skip navigation

UTAH V. EVANS (01-714) 536 U.S. 452 (2002)
182 F. Supp. 2d 1165, affirmed.
[ Breyer ]
[ Scalia ]
[ O’Connor ]
[ Opinion of Thomas ]
HTML version
PDF version
HTML version
PDF version
HTML version
PDF version
HTML version
PDF version
HTML version
PDF version

Opinion of the Court

NOTICE:  This opinion is subject to formal revision before publication in the preliminary print of the United States Reports. Readers are requested to notify the Reporter of Decisions, Supreme Court of the United States, Washington, D. C. 20543, of any typographical or other formal errors, in order that corrections may be made before the preliminary print goes to press.


No. 01—714



[June 20, 2002]

    Justice Breyer delivered the opinion of the Court.

    The question before us is whether the Census Bureau’s use in the year 2000 census of a methodology called “hot-deck imputation” either (1) violates a statutory provision forbidding use of “the statistical method known as ‘sampling’ ” or (2) is inconsistent with the Constitution’s statement that an “actual Enumeration” be made. 13 U.S.C. § 195; U.S. Const., Art. I, §2, cl. 3. We conclude that use of “hot-deck imputation” violates neither the statute nor the Constitution.



    “Hot-deck imputation” refers to the way in which the Census Bureau, when conducting the year 2000 census, filled in certain gaps in its information and resolved certain conflicts in the data. The Bureau derives most census information through reference to what is, in effect, a nationwide list of addresses. It sends forms by mail to each of those addresses. If no one writes back or if the information supplied is confusing, contradictory, or incomplete, it follows up with several personal visits by Bureau employees (who may also obtain information on addresses not listed). Occasionally, despite the visits, the Bureau will find that it still lacks adequate information or that information provided by those in the field has somehow not been integrated into the master list. The Bureau may have conflicting indications, for example, about whether an address on the list (or a newly generated address) represents a housing unit, an office building, or a vacant lot; about whether a residential building is vacant or occupied; or about the number of persons an occupied unit contains. These conflicts and uncertainties may arise because no one wrote back, because agents in the field produced confused responses, or because those who processed the responses made mistakes. There may be too little time left for further personal visits. And the Bureau may then decide “imputation” represents the most practical way to resolve remaining informational uncertainties.

    The Bureau refers to different kinds of “imputation” depending upon the nature of the missing or confusing information. Where, for example, the missing or confused information concerns the existence of a housing unit, the Bureau speaks of “status imputation.” Where the missing or confused information concerns whether a unit is vacant or occupied, the Bureau speaks of “occupancy imputation.” And where the missing or confused information concerns the number of people living in a unit, the Bureau refers to “household size imputation.” In each case, however, the Bureau proceeds in a somewhat similar way: It imputes the relevant information by inferring that the address or unit about which it is uncertain has the same population characteristics as those of a “nearby sample or ‘donor’ ” address or unit–e.g., its “geographically closest neighbor of the same type (i.e., apartment or single-family dwelling) that did not return a census questionnaire” by mail. Brief for Appellants 7—8, 11. Because the Bureau derives its information about the known address or unit from the current 2000 census rather than from prior censuses, it refers to its imputation as “hot-deck,” rather than “cold-deck,” imputation.

    These three forms of imputation increased the final year 2000 count by about 1.2 million people, representing 0.4% of the total population. But because this small percentage was spread unevenly across the country, it makes a difference in the next apportionment of congressional Representatives. In particular, imputation increased North Carolina’s population by 0.4% while increasing Utah’s population by only 0.2%. And the parties agree that that difference means that North Carolina will receive one more Representative, and Utah will receive one less Representative, than if the Bureau had not used imputation but instead had simply filled relevant informational gaps by counting the related number of individuals as zero.


    After analyzing the census figures, Utah brought this lawsuit against the Secretary of Commerce and the Acting Director of the Census Bureau, the officials to whom the statutes delegate authority to conduct the census. 28 U.S.C. § 2284. Utah claimed that the Bureau’s use of “hot-deck imputation” violates the statutory prohibition against use of “the statistical method known as ‘sampling,’ 13 U.S.C. § 195 and is inconsistent with the Constitution’s statement that an “actual Enumeration” be made, Art. I, §2, cl. 3. Utah sought an injunction compelling the census officials to change the official census results. North Carolina intervened. The District Court found in the Census Bureau’s favor. 182 F. Supp. 2d 1165 (Utah 2001). Utah appealed. 28 U.S.C. § 1253. And we postponed consideration of jurisdiction pending hearing the case on the merits. 534 U.S. 1112 (2002).


    North Carolina argues at the outset that the federal courts lack the constitutional power to hear this case. Article III, §2 of the Constitution extends the “judicial Power” of the United States to actual “Cases” and “Controversies.” A lawsuit does not fall within this grant of judicial authority unless, among other things, courts have the power to “redress” the “injury” that the defendant allegedly “caused” the plaintiff. Lujan v. Defenders of Wildlife, 504 U.S. 555, 561 (1992); Allen v. Wright, 468 U.S. 737, 751 (1984). And, in North Carolina’s view, the courts cannot “redress” the injury that Utah claims to have suffered here. Hence Utah does not have the “standing” that the Constitution demands.

    In Franklin v. Massachusetts, 505 U.S. 788 (1992), this Court considered, and rejected, a similar claim. A private plaintiff had sued the Secretary of Commerce, challenging the legality of a 1990 census counting method as “arbitrary and capricious” and contrary to certain specific statutes. Id., at 790—791. That plaintiff sought to require the Secretary to recalculate the numbers and recertify the official results. The plaintiff hoped that would ultimately lead to a reapportionment that would assign an additional Representative to his own State.

    Eight Members of the Court found that the plaintiff had standing. Four Justices considered only whether the law permitted courts to review Census Bureau decisions under the Administrative Procedure Act. They concluded that it did. And they saw no further standing obstacle. Id., at 807 (Stevens, J., concurring in part and concurring in judgment).

    Four other Justices went further. They found that the controversy between the plaintiff and the Secretary was concrete and adversary. They said:

“The Secretary certainly has an interest in defending her policy determinations concerning the census; even though she cannot herself change the reapportionment, she has an interest in litigating its accuracy.” Id., at 803 (opinion of O’Connor, J.).

They also found that, as a practical matter, redress seemed likely. They said:

“[A]s the Solicitor General has not contended to the contrary, we may assume it is substantially likely that the President and other executive and congressional officials would abide by an authoritative interpretation of the census statute and constitutional provision . . . even though they would not be directly bound by such a determination.” Ibid.

They saw no further potential obstacle to standing. Ibid.

    We can find no significant difference between the plaintiff in Franklin and the plaintiff (Utah) here. Both brought their lawsuits after the census was complete. Both claimed that the Census Bureau followed legally improper counting methods. Both sought an injunction ordering the Secretary of Commerce to recalculate the numbers and recertify the official result. Both reasonably believed that the Secretary’s recertification, as a practical matter, would likely lead to a new, more favorable, apportionment of Representatives. Given these similarities, North Carolina must convince us that we should reconsider Franklin. It has not done so.

    North Carolina does not deny that the courts can order the Secretary of Commerce to recalculate the numbers and to recertify the official census result. Rather it points out that Utah suffers, not simply from the lack of a proper census “report” (a document), but more importantly from the lack of the additional congressional Representative to which North Carolina believes itself entitled as a consequence of the filing of that document. Whatever we may have said in Franklin, North Carolina argues, court-ordered relief simply cannot reach beyond the “report” and, here, a proper “report” cannot help bring about that ultimate “redress.”

    The reason North Carolina believes that court-ordered relief, i.e., the new document, cannot help is that, in its view, the statutes that set forth the census process make ultimate redress legally impossible. Those statutes specify that the Secretary of Commerce must “take a decennial census of population as of the first day of April” 2000, 13 U.S.C. § 141(a); he must report the results to the President by January 1, 2001, §141(b); the President must transmit to Congress by January 12, 2001, a statement showing the “whole number of persons in each State … and the number of Representatives to which each State would be entitled,” 2 U.S.C. § 2a(a); and, within 15 days of receiving that statement, the Clerk of the House of Representatives, must “send to the executive of each State a certificate of the number of Representatives to which such State is entitled,” §2a(b). The statutes also say that, once all that is done, each State “shall be entitled” to the number of Representatives that the “certificate” specifies “until the taking effect of a reapportionment under this section or subsequent statute.” Ibid.

    North Carolina points out that all of this was done by January 16, 2001. And North Carolina concludes that it is “entitled” to the number of Representatives that the “certificate” specifies (i.e., one more than Utah would like)–come what may.

    We disagree with North Carolina because we do not read these statutes so absolutely–as if they barred a certificate’s revision in all cases no matter what. The statutes themselves do not expressly say what is to occur should the “report” or the “statement” upon which the Clerk’s “certificate” rests turn out to contain, or to reflect, a serious mistake. The language is open to a more flexible reading that would permit correction of a certificate found to rest upon a serious error–say, a clerical, a mathematical, or a calculation error, in census data or in its transposition. And if that error is uncovered before new Representatives are actually selected, and its correction translates mechanically into a new apportionment of Representatives without further need for exercise of policy judgment, such mechanical revision makes good sense. In such cases, the “certificate” previously sent would have turned out not to have been a proper or valid certificate, it being understood that these statutes do not bar the substitution of a newer, more accurate version. Guided by Franklin, which found standing despite the presence of this statute, we read the statute as permitting “certificate” revision in such cases of error, and we include among them cases of court-determined legal error leading to a court-required revision of the underlying Secretarial “report.” So read, the statute poses no legal bar to “redress.”

    North Carolina adds that another statute, enacted after Franklin, nonetheless bars our consideration of this case. That statute authorizes “[a]ny person aggrieved by the use of any [unlawful] statistical method” to bring “a civil action” for declaratory or injunctive “relief against the use of such method.” Pub L. 105—119, Title II, §209(b), 111 Stat. 2481. North Carolina argues that this statute, by directly authorizing a lawsuit prior to conclusion of the census, implicitly forbids a lawsuit after its conclusion. And it supports this reading by pointing to a legislative finding that it would “be impracticable” to provide relief “after” that time. Id., §209(a)(8).

    This statute, however, does not say that it bars post-census lawsuits. It does not explain why Congress would have wished to deprive of its day in court a State that did not learn about a counting method’s representational consequences until after the census is complete–and hence had little, if any, incentive to bring a precensus action. Nor (as we have just explained), if a lawsuit is brought soon enough after completion of the census and heard quickly enough, is relief necessarily “impracticable.” We read limitations on our jurisdiction to review narrowly. See Webster v. Doe, 486 U.S. 592, 603 (1988); see also Bowen v. Michigan Academy of Family Physicians, 476 U.S. 667, 670 (1986). But see National Railroad Passenger Corporation v. National Assn. of Railroad Passengers, 414 U.S. 453 (1974) (special circumstances warrant reading statute as limiting the persons authorized to bring suit). We do not normally read into a statute an unexpressed congressional intent to bar jurisdiction that we have previously exercised. Franklin; Department of Commerce v. Montana, 503 U.S. 442 (1992). And we shall not do so here.

    Neither statute posing an absolute legal barrier to relief, we believe it likely that Utah’s victory here would bring about the ultimate relief that Utah seeks. Victory would mean a declaration leading, or an injunction requiring, the Secretary to substitute a new “report” for the old one. Should the new report contain a different conclusion about the relative populations of North Carolina and Utah, the relevant calculations and consequent apportionment-related steps would be purely mechanical; and several months would remain prior to the first post-2000 census congressional election. Under these circumstances, it would seem, as in Franklin, “substantially likely that the President and other executive and congressional officials would abide by an authoritative interpretation of the census statute and constitutional provision … .” 505 U.S., at 803 (opinion of O’Connor, J.).

    Moreover, in terms of our “standing” precedent, the courts would have ordered a change in a legal status (that of the “report”), and the practical consequence of that change would amount to a significant increase in the likelihood that the plaintiff would obtain relief that directly redresses the injury suffered. We have found standing in similar circumstances. See, e.g., Federal Election Comm’n v. Akins, 524 U.S. 11, 25 (1998) (standing to obtain court determination that the organization was a “political committee” where that determination would make agency more likely to require reporting, despite agency’s power not to order reporting regardless); Bennett v. Spear, 520 U.S. 154, 169—171 (1997) (similar in respect to determination of the lawfulness of an agency’s biological report); Metropolitan Washington Airports Authority v. Citizens for Abatement of Aircraft Noise, Inc., 501 U.S. 252, 264—265 (1991) (similar in respect to determination that transfer of airport control to local agency is unlawful). And related cases in which we have denied standing involved a significantly more speculative likelihood of obtaining ultimate relief. See Lujan, 504 U.S., at 564—565, n. 2 (obtaining ultimate relief “speculative”); Simon v. Eastern Ky. Welfare Rights Organization, 426 U.S. 26, 42 (1976) (same). We consequently conclude that Utah has standing here, and we have jurisdiction.


    Utah rests its statutory claim on a federal sampling statute which reads as follows:

“Except for the determination of population for purposes of apportionment of Representatives in Congress among the several States, the Secretary shall, if he considers it feasible, authorize the use of the statistical method known as ‘sampling’ … .” 13 U.S.C. § 195.

We have previously read this language as forbidding apportionment-related use of “the statistical method known as ‘sampling.’ Department of Commerce v. United States House of Representatives, 525 U.S. 316, 343 (1999). Utah claims that imputation, as practiced by the Census Bureau, is a form of that forbidden “sampling” method.

    The Government argues that imputation is not “sampling.” And it has used a simplified example to help explain why this is so. Imagine a librarian who wishes to determine the total number of books in a library. If the librarian finds a statistically sound way to select a sample (e.g., the books contained on every 10th shelf) and if the librarian then uses a statistically sound method of extrapolating from the part to the whole (e.g., multiplying by 10), then the librarian has determined the total number of books by using the statistical method known as “sampling.” If, however, the librarian simply tries to count every book one by one, the librarian has not used sampling. Nor does the latter process suddenly become “sampling” simply because the librarian, finding empty shelf spaces, “imputes” to that empty shelf space the number of books (currently in use) that likely filled them–not even if the librarian goes about the imputation process in a rather technical way, say by measuring the size of nearby books and dividing the length of each empty shelf space by a number representing the average size of nearby books on the same shelf.

    This example is relevant here both in the similarities and in the differences that it suggests between sampling and imputation. In both, “ ‘information on a portion of a population is used to infer information on the population as a whole.’ ” Brief for Appellants 18. And in Utah’s view, and that of Justice O’Connor, see post, at 4 (opinion concurring in part and dissenting in part), that similarity brings the Census Bureau imputation process within the relevant statutory phrase.

    On the other hand, the two processes differ in several critical respects: (1) In respect to the nature of the enterprise, the librarian’s sampling represents an overall approach to the counting problem that from the beginning relies on data that will be collected from only a part of the total population, Declaration of Howard Hogan ¶¶19—23, App. 257—259 (hereinafter Hogan); (2) in respect to methodology, the librarian’s sampling focuses on using statistically valid sample-selection techniques to determine what data to collect, ¶¶29—30, id., at 261—262; Declaration of Joseph Waksberg ¶¶6, 10, id., at 290—294 (hereinafter Waksberg); and (3) in respect to the immediate objective, the librarian’s sampling seeks immediately to extrapolate the sample’s relevant population characteristics to the whole population, Hogan ¶30, id., at 262; Declaration of David W. Peterson ¶8, id., at 352 (hereinafter Peterson).

    By way of contrast, the librarian’s imputation (1) does not represent an overall approach to the counting problem that will rely on data collected from only a subset of the total population, since it is a method of processing data (giving a value to missing data), not its collection, ¶¶21, 29, id., at 257—258, 261—262; it (2) does not rely upon the same statistical methodology generally used for sample selection, U.S. Dept. of Commerce, Decennial Statistical Studies Division, Census 2000 Procedures and Operations, Memorandum Series B—17, Feb. 28, 2001, id., at 194—196; Waksberg ¶¶6, 10, id., at 290, 293—294; and it (3) has as its immediate objective determining the characteristics of missing individual books, not extrapolating characteristics from the sample to the entire book population, Hogan ¶17, id., at 256—257; Peterson ¶9, id., at 352.

    These same differences distinguish Bureau imputation in the year 2000 census from “the statistical method known as ‘sampling.’ 13 U.S.C. § 195. The nature of the Bureau’s enterprise was not the extrapolation of the features of a large population from a small one, but the filling in of missing data as part of an effort to count individuals one by one. But cf. post, at 4 (O’Connor, J., concurring in part and dissenting in part) (suggesting the contrary). The Bureau’s methodology was not that typically used by statisticians seeking to find a subset that will resemble a whole through the use of artificial, random selection processes; but that used to assure that an individual unit (not a “subset”), chosen nonrandomly, will resemble other individuals (not a “whole”) selected by the fortuitous unavailability of data. L. Kish, Survey Sampling 26 (1965) (“In statistical literature [sampling] is generally synonymous with random sampling”). And the Bureau’s immediate objective was the filling in of missing data; not extrapolating the characteristics of the “donor” units to an entire population.

    These differences, whether of degree or of kind, are important enough to place imputation outside the scope of the statute’s phrase “the statistical method known as ‘sampling.’ ” For one thing, that statutory phrase–using the words “known as” and the quotation marks that surround “sampling”–suggests a term of art with a technical meaning. And the technical literature, which we have consequently examined, see Corning Glass Works v. Brennan, 417 U.S. 188, 201 (1974), contains definitions that focus upon differences of the sort discussed above. One text, for example, says that “[s]urvey sampling, or population sampling, deals with methods for selecting and observing a part (sample) of the population in order to make inferences about the whole population.” Kish, supra, at 18. Another says that “sample, as it is used in the [statistics] literature … means a subset of the population that is used to gain information about the entire population,” G. Henry, Practical Sampling 11 (1990), or, in other words, “a model of the population.” Ibid. Yet another says that a “sampling method is a method of selecting a fraction of the population in a way that the selected sample represents the population.” P. Sukhatme, Sampling Theory of Surveys with Applications 1 (1954). A 1953 treatise, to which Utah refers, says that a broader definition of “sample” is imprecise, adding that the term “should be reserved for a set of units … which has been selected in the belief that it will be representative of the whole aggregate.” F. Yates, Sampling Methods for Censuses and Surveys §1.1, p. 2 (2d rev. ed. 1953) (hereinafter Yates). And Census Bureau documents state that “professional statisticians” reserve the term “ ‘sample’ … for instances when the selection of the smaller population is based on the methodology of their science.” Report to Congress–The Plan for Census 2000, p. 3 (revised and reissued Aug. 1997) (thereinafter Report to Congress) 23.

    These definitions apply easily and naturally to what we called “sampling” in the librarian example, given its nature, methods, and immediate objectives. These definitions do not apply to the librarian’s or to the Bureau’s imputation process–at least not without considerable linguistic squeezing.

    For another thing, Bureau statisticians testified in the District Court that, in their expert opinion, Bureau imputation was not “sampling” as that term is used in the field of statistics. Hogan ¶¶18—30, App. 257—262; Waksberg ¶¶6—10, id., at 290—294 (former Bureau statistician). Their reasons parallel those to which we have referred. Ibid. Although Utah presented other experts who testified to the contrary, Utah has not relied upon their testimony or expert knowledge here. Insofar as the parties now rely on expert opinion, that opinion uniformly favors the

    Further, the history of the sampling statute suggests that Congress did not have imputation in mind in 1958 when it wrote that law. At that time, the Bureau already was engaged in what it called “sampling,” a practice that then involved asking a small subset of the population subsidiary census questions about, say, automobiles, telephones, or dishwashers, and extrapolating the responses to produce national figures about, say, automobile ownership. See M. Anderson, The American Census: A Social History 199 (1988) (discussing “long form” survey, sent in 1950 to about 20% of population). The Secretary of Commerce asked Congress to enact a law that would make clear the Bureau had legal authority to engage in this “practice.” Amendment of Title 13, United States Code, Relating to Census: Hearing on H. R. 7911 before the House Committee on the Post Office and Civil Service, 85th Cong., 1st Sess., 7 (1957) (Statement of Purpose and Need) (Secretary of Commerce, describing Bureau’s ability to obtain “some … information … efficiently through a sample survey … rather than a complete enumeration basis”). The Secretary did not object to a legislative restriction that would, in effect, deny the Bureau sampling authority in the area of apportionment. And Congress, in part to help achieve cost savings, responded with the present statute which provides that limited authority. See S. Rep. No. 698, 85th Cong., 1st Sess., 3 (1957) (“[P]roper use of sampling methods can result in substantial economies in census taking”); S. Rep. No. 94—1256, p. 5 (1976) (“use of sampling procedures and surveys . . . urged for the sake of economy and reducing respondent burden”).

    This background suggests that the “sampling” to which the statute refers is the practice that the Secretary called “sampling” at the time–for that is what Congress considered. And it suggests that the statutory word does not apply to imputation–for that is a matter that Congress did not consider. Indeed, had the Secretary believed that Congress intended to restrict the Bureau’s authority to engage in apportionment-related imputation, he would likely have expressed an objection, for the Bureau had used such imputation in the past and intended to use it in the future. Hogan ¶39, App. 266—267. Moreover, the Bureau’s rationale for using sampling was quite different from its rationale for using imputation. An advance plan to sample a subset saves money, for it restricts a survey’s potential scope. Bureau imputation does not save money, for the Bureau turns to imputation only after ordinary questionnaires and interviews have failed. Rather, imputation reflects a Bureau decision to spend at least a small amount of additional money in order to avoid placing the figure “zero” next to a listed address when it is possible to do better. See ¶34, id., at 264 (“The goal in Census 2000 was to conduct a census that was both numerically and distributively accurate”).

    Finally, Utah provides no satisfactory alternative account of the meaning of the phrase “the statistical method known as ‘sampling.’ ” Its arguments suggest that the phrase should apply to any use of statistics that would help the Bureau extrapolate from items about which the Bureau knows to other items, the characteristics of which it does not know. Brief for Appellants 9. But that definitional view would include within the statutory phrase matters that could not possibly belong there–for example, the use of statistics to determine whether it is better to ask a postal worker or a neighbor about whether an apparently empty house is occupied. And it would come close to forbidding the use of all statistics, not simply one statistical method (“sampling”). Utah’s express definitional statement–that “sampling” occurs whenever “information on a portion of a population is used to infer information on the population as a whole”–suffers from a similar defect. Indeed, it is even broader, coming close to a description of the mental process of inference itself. While the Census Bureau and at least one treatise have used somewhat similar language to define “sampling,” they have immediately added the qualification that such is the “layman’s” view, while professional statisticians, when speaking technically, speak more narrowly and more precisely. Report to Congress 23; Yates 1—2.

    Utah makes several additional arguments. It says that in House of Representatives, the Court found that two methods, virtually identical to imputation, constituted “sampling.” It says that the Bureau, if authorized to engage in imputation, might engage in wide-scale substitution of imputation for person-by-person counting. And it says that, in any event, the Bureau’s methods for imputing status and occupancy, see supra, at 2, are inaccurate.

    In our view, however, House of Representatives is distinguishable. The two instances of Bureau methodology at issue there satisfied the technical criteria for “sampling” in ways that the imputation here at issue does not. In both instances, the Bureau planned at the outset to produce a statistically sound sample from which it extrapolated characteristics of an entire population. In the first instance it did so by selecting census blocks randomly from which to extrapolate global census figures in order to compare (and adjust) the accuracy of figures obtained in traditional ways with figures obtained through statistical sampling. 525 U.S., at 325—326. In the second instance it used a sample drawn from questionnaire nonrespondents in particular census tracts in order to obtain the population figure for the entire tract. The “sampling” in the second instance more closely resembles the present effort to fill in missing data, for the “sample” of nonrespondents was large (about 20% of the tract) compared to the total nonresponding population (about 30% of the entire tract). Id., at 324—325. Nonetheless, we believe that the Bureau’s view of the enterprise as sampling, the deliberate decision taken in advance to find an appropriate sample, the sampling methods used to do so, the immediate objective of determining through extrapolation the size of the entire nonresponding population, and the quantitative figures at issue (10% of the tract there; 0.4% here), all taken together, distinguish it–in degree if not in kind–from the imputation here at issue.

    Nor are Utah’s other two arguments convincing. As to the first, Utah has not claimed that the Bureau has used imputation to manipulate results. It has not explained how census-taking that fills in ultimate blanks through imputation is more susceptible to manipulation than census-taking that fills in ultimate blanks with a zero. And given the advance uncertainties as to what States imputation might favor, manipulation would seem difficult to arrange. If Justice O’Connor’s speculation comes to pass–that the Bureau would decide, having litigated this case and utilized imputation in a subsequent census, to forgo the benefits of that process because of its results– the Court can address the problem at that time. As to the second, Utah’s claim concerns the nature of the imputation method, not its accuracy as applied–though we add that neither the record, see infra, at 21, nor Justice O’Connor’s opinion, see post, at 9, gives us any reason to doubt that accuracy here.

    We note one further legal hurdle that Utah has failed to overcome–the Bureau’s own interpretation of the statute. The Bureau, which recommended this statute to Congress, has consistently, and for many years, interpreted the statute as permitting imputation. Hogan ¶¶39, 41, 43, 46, 47, 52, App. 266—273. Congress, aware of this interpretation, has enacted related legislation without changing the statute. See, e.g., Census Address List Improvement Act of 1994, Pub. L. 103—430, 108 Stat. 4393; Foreign Direct Investment and International Financial Data Improvements Act of 1990, Pub. L. 101—533, 104 Stat. 2344; Act of Oct. 14, 1986, Pub. L. 99—467, 100 Stat. 1192. (Indeed, the Bureau told Congress of its planned use of imputation in the year 2000 census without meeting objection.) And the statute itself delegates to the Secretary the authority to conduct the decennial census “in such form and content as he may determine.” 13 U.S.C. § 141(a). Although we do not rely on it here, under these circumstances we would grant legal deference to the Bureau’s own legal conclusion were that deference to make the difference. Chevron U.S. A. Inc. v. Natural Resources Defense Council, Inc., 467 U.S. 837, 842—845 (1984).

    In sum, imputation differs from sampling in respect to the nature of the enterprise, the methodology used, and the immediate objective sought. And as we have explained, these differences are of both kind and degree. That the differences may be of degree does not lessen their significance where we are charged with interpreting statutory language and we are faced with arguments that suggest that it covers even the most ordinary of inferences. Since that cannot be so, we have found the keys to understanding the operative phrase in its history: the fact that the Bureau itself believed imputation to stand outside the prohibition it requested Congress pass, the fact that the Bureau has consistently used imputation, and the fact that Congress, on notice of that use, has not suggested otherwise. For these reasons, we conclude that the statutory phrase “the statistical method known as ‘sampling’ ” does not cover the Bureau’s use of imputation.


    Utah’s constitutional claim rests upon the words “actual Enumeration” as those words appear in the Constitution’s Census Clause. That Clause, as changed after the Civil War (in ways that do not matter here), reads as follows:

“Representatives and direct Taxes shall be apportioned among the several States … according to their respective Numbers … counting the whole number of persons in each State… . The actual Enumeration shall be made within three Years after the first Meeting of the Congress of the United States, … in such Manner as they shall by Law direct.” Art. I, §2, cl. 3 (emphasis added); see also Amdt. 14, §2.

Utah argues that the words “actual Enumeration” require the Census Bureau to seek out each individual. In doing so, the Bureau may rely upon documentary evidence that an individual exists, say a postal return, or upon eyewitness evidence, say by a census taker. It can fill in missing data through the use of testimonial reports, including secondhand or thirdhand reports, made by a family member, neighbor, or friend. But it may not rely upon imputation, which fills in data by assuming, for example, that an unknown house has the same population characteristics as those of the closest similar house nearby.

    We do not believe the Constitution makes the distinction that Utah seeks to draw. The Constitution’s text does not specify any such limitation. Rather the text uses a general word, “enumeration,” that refers to a counting process without describing the count’s methodological details. The textual word “actual” refers in context to the enumeration that will be used for apportioning the Third Congress, succinctly clarifying the fact that the constitutionally described basis for apportionment will not apply to the First and Second Congresses. The final part of the sentence says that the “actual Enumeration” shall take place “in such Manner as” Congress itself “shall by Law direct,” thereby suggesting the breadth of congressional methodological authority, rather than its limitation. See, e.g., Wisconsin v. City of New York, 517 U.S. 1, 19 (1996).

    The history of the constitutional phrase supports our understanding of the text. The Convention sent to its Committee of Detail a draft stating that Congress was to “regulate the number of representatives by the number of inhabitants, … which number shall … be taken in such manner as … [Congress] shall direct.” 2 M. Farrand, Records of the Federal Convention of 1787, pp. 178, 182—183 (rev. ed. 1966) (hereinafter Farrand). After making minor, here irrelevant, changes, the Committee of Detail sent the draft to the Committee of Style, which, in revising the language, added the words “actual Enumeration.” Id., at 590, 591. Although not dispositive, this strongly suggests a similar meaning, for the Committee of Style “had no authority from the Convention to alter the meaning” of the draft Constitution submitted for its review and revision. Powell v. McCormack, 395 U.S. 486, 538—539 (1969); see 2 Farrand 553; see also Nixon v. United States, 506 U.S. 224, 231 (1993). Hence, the Framers would have intended the current phrase, “the actual Enumeration shall be made … in such Manner as [Congress] … shall by Law direct,” as the substantive equivalent of the draft phrase, “which number [of inhabitants] shall … be taken in such manner as [Congress] shall direct.” 2 Farrand 183. And the Committee of Style’s phrase offers no linguistic temptation to limit census methodology in the manner that Utah proposes.

    Moreover, both phrases served to distinguish the census from the process of apportionment for the first Congress. Read in conjunction with the proceedings of the Constitutional Convention, the text of Article I makes clear that the original allocation of seats in the House was based on a kind of “conjectur[e],” 1 id., at 578—579, in contrast to the deliberately taken count that was ordered for the future. U.S. Const., Art. I, §2, cl. 3; 1 Farrand 602; 2 id., at 106; 2 The Founders’ Constitution 135—136, 139 (P. Kurland & R. Lerner eds. 1987) (hereinafter Kurland & Lerner); see also Department of Commerce, 503 U.S., at 448, and n. 15; post, at 11—13 (Thomas, J., concurring in part and dissenting in part) (describing colonial estimates). What was important was that contrast–rather than the particular phrase used to describe the new process.

    Contemporaneous general usage of the word “enumeration” adds further support. Late-18th-century dictionaries define the word simply as an “act of numbering or counting over,” without reference to counting methodology. 1 S. Johnson, A Dictionary of the English Language 658 (4th rev. ed. 1773); N. Bailey, An Etymological English Dictionary (26th ed. 1789) (“numbering or summing up”); see also Webster’s Third New International Dictionary __ (1993 ed.) (“the act of counting,” “a count of something (as a population)”). Utah’s strongest evidence, a letter from George Washington contrasting a population “estimate” with a “census” or “enumeration,” does not demonstrate the contrary, for one can indeed contrast, say a rough estimate, with an enumeration, without intending to encompass in the former anything like the Bureau’s use of imputation to fill gaps or clarify confused information about individuals. 31 Writings of George Washington 329 (J. Fitzpatrick ed. 1931); see 8 Writings of Thomas Jefferson 236 (A. Lipscomb ed. 1903) (comparing the “actual returns” with “conjectures”); 1 Farrand 602; 2 id., at 106; Kurland & Lerner 135—136. And the evidence Justice Thomas sets forth, post, at 11—13 (opinion concurring in part and dissenting in part), demonstrates the same. The kinds of estimates to which his sources refer are those based on “the number of taxable polls, or the number of the militia.” Post., at 7 (internal quotation marks omitted). Such sources show nothing other than that “enumeration” may be “incompatible (or at least arguably incompatible … ) with gross statistical estimates,” United States House of Representatives, 525 U.S., at 347 (Scalia, J., concurring in part), but such “gross statistical estimates” are not at stake here.

    Contemporaneous legal documents do not use the term “enumeration” in any specialized way. The Constitution itself, in a later article, refers to the words “actual Enumeration” as meaning “Census or Enumeration,” Art. I, §9, cl. 4, thereby indicating that it did not intend the term “actual Enumeration” as a term of art requiring, say, contact (directly or through third parties) between a census taker and each enumerated individual. The First Census Act uses the term “enumeration” almost interchangeably with the phrase “cause the number of the inhabitants … to be taken.” And the marshals who implemented that Act did not try to contact each individual personally, as they were required only to report the names of all heads of households. Act of Mar. 1, 1790, ch. 2, §1, 1 Stat. 102. Cf. House of Representatives, 525 U.S., at 347 (Scalia, J., concurring in part) (noting that the Census Acts of 1810 through 1950 required census workers to “visit each home in person”); see also post, at 17—18 (Thomas, J., concurring in part and dissenting in part).

    Of course, this last limitation suggests that the Framers expected census enumerators to seek to reach each individual household. And insofar as statistical methods substitute for any such effort, it may be argued that the Framers did not believe that the Constitution authorized their use. See House of Representatives, supra, at 346—349 (Scalia, J., concurring in part). But we need not decide this matter here, for we do not deal with the substitution of statistical methods for efforts to reach households and enumerate each individual. Here the Census Bureau’s method is used sparingly only after it has exhausted its efforts to reach each individual, and it does not differ in principle from other efforts used since 1800 to determine the number of missing persons. Census takers have long asked heads of households, “neighbors, landlords, postal workers, or other proxies” about the number of inhabitants in a particular place, Hogan ¶11, App. 253. Such reliance on hearsay need be no more accurate, is no less inferential, and rests upon no more of an individualized effort for its inferences than the Bureau’s method of imputation.

    Nor can Utah draw support from a consideration of
the basic purposes of the Census Clause. That Clause reflects several important constitutional determinations: that comparative state political power in the House would reflect comparative population, not comparative wealth; that comparative power would shift every 10 years to reflect population changes; that federal tax authority would rest upon the same base; and that Congress, not the States would determine the manner of conducting the census. See Wesberry v. Sanders, 376 U.S. 1, 9—14, and n. 34 (1964); 1 Farrand 35—36, 196—201, 540—542, 559—560, 571, 578—588, 591—597, 603; 2 id., at 2—3, 106; Kurland & Lerner 86—144; see The Federalist No. 54, pp. 336—341 (C. Rossiter ed. 1961) (J. Madison); id., No. 55, at 341—350 (J. Madison); id., No. 58, at 356—361 (J. Madison); 31 Writings of George Washington, supra, at 329. These basic determinations reflect the fundamental nature of the Framers’ concerns. Insofar as Justice Thomas proves that the Framers chose to use population, rather than wealth or a combination of the two, as the basis for representation, post, at 14—16, we agree with him. What he does not show, however, is that, in order to avoid bias or for other reasons, they prescribed, or meant to prescribe, the precise method by which Congress was to determine the population. And he cannot show the latter because, for the most part, the choice to base representation on population, like the other fundamental choices the Framers made, are matters of general principle that do not directly help determine the issue of detailed methodology before us. Declaration of Jack N. Rakove, in Department of Commerce v. United States House of Representatives, O. T. 1998, No. 98—404, p. 387 (“What was at issue … were fundamental principles of representation itself … not the secondary matter of exactly how census data was to be compiled”).

    Nonetheless, certain basic constitutional choices may prove relevant. The decisions, for example, to use population rather than wealth, to tie taxes and representation together, to insist upon periodic recounts, and to take from the States the power to determine methodology all suggest a strong constitutional interest in accuracy. And an interest in accuracy here favors the Bureau. That is because, as we have said, the Bureau uses imputation only as a last resort–after other methods have failed. In such instances, the Bureau’s only choice is to disregard the information it has, using a figure of zero, or to use imputation in an effort to achieve greater accuracy. And Bureau information provided in the District Court suggests that those efforts have succeeded. U.S. Dept. of Commerce, Economics and Statistics Admin., Census 2000 Informational Memorandum No. 110, App. 445—448 (concluding that postcensus research confirms that imputation appropriately included individuals in the census who would otherwise have been excluded).

    Of course, the Framers did not consider the imputation process. At the time they wrote the Constitution “statisticks” referred to “ ‘a statement or view of the civil condition of a people,’ ” not the complex mathematical discipline it has become. P. Cohen, A Calculating People 150—151 (1982). Yet, however unaware the Framers might have been of specific future census needs, say, of automobiles for transport or of computers for calculation, they fully understood that those future needs might differ dramatically from those of their own times. And they were optimists who might not have been surprised to learn that a year 2000 census of the Nation that they founded required “processed data for over 120 million households, including over 147 million paper questionnaires and 1.5 billion pages of printed material.” Hogan ¶8, App. 251. Consequently, they did not write detailed census methodology into the Constitution. As we have said, we need not decide here the precise methodological limits foreseen by the Census Clause. We need say only that in this instance, where all efforts have been made to reach every household, where the methods used consist not of statistical sampling but of inference, where that inference involves a tiny percent of the population, where the alternative is to make a far less accurate assessment of the population, and where consequently manipulation of the method is highly unlikely, those limits are not exceeded.

    For these reasons the judgment of the District Court is