Opinion of Breyer, J.

Justice Breyer, concurring in part and dissenting in part.

I join Part II of the majority opinion concerning standing, and I join Parts II and III of Justice Stevens's dissent. I also agree with Justice Stevens's conclusion in Part I that the plan for the 2000 census presented by the Secretary of Commerce is not barred by the Census Act. In my view, however, the reason that 13 U.S.C. § 195 does not bar the statistical sampling at issue here is that §195 focuses upon sampling used as a substitute for traditional enumeration methods, while the proposal at the heart of the Secretary's plan for the 2000 census (namely, Integrated Coverage Measurement, or ICM) is not so intended. Rather, ICM uses statistical sampling to supplement traditional enumeration methods in order to achieve the very accuracy that the census seeks and the Census Act itself demands. See, e.g., Decennial Census Improvement Act of 1991, §2(a)(1), 105 Stat. 635, note following 13 U.S.C. § 141 (directing the Secretary to contract with the National Academy of Sciences to study "means by which the Government could achieve the most accurate population count possible").

The language of §195 permits a distinction between sampling used as a substitute and sampling used as a supplement. The literal wording of its "except" clause focuses upon the use of sampling "for the determination of population for purposes of apportionment of Representatives in Congress among the several States." 13 U.S. C. §195 (emphasis added). One can read those words as the majority does-applying to apportionment-connected sampling irrespective of use or kind. But one can also read them as applicable only to the use of sampling in place of the traditional "determination of population for purposes of apportionment." The "except" clause does not necessarily apply to every conceivable use of statistical sampling any more than, say, a statutory rule forbidding "vehicles" in the park applies to everything that could possibly be characterized as a "vehicle." See generally H.L.A. Hart, The Concept of Law 124-136 (2d ed. 1994) (discussing the "open texture of law"). Context normally informs the meaning of a general statutory phrase and often limits its scope.

The history and context of §195 favors an interpretation that so limits the scope of that section. Cf. Brief for Appellants in No. 98-404, p. 36, n. 19; Brief for Appellees Gephardt et al. in No. 98-404, pp. 9-10, 22-23, 33-38; Young v. Klutznick, 497 F. Supp. 1318, 1335 (ED Mich. 1980) ("All that §195 does is prohibit the use of figures derived solely by statistical techniques. It does not prohibit the use of statistics in addition to the more traditional measuring tools to arrive at a more accurate population count"), rev'd on other grounds, 652 F.2d 617 (CA6 1981); Carey v. Klutznick, 508 F. Supp. 404, 415 (SDNY 1980) (Census Act permits sampling in the context of apportionment as long as it is used only in addition to more traditional methods of enumeration).     In the 1940's the Census Bureau began using statistical sampling in the collection of a variety of demographic information. U.S. Dept. of Commerce, Bureau of the Census, 200 Years of Census Taking: Population and Housing Questions, 1790-1990, p. 5 (Nov. 1989). Thus, during the 1940's and 1950's, each American family was asked to complete a short form containing a few information-gathering questions. In addition, the Bureau also used a long form that contained additional questions about individuals and families, but it asked only 1 family in 20 to complete this form. Ibid.; R. Jenkins, Procedural History of the 1940 Census of Housing and Population 13-15 (1985). The Census Bureau used those long-form answers, from 5 percent of the population, as a basis for extrapolating statistics and trends, about, say, unemployment or housing conditions, for the Nation as a whole.

In 1957 Congress focused upon this kind of sampling-a long form completed by only 1 American household in 20-as a model of what §195 would authorize the Secretary to do-"[e]xcept for the determination of population for purposes of apportionment." 13 U.S.C. § 195. When explaining the need for the proposed §195, the Secretary of Commerce spoke of a "sample enumeration or a sample census [that] might be substituted for a full census." Amendment of Title 13, United States Code, Relating to Census, Hearing on H. R. 7911 before the House Committee on Post Office and Civil Service, 85th Cong., 1st Sess., 7 (1957) (Statement of Purpose and Need) (emphasis added). He added that "[e]xperience has shown that some of the information which is desired in connection with a census could be secured efficiently through a sample survey . . . [and] that in some instances a portion of the universe to be included might be efficiently covered on a sample rather than a complete enumeration basis . . . ." Ibid. The House Report spoke in the same terms: "The purpose of section 195 in authorizing the use of sampling procedures is to permit the utilization of something less than a complete enumeration, as implied by the word 'census,' when efficient and accurate coverage may be effected through a sample survey." H. R. Rep. No. 1043, 85th Cong., 1st Sess., 10 (1957) (emphasis added); accord, S. Rep. No. 94-1256, p. 1 (1976) (1976 amendments added new language "to direct the Secretary . . . to use sampling and special surveys in lieu of total enumeration in the collection of statistical data whenever feasible" (emphasis added)). The discussion thus linked the authorization-and hence the exception-to sampling as a substitute for a headcount.

Census Bureau practice also helps to support this limited interpretation of the section's scope. Both before and after §195 was enacted in 1957, the census has used sampling techniques in one capacity or another in connection with its determination of population, most often as a quality check on the headcount itself. See, e.g., Declaration of Margo J. Anderson ¶12, App. in No. 98-404, p. 348 (first post-enumeration survey was performed following the 1950 census to check for inaccuracies).

The Census Bureau has also used a form of statistical estimation to adjust or correct its actual headcount. Since at least 1940, the Census Bureau has used an estimation process called "imputation" to fill in gaps in its headcount. U.S. Dept. of Commerce, Bureau of the Census, Report to Congress: The Plan for Census 2000, p. 23 (Aug. 1997) (hereinafter Census 2000 Report). When an enumerator believes a residence is occupied but is unable to obtain any information about how many people live there, the Census Bureau "imputes" that information based upon the demographics of nearby households. Imputation was responsible, for example, for adding 761,000 people to the Nation's total population in 1980 and 53,590 people in 1990. Ibid. In 1970, when the Census Bureau discovered at the last minute that it had mistakenly assumed that a significant number of housing units were vacant, it adjusted the headcount to add 1,068,882 people, or 0.5% of the total population. Ibid.

Integrated Coverage Measurement would not substitute for, but rather would supplement, a traditional headcount, and it would do so to achieve the basic purpose of the statutes that authorize the headcount-namely, accuracy. The Census Bureau has learned over time that certain portions of the population-for example, children, racial and ethnic minorities, and those who rent rather than own their homes-are systematically undercounted in a traditional headcount. Census 2000 Report 2-4; see also Wisconsin v. City of New York, 517 U.S. 1, 6-8 (1996). The ICM program is the Census Bureau's effort to correct for this problem. As I understand it, this proposal would use statistical sampling to check headcount results, State by State, by intensively investigating sample blocks in each State, comparing the results from that investigation with the results of the headcount, and using that information to estimate to what extent different groups of persons were undercounted during the headcount. The undercount rates-which will be calculated separately for every State in the Union-will then be used to adjust the headcount totals in an effort to correct for those inaccuracies.

I recognize that the use of statistical sampling to correct or reduce headcount inaccuracies is a complicated matter. An overall national improvement in accuracy does not necessarily tell the whole story. Apportionment demands comparable accuracy State by State. A count that reflected evenly distributed error (say, if the population in every State were undercounted by 20%) would produce the same congressional apportionment as a perfectly accurate count; a count that is less comparatively accurate could make matters worse. Although earlier attempts at ICM-like adjustments apparently failed to take some of these difficulties into account, the Secretary believes the present proposal does so. Census 2000 Report 30 (strata crossed state lines in 1990, but in 2000, strata will be defined on a state-by-state basis); cf. id., at 29 (explaining that the ICM methodology, which was used in the past two censuses to evaluate census quality, has "undergone substantial review and improvement" and "is generally accepted as the most reliable method to improve census results"). And, as I understand it, ICM will help to uncover and to correct undercounting not only among minority but also among majority populations. Any special emphasis the Census Bureau might place on including racial and ethnic minority neighborhoods among its samples would be justified as an effort to ensure proper counts among groups that history shows have been undercounted. Although some amici express concerns about the possibility of error in the execution of the statistical program, the Census Bureau itself, aware of potential difficulties, has created an expert panel of statisticians and social scientists, which will guide the Census Bureau's execution of its plan for the 2000 census, particularly with respect to its use of sampling. See Census 2000 Report 49-51. And, of course, unadjusted headcounts are also subject to error or bias-the very fact that creates the need for a statistical supplement. See, e.g., id., at 3-4 (describing the problem of differential undercount under the traditional headcount method); id., at 37 (without ICM, the 2000 census will be less accurate than the 1990 census).

Finally, as Justice Stevens points out, Congress has changed the statute considerably since it enacted §195 in 1957. Each change tends to favor the use of statistical sampling. In 1964, for example, Congress repealed former §25(c) of the Census Act, see Act of Aug. 31, 1964, 78 Stat. 737, which had required that each enumerator obtain "every item of information" through a personal visit to each household, 68 Stat. 1015, thereby permitting census taking by mail. In 1976, Congress amended §141(a) ("Population and other census information") to authorize the Secretary 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." At the same time, Congress strengthened §195's position on sampling, providing that the Secretary "shall" use sampling for purposes other than "for the determination of population for purposes of apportionment." 13 U.S.C. § 195. Given the legal need to interpret subsections of a single statute as creating a single coherent whole, these changes strengthen the case for an interpretation that restricts the scope of §195 to the kind and use of sampling that called it into being, placing beyond its outer limits a conceptually different (i.e., supplementary) use needed to achieve that statute's basic goal-greater census accuracy.

The Secretary's further proposal, the Nonresponse Followup program, uses statistical sampling not simply to verify a headcount but to determine the last 10% of population in each census tract. I concede that this kind of statistical "follow-up" is conceptually similar to the kind of sampling that was before Congress in 1957, in the sense that it involves determining a portion of the total population based upon a sample. But one can consider it supplementary for a different reason-because it simply does not have a great enough impact upon the headcount to be considered a "substitute" falling within §195's "except" clause.

I note that the Census Bureau has never relied exclusively upon headcounts to determine population. As discussed above, for example, the Census Bureau has supplemented its headcounts with imputation to some degree for at least the last 50 years. Section 195 of the Census Act, at least in my view, could not have been intended as a prohibition so absolute as to stop the Census Bureau from imputing the existence of a living family behind the closed doors of an apparently occupied house, should that family refuse to answer the bell. Similarly, I am not convinced that the Act prevents the use of sampling to ascertain the existence of a certain number of the families that fail to mail back their census forms.

The question, then, is what "number" of housing units will be assigned a population through sampling. Whether the Nonresponse Followup program is sufficiently like imputation in terms of its degree of impact so as to be a supplement to the headcount-or rather whether it is more like the way in which the Bureau uses sampling in connection with the "long form," as a substitute for a headcount-is here a matter of degree, not kind. Is the use of that method in the Nonresponse Followup, limited to the last 10%, sufficiently small, as a portion of the total population, and sufficiently justified, through the need to avoid disproportionately prohibitive costs, that it remains, effectively, a "supplement" to the traditional headcount?

For each census tract (made up of roughly 1,700 housing units), the Nonresponse Followup program will assign population figures to no more than 170 housing units. Census Bureau enumerators will personally visit enough of the housing units in each census tract to ensure that 90% of all housing units have been counted either by mail or in person. The Census Bureau will then use the information gathered from the housing units that the census enumerators actually visited in that tract to arrive at a number for the remaining 10%. See generally Census 2000 Report 26-29. The primary advantage of this program is financial; it is considerably cheaper than a personal search by enumerators to take account of the last few of the households that do not respond by mail. See, e.g., National Research Council, Panel to Evaluate Alternative Census Methods, Counting People in the Information Age 100 (D. Steffey & N. Bradburn eds. 1994). But the Secretary also believes that this program addresses other concerns-concerns related to the immense difficulties involved in personally visiting every home that does not respond by mail-and that, overall, the Nonresponse Followup plan "will increase the accuracy of the census as a whole." Reply Brief for Appellants in No. 98-564, p. 4; see also Census 2000 Report 27; id., at 7 (quoting the National Academy of Sciences Panel on Requirements as concluding that "[i]t is fruitless to continue trying to count every last person with traditional Census methods of physical enumeration").

In answering the question whether this use of sampling remains a "supplement" because of its limited impact on the total headcount, I would give considerable weight to the views of the Secretary, to whom the Act entrusts broad discretionary authority. See 13 U.S.C. § 141(a). The Secretary's decision to draw the line at the last 10%, rather than at the last 5% or 1%, of each census tract's population may well approach the limit of his discretionary authority. But I cannot say that it exceeds that limit. Consequently, I would not set aside the Census Bureau's Nonresponse Followup proposal on this basis.

For these reasons, I respectfully dissent.