirement highlights what should be obvious: Whether adding a citizenship question to the short form would produce more accurate citizenship data is a relevant factor-indeed, a critically important factor-that the Secretary was required to consider. Here, the Secretary did not adequately explain why he rejected the evidence that adding the question would yield less accurate data. He did not even acknowledge that the Census Act obliged him to use administrative records rather than asking a question to the extent possible. And he did not explain how obtaining citizenship data that is no better or worse than the data otherwise available could justify jeopardizing the accuracy of the census count.
In these respects, the Secretary failed to consider "important aspect[s] of the problem" and "offered an explanation for [his] decision that runs counter to the evidence before the agency." State Farm , 463 U.S. at 43, 103 S.Ct. 2856.
C
The Secretary's failure to consider this evidence-that adding the question would harm the census count in the interest of obtaining less accurate citizenship data-provides a sufficient basis for setting the decision aside. But there is more. The reason that the Secretary provided for needing more accurate citizenship information in the first place-to help the DOJ
*2594enforce the Voting Rights Act-is unconvincing.
The Secretary stated that adding the citizenship question was "necessary to provide complete and accurate data in response to the DOJ request." App. to Pet. for Cert. 562a. The DOJ's request in turn asserted that the citizenship data currently available from the ACS was not "ideal" for enforcing the Voting Rights Act. Id. , at 567a. One of the DOJ's principal complaints was that ACS data is reported for groups of census blocks rather than for each census block itself. The DOJ letter stated that adding a citizenship question could provide it with individual block-by-block data which, the DOJ maintained, would allow it to better enforce the Voting Rights Act's protections for minority voters. Id. , at 568a.
This rationale is difficult to accept. One obvious problem is that the DOJ provided no basis to believe that more precise data would in fact help with Voting Rights Act enforcement. Congress enacted the Voting Rights Act in 1965-15 years after the census last asked every household about citizenship. Actions to enforce the Act have therefore always used citizenship data derived from sampling. Yet I am aware of no one-not in the Department of Commerce proceeding, in the District Court, or in this Court-who has provided a single example in which enforcement of the Act has suffered due to lack of more precise citizenship data. Organizations with expertise in this area tell us that asking the citizenship question will not help enforce the Act. See, e.g. , Brief for NAACP Legal Defense & Educational Fund, Inc., as Amicus Curiae 30-36. Rather, the question will, by depressing the count of minority groups, hurt those whom the Act seeks to help. See, e.g. , Brief for Leadership Conference on Civil and Human Rights et al. as Amici Curiae 21-29.
Another problem with the Secretary's rationale is that, even assuming the DOJ needed more detailed citizenship data, there were better ways of obtaining the needed data. The Census Bureau offered to provide the DOJ with data using administrative records, which, as I have pointed out, are likely just as accurate, if not more accurate, than responses to a citizenship question. The Census Bureau offered to provide this data at the census block level, which would resolve each of the DOJ's complaints about the existing ACS data. See Administrative Record 3289. But the Secretary rejected this alternative without explaining why it would not fully respond to the DOJ's request. That failure was particularly problematic given that the Census Act requires the Secretary to use other methods of obtaining demographic information if at all possible. See §§ 6(c), 195.
Normally, the Secretary would be entitled to place considerable weight upon the DOJ's expertise in matters involving the Voting Rights Act, but there are strong reasons for discounting that expertise here. The administrative record shows that DOJ's request to add a citizenship question originated not with the DOJ, but with the Secretary himself. See Administrative Record 3710. The Voting Rights Act rationale was in fact first proposed by Commerce Department officials. See ibid. DOJ officials, for their part, were initially uninterested in obtaining more detailed citizenship data, App. 414, and they agreed to request the data only after the Secretary personally spoke to the Attorney General about the matter, see Administrative Record 2651. And when the acting director of the Census Bureau proposed alternative means of obtaining better citizenship data, DOJ officials declined to meet to discuss the proposal. See id. , at 3460.
*2595Taken as a whole, the evidence in the administrative record indicates that the Voting Rights Act rationale offered by the Secretary was not just unconvincing, but pretextual. And, as the Court concludes, further evidence outside the administrative record but present in the trial record supports the finding of pretext. See Part V, ante. Among other things, that evidence reveals that the DOJ official who wrote the letter agreed that adding the question "is not necessary for DOJ's VRA enforcement efforts." App. 1113. And that official further acknowledged that he did not "know whether or not [citizenship] data produced from responses to the citizenship question ... will, in fact, be more precise than the [citizenship] data on which the DOJ is currently relying for purposes of VRA enforcement." Id. , at 1102.
The Court explains, and I agree, that a court normally should not "reject an agency's stated reasons for acting simply because the agency might also have had other unstated reasons." Ante , at 2573. But in this case, "the evidence tells a story that does not match the explanation the Secretary gave for his decision." Ante , at 2575. This evidence strongly suggests that the Secretary's stated rationale was pretextual. I consequently join Part V of the Court's opinion (except insofar as it concludes that the Secretary's decision was reasonable apart from the question of pretext). And I agree that the pretextual nature of the Secretary's decision provides a sufficient basis to affirm the District Court's decision to send the matter back to the agency.
* * *
I agree with the Court that the APA gives agencies broad leeway to carry out their legislatively delegated duties. And I recognize that Congress has specifically delegated to the Secretary of Commerce the authority to conduct a census of the population "in such form and content as he may determine." § 141(a). But although this delegation is broad, it is not without limits. The APA supplies one such limit. In an effort to ensure rational decisionmaking, the APA prohibits an agency from making decisions that are "arbitrary, capricious, [or] an abuse of discretion." 5 U. S. C. § 706(2)(A).
This provision, of course, does not insist that decisionmakers think through every minor aspect of every problem that they face. But here, the Secretary's decision was a major one, potentially affecting the proper workings of our democratic government and the proper allocation of hundreds of billions of dollars in federal funds. Cf. ante , at 2565 - 2566. Yet the decision was ill considered in a number of critically important respects. The Secretary did not give adequate consideration to issues that should have been central to his judgment, such as the high likelihood of an undercount, the low likelihood that a question would yield more accurate citizenship data, and the apparent lack of any need for more accurate citizenship data to begin with. The Secretary's failures in considering those critical issues make his decision unreasonable. They are the kinds of failures for which, in my view, the APA's arbitrary and capricious provision was written.
As I have said, I agree with the Court's conclusion as to pretext and with the decision to send the matter back to the agency. I do not agree, however, with several of the Court's conclusions concerning application of the arbitrary and capricious standard. In my view, the Secretary's decision-whether pretextual or not-was arbitrary, capricious, and an abuse of his lawfully delegated discretion. I consequently concur in the Court's judgment to the extent that it affirms the judgment of the District Court.
The annual alien registration requirement was repealed in 1981. See § 11, 95 Stat. 1617 (1981).
Several months after the Secretary made his decision, the Bureau updated its prediction to 5.8%, the figure the District Court later relied on in its standing analysis. See 351 F.Supp.3d 502, 579 (SDNY 2019).
The full text of subsections (a) and (b) provides:
"(a) The Secretary, whenever he considers it advisable, may call upon any other department, agency, or establishment of the Federal Government, or of the government of the District of Columbia, for information pertinent to the work provided for in this title.
"(b) The Secretary may acquire, by purchase or otherwise, from States, counties, cities, or other units of government, or their instrumentalities, or from private persons and agencies, such copies of records, reports, and other material as may be required for the efficient and economical conduct of the censuses and surveys provided for in this title." 13 U. S. C. § 6.
As a 2016 Census Bureau guidance document explained, obtaining citizenship statistics is "essential for agencies and policy makers setting and evaluating immigration policies and laws, understanding how different immigrant groups are assimilated, and monitoring against discrimination." Dept. of Commerce, Census Bureau, American Community Survey, Why We Ask: Place of Birth, Citizenship and Year of Entry, www2.census.gov/programs-surveys/acs/about/qbyqfact/2016/Citizenship.pdf (all Internet materials as last visited June 25, 2019).
United Nations, Dept. of Economic and Social Affairs Statistics Div., Principles and Recommendations for Population and Housing Censuses 163, 191 (rev. 3, 2017).
See, e.g. , Brief for Petitioners 29 (" '[O]ther major democracies inquire about citizenship on their census, including Australia, Canada, France, Germany, Indonesia, Ireland, Mexico, Spain, and the United Kingdom, to name a few' " (quoting App. to Pet. for Cert. 561a)).
We do not have before us a claim that information outside the administrative record calls into question the legality of an agency action based on an unstated, unlawful bias or motivation (e.g. , a claim of religious discrimination under the Free Exercise Clause). But to the extent such a claim is viable, the analysis would have nothing to do with the arbitrary-and-capricious review pressed by respondents. See §§ 706(2)(A)-(C) (addressing agency actions that violate "constitutional" or "statutory" requirements, or that "otherwise [are] not in accordance with law").
Insofar as Overton Park authorizes an exception to review on the administrative record, it has been criticized as having "no textual grounding in the APA" and as "created by the Court, without citation or explanation, to facilitate Article III review." Gavoor & Platt, Administrative Records and the Courts, 67 U. Kan. L. Rev. 1, 44 (2018) ; see id., at 22 (further arguing that the exception was "neither presented by the facts of the case nor briefed by the parties"). The legitimacy and scope of the exception-which by its terms contemplates only "administrative officials who participated in the decision ... giv[ing] testimony explaining their action," Overton Park , 401 U.S. at 420, 91 S.Ct. 814 -is an important question that may warrant future consideration. But because the Court's holding is incorrect regardless of the validity of the Overton Park exception, I will apply it here.