P.E. BAZEMORE, ET AL., PETITIONERS V. WILLIAM C. FRIDAY, ET AL. UNITED STATES OF AMERICA, ET AL., PETITIONERS V. WILLIAM C. FRIDAY, ET AL. No. 85-93 and No. 85-428 In the Supreme Court in the United States October Term, 1985 On Writ of Certiorari to the United States Court of Appeals for the Fourth Circuit Brief for the Federal Petitioners TABLE OF CONTENTS Opinions below Jurisdiction Statutes involved Questions presented Statement Summary of argument Argument: I. Black state employees establish a claim under Title VII by identifying current salary disparities between themselves and white employees holding the same jobs and demonstrating that such disparities result from a state policy before 1965 of paying blacks lower salaries than whites II. The regression analyses showed racial discrimination, and respondents did not refute that showing A. In order to establish a prima facie of salary discrimination, a regression analysis must control for factors that normally affect salary B. The court of appeals erred in analyzing the statistical proof offered in this case III. The service retains joint responsibility for the selection of county chairmen, and is therefore liable under Title VII for discrimination in those selections IV. Prior segregation in the 4-H and extension homemaker clubs was fully cured by respondents' adoption of a genuinely nondiscriminatory admissions policy Conclusion OPINIONS BELOW The opinion of the court of appeals (Pet. App. 346a-481a) /1/ is reported at 751 F.2d 662. The opinions of the district court (Pet. App. 3a-207a, 216a-345a) are unreported. JURISDICTION The judgment of the court of appeals (Supp. Pet. App. 1a-3a) was entered on December 10, 1984. Rehearing was denied on April 15, 1985 (Pet. App. 482a-483a). On July 5, 1985, the Chief Justice extended the Government's time for filing a petition for a writ of certiorari to and including September 12, 1985. The petition in No. 85-93 was filed on July 15, 1985, and the petition in No. 85-428 was filed on September 12, 1985. Both petitions were granted on November 12, 1985 (J.A. 181-182). The jurisdiction of this Court is invoked under 28 U.S.C. 1254(1). STATUTES INVOLVED The relevant portions of Section 601 of Title VI of the Civil Rights Act of 1964, 42 U.S.C. 2000d, and Section 703(a)(1) of Title VII of the Civil Rights Act of 1964, 42 U.S.C. 2000e-2(a)(1), are reproduced at pages 3-4 of the petition in No. 85-93. QUESTIONS PRESENTED 1. Whether black state employees establish a claim under Section 703(a) of the Civil Rights Act of 1964, 42 U.S.C. 2000e-2(a), by identifying current salary disparities between themselves and white employees holding the same jobs and demonstrating that such disparities result from a pre-1965 state policy of paying blacks lower salaries than whites. 2. Whether the court below was incorrect in concluding that the statistical evidence of salary disparities in this case failed to meet the petitioners' burden of accounting for variables which ought to be reasonably viewed as determinants of salary. 3. Whether respondents are liable under Title VII, on the basis of their joint responsibility for final selections of employees to fill county chairman positions, for discrimination in those selections. 4. Whether, in providing financial assistance to voluntary single-race clubs, respondents comply with Title VI of the Civil Rights Act of 1964, 42 U.S.C. 2000d et seq., and the Fourteenth Amendment by requiring that membership in those clubs be open to all without regard to race. 5. Whether a plaintiff class should have been certified in this case. /*/ STATEMENT 1. The North Carolina Agricultural Extension Service (the Service) provides services to state residents involving the dissemination of "useful and practical information on subjects relating to agriculture and home economics," e.g., through educational programs for farmers and sponsoring 4-H and extension homemaker clubs (Pet. App. 7a, 12a-20a). It is funded jointly by the United States Department of Agriculture, the State of North Carolina, and the various counties in the State (Pet. App. 7a-8a). The Service employs agricultural extension agents, professional employees at the county level, of which there are three ranks: full agent, associate agent, and assistant agent (Pet. App. 17a). The three ranks perform "essentially the same types of tasks," but the full agents have more responsibility and are expected to maintain higher performance levels than associate agents, the intermediate position, or assistant agents, the entry level position (ibid.). /2/ Until August 1965, the Service was divided into "a white branch * * * and a Negro branch * * * composed entirely of black personnel and serv(ing) only black farmers, homemakers and youth" (Pet. App. 27a). Although black and white county agents had identical responsibilities and job descriptions (Pet. App. 29a), "(t)he salaries of black agents in the segregated system were lower than the salaries of their white counterparts" (Pet. App. 30a). The two branches of the Service were merged on August 1, 1965 (Pet. App. 30a, 359a), and a single minimum entry level salary was adopted for all agents hired after the merger. /3/ Shortly before the merger, the position of county extension chairman was created by making the white county agents responsible for coordinating the entire extension program in their respective counties (C.A. App. 1001-1002, 1783). In November 1972, the Service introduced a system of announcing job vacancies and accepting applications for county chairman positions (Pet. App. 24a-25a, 75a). /4/ Applicants who possess the minimum qualifications for county chairman are interviewed by Service officials, who then make a recommendation to the board of county commissions (Pet. App. 25a-26a, 76a-77a). The county generally accepts the recommendation (C.A. App. 171), but "all appointments are worked out jointly between the Extension Service and the commissioners and no official action can be taken unilaterally by either party with respect to filling a vacancy" (Pet. App. 77a). Prior to 1965, the Service had established separate all-white and all-black 4-H and extension homemaker clubs, and many clubs presently have only members of a single race (J.A. 103; C.A. App. 1807), although the number of integrated Clubs increased nearly three-fold between 1972 and 1980 (C.A. App. 1807, 180, 1140). After 1965, the Service requested a formal assurance from each club that it would not discriminate on the basis of race, color, or national origin (GX 115, at 3). The Service has also published in the media its policy that all voluntary clubs be organized without regard to race, instructed its agents to encourage formation of new clubs on that basis (Pet. App. 181a), and integrated all other aspects on the 4-H program. 2. a. This suit was initiated in November 1971 by more than 50 black employees of the Service, alleging, inter alia, intentional racial discrimination in employment and services in violation of the Fourteenth Amendment to the Constitution, 42 U.S.C. 1981, and Section 601 of Title VI of the Civil Rights Act of 1964, 42 U.S.C. 2000d (Pet. App. 3a-4a). After Title VII of the Civil Rights Act of 1964 became applicable to the states in 1972, the complaint was amended to include claims under Section 703 of that Act, 42 U.S.C. 2000e-2, and the United States intervened in the action. The complaint in intervention, as amended, also alleged racial discrimination against black employees and recipients of services in violation of the Fourteenth Amendment, Title VI, and Title VII (Pet. App. 5a, 35a). Plaintiffs asserted intentional racial discrimination in various incidents of employment, including salaries, job assignments and promotions, and in the selection of county chairmen, as well as in the continued support of single-race 4-H and extension homemaker clubs (Pet. App. 49a-51a). The employment-related claims included individual claims of discriminatory treatment as well as allegations of intentional patterns and practices of discrimination (Pet. App. 49a-51a, 227a-339a). b. The case was tried to the court for 10 weeks starting in December 1981 (Pet. App. 4a). During discovery, the defendants had asserted that four factors were determinative of salary: education, tenure, job title and job performance (GX 159, at 90, 96 (Oct. 16, 1981, Deposition of Dr. Paul Dew, Assistant Director of County Operations)). At trial, the evidence introduced by the United States included multiple regression analyses comparing the salaries of black and white county agents in 1974, 1975 and 1981. Certain of these regressions used four independent variables -- race, education, tenure, and job title -- and showed a statistically significant racial effect for 1974 (C.A. App. 1601, 402-403) and 1975 (C.A. App. 1589, 416), and a smaller racial effect, without statistical significance, for 1981 (C.A. App. 1578). /5/ The district court indicated that, based on this evidence, plaintiffs would prevail unless defendants produced statistical evidence demonstrating that the addition of other variables would reduce or eliminate the racial effect. /6/ Accordingly, defendants also introduced multiple regression analyses, for 1975 and 1981. Certain of these regressions used the same variables as plaintiffs' regressions, but excluded county chairmen from the data base, and these analyses produced results similar to plaintiffs' (C.A. App. 1716 (analyses), 1711-1712, 1691-1692). In addition, defendants presented regressions that added quartile rank, a measure of job performance, as an independent variable. /7/ Doing so for 1981 produced statistically insignificant racial disparities, but doing so for 1975 increased the racial effect, and this result was statistically significant (C.A. App. 1716 (analysis 6), 1713-1714)). Plaintiffs also introduced evidence showing that although the size of the 4-H club system has varied considerably over the years, there have been more than 1,000 all-white clubs each year since 1972 (C.A. App. 2237; GX 11), and more than 850 single-race clubs in communities identified by defendants as "ethnically mixed" (C.A. App. 1807). The extension homemaker clubs also remain largely single-race clubs (J.A. 103-113; C.A. App. 1806-1807; Tr. 941-942, 1524-1525, 2390, 2449-2450). /8/ c. The district court rejected all claims of the private plaintiffs and the United States. With respect to the salaries of county-level employees, the court held that no pattern or practice of racially discriminatory treatment had been shown (Pet. App. 150a). The court noted that "(i)t is undisputed" that before the merger of its black and white branches in 1965, the Service paid black agents less than white ones, and recognized that, although "steps were taken to begin (the) elimination" of this disparity before the Service was covered by Title VII in 1972, "the government has offered evidence tending to show that as of January 1973, the salaries of numerous black agents throughout the system were less than those of white agents in the same counties who were in comparable or lower positions and who had comparable or less tenure * * * (and) defendants' own exhibit (showed) some salary disparities between blacks and whites as late as October, 1974" (Pet. App. 120a-121a). The district court nevertheless found that "while on its face (the) evidence unquestionably establishes salary disparities, when viewed in the light of defendants' explanatory evidence it fails to prove discrimination" (Pet. App. 122a-123a). /9/ Focusing on certain regressions that controlled only for tenure and education (Pet. App. 131a), rather than on those also including job title and job performance as independent variables, the court described the regression analyses as flawed, primarily because the raw data on which they were based included the salaries of higher-paid county chairmen, most of whom were white, as well as county agents (Pet. App. 136a-138a), and because they failed to account for "several unmeasured factors, notably job performance" (Pet. App. 141a). /10/ In sum, the court "conclude(d) that the plaintiffs had probably made out a prima facie case with respect to defendants' promotion and salary practices * * * (but) the defendant * * * articulat(ed) plausible reasons for its actions * * * which the court found convincing" (Pet. App. 190a). As to the claim that the Service permitted segregated 4-H clubs and extension homemaker clubs to be maintained in North Carolina, recognizing and providing services to such clubs, the district court found that "there are many clubs to which members of both races belong" (Pet. App. 165a), and that "(i)f any individual has become a member of a club composed only of members of his or her own race, it has been an entirely voluntary act" (Pet. App. 172a). The court found no evidence of any denial of membership or discrimination in services on the basis of race, and concluded that the law does not require that these clubs be integrated (Pet. App. 165a-185a). The court ruled that the evidence did not demonstrate any discriminatory intent on the Service's part in tolerating the single-race clubs (Pet. App. 179a-182a), and that the Service accordingly did not violate the law in continuing to provide services to such clubs (Pet. App. 184a-185a). /11/ 3.a. The court of appeals affirmed (Pet. App. 346a-425a), Judge Phillips dissenting in part (Pet. App. 425a-481a). The panel majority adopted the district court's findings that under policies in effect when the Service maintained two separate racially segregated branches, black employees were paid less than white employees performing the same job because of their race, and that even after the Service became subject to Title VII in 1972, "(s)ome pre-existing salary disparities continued to linger on" (Pet. App. 360a). However, the court stated succinctly (Pet. App. 380a): The plaintiffs claim that the pre-act discriminatory difference in salaries should have been affirmatively eliminated but has not. We do not think this is the law. The panel majority relief for its view on United Air Lines, Inc. v. Evans, 431 U.s. 553 (1977), and Hazelwood School District v. United States, 433 U.S. 299 (1977), as well as several court of appeals decisions that followed Evans in rejecting time-barred claims despite the continuing effects of the alleged discriminatory acts on seniority rights (Pet. App. 380a-382a). This view of Title VII's requirement led the panel to fault all of the regression analyses of the salaries of current county employees -- recent hires as well as pre-Act hires -- because the figures analyzed "reflect the effect of pre-Act discrimination" (Pet. App. 389a). For this reason, as well as because "both experts omitted from their respective analys(e)s variables which ought to be reasonably viewed as determinants of salary," the analyses were deemed "unacceptable as evidence of discrimination" (Pet. App. 391a). With respect to the selection of county chairmen, plaintiffs challenged on appeal the district court's analysis of selection rates, arguing that the court erred in excluding vacancies for which only whites applied, while including vacancies for which only blacks applied (Pet. App. 411a-412a). The court of appeals majority found it unnecessary to consider these objections, because it concluded that "the employment decisions made by the Extension Service with respect to the selection of County Chairmen were made when the Service either recommended or did not recommend an applicant for an existing vacancy to the County Commissions" (Pet. App. 405a-406a). It therefore examined the data as to the Service's recommendations, rather than the selection statistics relied on by plaintiffs and the district court, and found no discrimination (Pet. App. 418a-423a). /12/ Finally, the panel majority held that the district court "correctly denied the plaintiffs' claim with respect to the alleged affirmative duty to require integrated membership" in 4-H clubs and extension homemaker clubs because, absent any proof of discrimination, "the mere existence of all white and all black * * * (c)lubs in some racially mixed communities" does not violate the law (Pet. App. 424a n.128). b. Judge Phillips dissented from the majority's dismissal of the salary claims (Pet. App. 425a, 433a-469a), /13/ noting that it was undisputed that prior to the 1965 merger "the salaries of black professionals were intentionally and quite openly simply set lower than those of white colleagues in the same employment positions" (Pet. App. 437a), and that these salary differential continued "well past 1968 (the earliest limitation date applicable to the salary claim)" (Pet. App. 438a-439a). /14/ The regression analyses of both plaintiffs' and defendants' experts were, moreover, in his view, "wholly consistent" in showing a substantial, across-the-board race-based disparity (Pet. App. 449a-450a). Because these analyses "employ(ed) the most obvious alternative variables of tenure, education, and job position" (Pet. App. 449a), Judge Phillips found no authority for rejecting such analyses "for failure to include a number of other independent variables merely hypothesized by defendants" (Pet. App. 448a). In sum, Judge Phillips concluded that "the only rational assessment to be made of the evidence in this record" is that "the general pattern of pre-1965 overt discrimination in salary continued in substantial, if gradually diminishing, degree until at least 1976 and perhaps beyond," and that responsible Service officials knew that such a race-based pattern continued and failed to correct it (Pet. App. 455a-456a). In his view, the majority's failure to award relief on such a record resulted from "misapprehensions of controlling legal principle" (Pet. App. 456a), including the "relevant time frame within which the existence of a pattern or practice of salary discrimination was to be assessed" (Pet. App. 457a). /15/ 4. Rehearing en banc was denied by an equally divided court, without opinion, on April 15, 1985, and the panel, Judge Phillips again dissenting, also declined to rehear the case (Pet. App. 482a-483a). SUMMARY OF ARGUMENT Hiring and promotion decisions are discrete acts, which, if taken before the effective date of Title VII or outside of the applicable statute of limitations, cannot be the subject of a successful Title VII suit, even if the consequences of those actions continue to affect the employee until the time of suit. But Title VII does require the correction of unequal salaries that are the continuation of racially-based pay differentials originating pre-Act or in the time-barred period. This continuing salary discrimination is akin to intentionally discriminatory seniority systems, which, regardless of the date of their inception, afford no justification for race-based disparate treatment. With regard to the statistical evidence, the court of appeals articulated the correct rule -- that in a disparate treatment case challenging salary differences, plaintiff's statistical analysis must include "variables which ought to be reasonably viewed as determinants of salary" (Pet. App. 391a) -- but improperly applied that rule. The parties' multiple regression analyses, together with the other evidence introduced, proved that there was racial discrimination in salaries. None of the variable omitted from petitioners' analyses undermines that conclusion. As a corollary to the "reasonableness" standard for statistical proof, we urge that district courts be encouraged to make formal determinations at the earliest possible stage of the proceedings of the factors to be included in statistical analyses offered at trial. Although private petitioners ask this Court to consider whether Title VII permits an employer to delegate its hiring decisions to a third party that invariably acts in a discriminatory manner (85-93 Pet. 49-55), this case presents no such issue. It is clear that the Service was jointly responsible with the county commissioners for the selection of county chairmen; even the court of appeals recognized this (Pet. App. 403a). The court's reliance on statistics relating to the recommendations made by respondents, rather than the final selection statistics, was accordingly inappropriate. Finally, the court of appeals correctly found that respondents had satisfied their affirmative duty to desegregate the 4H and homemaker extension clubs by maintaining and publicizing a policy of entirely open admissions to such clubs. Although some single-race clubs remain, neither the Constitution nor Title VI requires any particular degree of racial balance, and maintenance of the traditional option of individuals to join any club that they choose does not suggest that respondents are perpetuating their prior segregative practices. ARGUMENT I. BLACK STATE EMPLOYEES ESTABLISH A CLAIM UNDER TITLE VII BY IDENTIFYING CURRENT SALARY DISPARITIES BETWEEN THEMSELVES AND WHITE EMPLOYEES HOLDING THE SAME JOBS AND DEMONSTRATING THAT SUCH DISPARITIES RESULT FROM A STATE POLICY BEFORE 1965 OF PAYING BLACKS LOWER SALARIES THAN WHITES. The court of appeals acknowledged that, before the merger in 1965, the Service maintained two separate racially-segregated branches and paid black employees less than white employees because of their race; that, after the merger, these race-based disparities were not immediately eliminated; and that these disparities continued after this suit was filed and after Title VII becamse applicable to the Service in March 1972 (Pet. App. 359a-360a, 389a-390a). As a result, since the effective date of the Act, black employees hired before 1965, because of their race, have received and continue to receive lower salaries than white employees who have been performing the same job for the same length of time. The court of appeals incorrectly decided that Title VII provides no remedy to these black employees (Pet. App. 380a-382a, 399a-400a). The court of appeals relied in large part on this Court's decisions in United Air Lines, Inc. v. Evans, 431 U.S. 553 (1977), and Hazelwood School District v. United States, 433 U.S. 299 (1977), interpreting those decisions as absolving an employer of any responsibility for affirmatively eliminating the continuing effects of pre-Act salary discrimination provided it has adopted a race-neutral policy in establishing salaries for post-Act hires. The appellate court's reliance is misplaced, however, as this Court's decisions in Evans and Hazelwood are readily distinguishable from the case at bar. In Evans, a female flight attendant forced to resign when she married in 1968, and rehired in 1972 after the "no marriage" policy for female flight attendants was discontinued, challenged United's refusal to credit her prior service towards her seniority. Noting the absence of any allegation that prior service is credited to rehired male employees under United's seniority system, this Court acknowledged that the denial of pre-1972 seniority "does indeed have a continuing impact on (Evans') pay and fringe benefits" (431 U.S. at 558), but rejected her claim that United was guilty of a continuing violation of Title VII, stating (id. at 558, 560): (T)he seniority system gives present effect to a past act of discrimination (the forced resignation). But United was entitled to treat that past act as lawful after respondent failed to file a charge of discrimination within the (statutory limitations period). A discriminatory act which is not made the basis for a timely charge is the legal equivalent of a discriminatory act which occurred before the statute was passed. It may constitute relevant background evidence in a proceeding in which the status of a current practice is at issue, but separately considered, it is merely an unfortunate event in history which has no present legal consequences. * * * * * The statute does not foreclose attacks on the current operation of seniority systems which are subject to challenge as discriminatory. But such a challenge to a neutral system may not be predicated on the mere fact that a past event which has no present legal significance has affected the calculation of seniority credit, even if the past event might at one time have justified a valid claim against the employer. The Court's decision in Evans thus turned on the fact that United's past act of discrimination -- forcing Evans to resign because she was a married female -- was a single, discrete act taken at a time outside of the applicable statutory limitations period, and as such was not actionable under Title VII. The Court acknowledged, as the court of appeals assumed here, that the Evans rule is equally applicable when the discriminatory act was taken before the effective date of Title VII. To similar effect is the Court's decision in Hazelwood. There this Court vacated a court of appeals judgment, based on statistical disparities between the racial composition of Hazelwood's teaching staff and that of the qualified public school teacher population in the relevant labor market, that the school district had engaged in a pattern and practice of hiring discrimination in violation of Title VII. Although this Court agreed that the court of appeals correctly rejected the district court's statistical analysis, it held that the statistical disparities on which the appellate court relied were not dispositive (433 U.S. at 309-310 (footnote omitted)): The Court of Appeals totally disregarded the possibility that this prima facie statistical proof in the record might at the trial level be rebutted by statistics dealing with Hazelwood's hiring after it became subject to Title VII. Racial discrimination by public employers was not made illegal under Title VII until March 24, 1972. A public employer who from that date forward made all its employment decisions in a wholly nondiscriminatory way would not violate Title VII even if it had formerly maintained an all-white work force by purposefully excluding Negroes. For this reason, the Court cautioned in the Teamsters opinion (International Brotherhood of Teamsters v. United States, 431 U.S. 324 (1977)) that once a prima facie case has been established by statistical workforce disparities, the employer must be given an opportunity to show that "the claimed discriminatory pattern is a product of pre-Act hiring rather than unlawful post-Act discrimination." 431 U.S., at 360. Thus, Hazelwood indicates that Title VII is not violated by disparities in the racial composition of an employer's staff which are the present effects of discriminatory hiring decisions, all of which occurred before the effective date of Title VII. Evans and Hazelwood thus establish that pre-Act or time-barred hiring and termination decisions cannot from the basis of a claim under Title VII, even when those decisions have continuing current effects due, for example, to the operation of a bona fide seniority system. /16/ Similarly, pre-Act or time-barred promotion decisions cannot be challenged under Title VII on the theory that the claimant who should have received the promotion now continues in a lesser job at a salary level below that which he would have obtained had he not been the victim of the pre-Act or time-barred discrimination. /17/ Hiring and promotion decisions are discrete acts, taken once and for all at a single moment in time. Evans and Hazelwood teach that if that moment occurred prior to the effective date of the Act or beyond the reach of the statute of limitations, the discriminatory decision cannot be the subject of a Title VII suit, even though the consequences of that decision may well continue to the present. Indeed, it could scarcely be otherwise unless the Act is, as a practical matter, to have retroactive application, and relief for past illegalities is to be available into the indefinite future. For better or worse, unlawful discriminatory hiring, promotion and termination decisions must either be timely complained of or be taken to have fixed a person's situation once and for all -- unless, of course, fresh illegalities are subsequently committed. No such practical and conceptual difficulties attend the correction of unequal salaries that are the continuation of pre-Act, admittedly racially-based pay differentials. One claim in the case at bar is that the Service has continued to pay certain black employees less than white employees holding the same job for the same length of time, simply because of their race. Until now, the courts of appeals -- including the Fourth Circuit -- have consistently held that such discrimination in compensation is a continuing violation of Title VII, and as such is actionable notwithstanding Evans and Hazelwood, even when the pay differentials originated before the effective date of the Act or outside of the statutory limitations period. Thus, for example, the Fourth Circuit stated in Jenkins v. Home Insurance Co., 635 F.2d 310, 312 (1980) (per curiam): Unlike Evans, the Company's alleged discriminatory violation occurred in a series of separate but related acts throughout the course of Jenkins' employment. Every two weeks, Jenkins was paid for the prior working period * * * an amount less than was paid her male counterparts for the same work covering the same period. Thus, the Company's alleged discrimination was manifested in a continuing violation which ceased only at the end of Jenkins' employment. Accord, Kim v. Coppin State College, 662 F.2d 1055, 1061 (4th Cir. 1981) ("This court * * * has consistently distinguished Evans when the discriminatory employment practice has continuously affected the complaining employee and is continuing."); Hall v. Ledex, Inc., 669 F.2d 397, 398 (6th Cir. 1982) ("(T)he discrimination was continuing in nature. Hall suffered a denial of equal pay with each check she received.") Satz v. ITT Financial Corp., 619 F.2d 738, 743 (8th Cir. 1980) ("The practice of paying discriminatorily unequal pay occurs not only when an employer set pay levels, but as long as the discriminatory differential continues."); Bartelt v. Berlitz School of Languages of America, Inc., 698 F.2d 1003, 1004 (9th Cir.), cert. denied, 464 U.S. 915 (1983) ("The policy of paying lower wages to female employees on each payday constitutes a 'continuing violation.'"). Cf. Perez v. Laredo Junior College, 706 F.2d 731 (5th Cir. 1983) (applying Title VII principles in suit under Sections 1981 and 1983). Viewing discrimination in compensation as a continuing violation of Title VII, the courts of appeals after Evans have held that pre-Act, intentional discrimination cannot be used to justify the post-Act, intentional discrimination cannot be used to justify the post-Act payment of lower salaries to minority employees than to other similarly situated employees. In Lamphere v. Brown University, 685 F.2d 743, 747 (1st Cir. 1982), for example, the First Circuit ruled that a female faculty member's "allegations that she received a discriminatorily low wage after 1972 (when Title VII became applicable to educational institutions) as a result of pre-1972 discrimination" were actionable, stating: "(A) decision to hire an individual at a discriminatorily low salary can, upon payment of each subsequent pay check, continue to violate the employee's rights." Cf. Berry v. Board of Supervisors of Louisiana State University, 715 F.2d 971, 980 (5th Cir. 1983). These court of appeals decisions reflect the proper construction of Title VII, and correctly distinguish challenges to salary discrimination originating before the Act or outside of the limitations period, but continuing after the effective date of the Act, from cases such as Evans and Hazelwood involving challenges to the post-Act effects of discrete pre-Act decisions such as hiring and termination. The continuing salary discrimination involved here is akin to the continued use of an intentionally discriminatory seniority system, which this Court has held is unlawful under Title VII even if the seniority system was adopted before the Act became effective. See Pullman-Standard v. Swint, 456 U.S. 273, 276-277 (1982); American Tobacco Co. v. Patterson, 456 U.S. 63, 76 (1982). /18/ Just as an intentionally discriminatory pre-Act seniority system affords no justification for current employment practices that have a race-based effect, so too an intentionally discriminatory pre-Act salary system affords no justification for current salary practices that have a race-based effect. To the extent that the court of appeals' justification of post-Act salary disparities as merely the "lingering effect" of pre-Act overt discrimination (Pet. App. 399a) represents a willingness to tolerate such practices, its decision cannot be allowed to stand. /19/ II. THE REGRESSION ANALYSES SHOWED RACIAL DISCRIMINATION, AND RESPONDENTS DID NOT REFUTE THAT SHOWING The court of appeals clearly would have erred if it had held, as private petitioners assert in their Questions Presented, "that statistics may not be treated as probative evidence of discrimination unless the statistical analysis considers every conceivable non-racial variable" (85-93 Pet. i). However, the court below did not impose this onerous "every conceivable variable" burden on petitioners in this case. Rather the appellate court articulated the correct rule, that a plaintiff's statistical analyses in a disparate treatment case must include "variables which ought to be reasonably viewed as determinants of salary" (Pet. App. 391a). We nevertheless agree with private petitioners that the result reached by the court below cannot be sustained because of a variety of errors of fact and law. It should be noted at the outset that reversal of the court below on the first Question Presented automatically requires reversal of the court of appeals' analysis of the statistical proof. One of the bases for the court's rejection of petitioners' regression analysis in this case was that "the analysis contained salary figures which reflect the effect of pre-Act discrimination" (Pet. App. 389a). Because the result we urge on the first Question Presented affirms the correctness of considering the effects of the pre-Act salary discrimination in this case, the holding of the court below cannot stand. At a minimum, the case should be remanded for consideration of whether, when the effect of the pre-Act discrimination on salary is considered, the petitioners' statistics demonstrate disparate treatment in fixing salaries. But other factual and legal errors in the court of appeals' analysis of the statistical proof in this case require entry of judgment for the petitioners on the issue of salary discrimination. Before discussing these errors, however, we outline the general standards for analyzing the legal sufficiency of a plaintiff's regression analysis in a disparate treatment pay disparity case. /20/ A. In Order To Establish A Prima Facie Case Of Salary Discrimination, A Regression Analysis Must Control For Factors That Normally Affect Salary Because this is a disparate treatment case, "(p)roof of discriminatory motive is critical" to a claim of classwide discrimination in fixing salaries. International Brotherhood of Teamsters v. United States, 431 U.S. at 335-336 n.15. But a plaintiff is not required to prove discriminatory motive directly. A plaintiff establishes a prima facie case of intentionally racial discrimination under Title VII if he "eliminates the most common nondiscriminatory reasons" for the challenged act. Texas Department of Community Affairs v. Burdine, 450 U.S. 248, 254 (1981). The reasoning underlying this standard was explained in Furnco Construction Corp. v. Waters, 438 U.S. 567, 577 (1978) (emphasis in original): (W)e are willing to presume (intentional discrimination) largely because we know from our experience that more often than not people do not act in a totally arbitrary manner, without any underlying reasons, especially in a business setting. Thus, when all legitimate reasons for rejecting an application have been eliminated as possible reasons for the employer's actions, it is more likely than not the employer, who we generally assume act only with some reason, based his decision on an impermissible consideration such as race. When plaintiffs present their proof in the form of statistical analyses, these basic principles should not change. Statistical methods should continue to reflect the premises that in a disparate treatment case plaintiff claims to be the victim of intentional discrimination, and that plaintiff bears the burden of proving that intention. Statistics are just a way of proving intention by indirect, inferential means. The touchstone of whether plaintiff has made out a prima facie case (that is, whether plaintiff has made a showing sufficient to permit the case to be presented to the trier of fact) is whether his statistical analysis eliminates the "most common nondiscriminatory reasons" for the disparate treatment, thus leaving racial discrimination as the logical inference. In order to apply these principles to statistical proof in a disparate treatment case, however, it is essential to understand the probative value of the statistics. "As this Court admonished in Teamsters: '(S)tatistics . . . come in infinite variety. . . . (T)heir usefulness depends on all of the surrounding facts and circumstances.' 43 U.S., at 340." Hazelwood School District v. United States, 433 U.S. at 312. In this case the primary statistical proof offered by the United States on behalf of the plaintiffs consisted of multiple regression analyses. The purpose of a multiple regression analysis in this setting is to determine whether the factor of race has sufficient correlation to salary differentials to satisfy plaintiff's burden of proving intentional discrimination. In the language of statistics, salary is referred to as the "dependent variable" in the calculation. To make the calculation "one first specifies the major variables (referred to as 'independent variables') that are believed to influence the dependent variable." Fisher, Multiple Regression in Legal Proceedings, 80 Colum. L. Rev. 702, 705 (1980). "The relationship between the dependent variable (here, salary) and the independent variable of interest (here, race) is then estimated by extracting the effects of the other major variables" (id. at 706). "The results of multiple regressions can be read as showing the effects of each variable on the dependent variable, holding the others constant. Moreover, those results allow one to make statements about the probability that the effect described has merely been observed as a result of change fluctuation" (ibid.). Thus in order to show that race is likely to have influenced salary, a multiple regression analysis must control for other major variables that are thought to influence salary. A multiple regression analysis that is so structured can meet the plaintiff's burden of proving a prima facie case of disparate treatment because the statistical proof eliminates the "most common nondiscriminatory reasons" for the disparate treatment. /21/ Conversely, if the plaintiff's multiple regression analysis does not account statistically for the "most common nondiscriminatory reasons" for differences in salary, the statistics cannot be said to give rise to an inference of racial discrimination and therefore do not make out a prima facie case. This principle was recognized in the context of hiring discrimination in Hazelwood School District v. United States, 433 U.S. at 308. The Court held in that case that in order to show racial discrimination in hiring school teachers the "proper (statistical) comparison was between the racial composition of Hazelwood's teaching staff and the racial composition of the qualified public school teacher population in the relevant labor market" (ibid.; emphasis added). The Court went on to explain that the statistical analysis must account for hiring qualifications in order to have probative value (id. at 308 n.13): In Teamsters, the comparison between the percentage of Negroes on the employer's work force and the percentage in the general areawide population was highly probative, because the job skill there involved -- the ability to drive a truck -- is one that many persons possess or can fairly readily acquire. When special qualifications are required to fill particular jobs, comparisons to the general population (rather than to the smaller group of individuals who possess the necessary qualifications) may have little probative value. It follows from the Hazelwood Court's analysis, and a number of cases have so held, /22/ that in a disparate treatment case if a plaintiff's statistics fails to account for the "most common nondiscriminatory reasons" for the employer's behavior (Burdine, 450 U.S. at 254) (that is, if they do not account for "variables which ought to be reasonably viewed as determinants of salary" (Pet. App. 391a), the defendant may prevail merely by pointing out that plaintiff's proof is not sufficient to give rise to an inference of discrimination. In those circumstances it should not be necessary for the employer to offer his own statistics in rebuttal. /23/ This is not to say that the plaintiff as part of his prima facie case must present a perfect statistical analysis which takes into account every conceivably relevant variable. Dothard v. Rawlinson, 433 U.S. 321, 331 (1977). Indeed, statistical theory presumes that even after the multiple regression analysis accounts for all the major factors thought to influence the dependent variable, other factors remain unaccounted for, and these factors may have a significant influence on the dependent variable. Fisher, supra, 80 Colum. L. Rev. at 705-706. If the plaintiff's statistics include the major non-discriminatory factors thought to influence salary, and if they show a statistically significant effect for race as a dependent variable, the plaintiff has made the required prima facie showing of disparate treatment. If at that point the defendant asserts that some additional factor should have been accounted for in the regression, the defendant must show that inclusion of the factor would have explained the salary disparity (that is, that inclusion of the factor would have eliminated the statistically significant effect for race). Requiring the defendant to offer statistical (or other) proof at this stage is consistent with the requirements of McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802 (1973). To dispel the adverse inference from a prima facie showing the employer must "articulate some legitimate, nondiscriminatory reason for the employee's rejection." Ibid. Accord Furnco, 438 U.S. at 578. As stated above, we believe that the court of appeals in this case articulated the correct standard, requiring the plaintiff's statistics in a disparate treatment case to include "variables which ought to be reasonably viewed as determinants of salary" (Pet. App. 391a). This rule of reason, like all general evidentiary standards, particularly those relating to statistical proof, will necessarily vary in application from case to case. See Hazelwood, 433 U.S. at 312; Teamsters, 431 U.S. at 340. In many situations, common sense will yield obvious answers. In this case, for example, it was reasonable to expect that the length of time the employer worked for the defendant (job tenure) would affect the employee's salary, because some of the pay raises the employer gave were across-the-board percentage increases. Therefore the statistics offered by the United States on behalf of the plaintiffs in this case did include job tenure as a variable in the multiple regression analysis. In general the "variables which ought to be reasonably viewed as determinants of salary" should reflect the factors that go into the employer's own salary decisions. In a disparate treatment case it is assumed that an employer is not required to make hiring, promotion on salary decisions on a basis common to most employers, or on any given basis. Title VII requires only that the basis of decision be nondiscriminatory. Burdine, 450 U.S. at 258-259; St. Marie v. Eastern R.R. Ass'n, 650 F.2d 395, 399 (2d Cir. 1981). Thus, for example, if the defendant chooses to give raises on the basis of job attendance, the plaintiffs' regression analysis should include job attendance as a variable. /24/ In disparate treatment cases where the parties dispute whether a particular factor or variable "ought to be reasonably viewed as (a) determinant( ) of salary" (Pet. App. 391a), if that dispute is not resolved before trial, one party or the other may be seriously disadvantaged by the trial court's ruling on that issue. If plaintiff's regression does not account for the variable and the district court rules that its inclusion was required as part of plaintiff's prima facie case, plaintiff will lose unless he has prepared back-up statistics. Conversely, if the district court rules that the variable was not required to be included, a defendant who had been relying on the inadequacy of plaintiff's case will lose if he had not prepared counter-statistics. But these harsh results are not inevitable. A preliminary judicial determination of the nondiscriminatory factors that are to be subjected to analysis will greatly aid in eliminating evidentiary and burden of proof problems because it will focus the court's and the parties' efforts on the same data from the outset. This approach will preclude wasted efforts (an important consideration because of the exceptional time and expense involved in preparing multiple regression analyses) and post hoc reallocation of burdens. We therefore urge, as a corrollary to the "reasonableness" standard we have outlined, that district courts be encouraged to make formal determinations at the earliest possible stage of proceedings as to the required (for the plaintiff) and permissible (for the defendant) data to be included in multiple regression analysis offered at trial. See Finkelstein, Regression Models in Administrative Proceedings, 86 Harv. L. Rev. 1442 (1973). B. The Court Of Appeals Erred In Analyzing The Statistical Proof Offered In This Case The allocation of proof outlined in the preceding section does not place excessive burdens on plaintiffs in disparate treatment cases. Indeed, application of these standards to the facts of this case compels the conclusion that the multiple regression analyses employed on behalf of petitioners, in conjunction with the other evidence introduced, proved racial discrimination in salaries. /25/ Although the nondiscriminatroy variables on which the courts below focused are generally ones that should be considered in an analysis of salary, the courts erred in their examination of these factors in the particular circumstances of this case. The United States' expert prepared multiple regression analyses concerning salaries for the years 1974, 1975 and 1981. Certain of these regression used four independent variables -- race, education, tenure, and job title. This model reflected the deposition testimony of a Service official who stated that the most important factors in determining salaries were tenure, job title, education degree and job performance (see page 5, supra); the model omitted only the factor of job performance, which was accounted for by other evidence in the case. /26/ The regressions showed that in 1974 the average black employee earned $331 less than a white employee with the same job title, education and tenure (C.A. App. 1601, 402-403), and that in 1975 the disparity was $395 (C.A. App. 1589, 416). /27/ Both of these racial disparities were statistically significant (C.A. App. 402-403, 416). /28/ The Service introduced multiple regression analyses prepared by its expert for the years 1975 and 1981. Using the same model that the petitioners had used, respondents' expert obtained substantially the same result for 1975, a statistically significant racial effect of $384 (C.A. App. 1716 (analysis 5), 1711-1712). Respondents also introduced regressions adding quartile rank as an independent variable. Doing so for 1975 actually increased the amount of the racial effect to $475 -- again a statistically significant result (C.A. App. 1716 (analysis 6), 1713-1714). Thus, the statistical evidence offered by respondents actually bolstered petitioners' case rather than rebutting it. Taken together, the regressions accounted for all of the elements respondents themselves identified during discovery as the most important determinants of salary: job level, tenure, education and quartile rank. Nevertheless, respondents criticized petitioners' statistical case as inadequate, listing additional variables which they hypothesized might, if included, have an effect on the racial disparities demonstrated by the regressions. As we demonstrate below, however, none of the "omitted variables" provides an appropriate basis for determining that petitioners had not met their burden of proof. 1. The only variables specifically discussed by the panel majority are those relating to differences among counties as to salaries (Pet. App. 387a-390a). /29/ It is undisputed in this case that salary levels varied among the various counties and thus the court of apeals is correct that such a major salary-affecting variable should have been accounted for as part of petitioners' case in chief. /30/ The court erred, however, because it failed to recognize that the county-by-county disparities were accounted for by petitioners' evidence, albeit not in the multiple regression analyses. The United States introduced unrebutted evidence that removed county differences as a possible explanation for the racial disparities in salaries shown by the regression analyses. This evidence included: (1) an exhibit establishing that black employees were not located disproportionately in the low-contributing counties (C.A. App. 189, 1611-1623); and (2) intra-county comparisons showing dozens of examples of black agents who were paid less than their white counterparts (C.A. App. 1563-1567 (exhibit based on 1973 data), 189-190 (proposed findings)). In addition, the same pattern -- whites earning more than blacks in the same county who had the same or superior job title, tenure and education level -- was shown in exhibits introduced by respondents (C.A. App. 1626-1680; DX 67). The court of appeals majority completely failed to acknowledge this evidence. Yet, as Judge Phillips pointed out in his dissent, such evidence "completely undercuts (respondents') suggestion * * * that the regression analyses were flawed because of different salary levels from county to county" (Pet. App. 455a n.137; see also Pet. App. 454a & n.136). Accordingly, the evidence shows clearly that the majority erred in using the issue of salary differences among counties as a ground for discounting the results of the regression analyses because overall salary differences among counties did not explain the lower salaries of black agents. 2. The district court stressed job performance as the most important factor determining post-entry salaries (Pet. App. 134a, 141a). According to the district court, "neither side" incorporated job performance in its regression analyses (Pet. App. 134a). This was a primary reason the district court refused to credit the regression data (Pet. App. 141a). The district court clearly erred in this ruling, and the court of appeals improperly sustained that error. As stated above, regression analyses prepared by respondents' own expert included the quartile rank measure of job performance. /31/ Inclusion of the quartile measurement resulted in an increased racial effect for 1975, /32/ Thus supporting the claim of salary discrimination. /33/ Of course, it does not matter that this proof came into the case as part of the respondents' evidence. After a full trial on the merits a reviewing court looks to the evidence as a whole. United States v. Aikens, supra. /34/ 3. The last two items on the Service's list relate to prior experience. /35/ Here, the Service had no existing standard and produced no evidence whatsoever relating to the kinds of experience, if any, that are relevant to extension work or affect salary levels (see C.A. App. 928-930; GX 166, at 52-53). /36/ As a result, there is no basis for measuring for prior experience of Service employees or for including that factor in a regression. /37/ In sum, the court of appeals majority erroneously refused to accept the multiple regression analyses (proffered by both petitioners and respondents) as proof of racial discrimination. The regressions, along with the other evidence discussed above, did establish racial discrimination in salaries, and, accordingly, petitioners should have prevailed on this claim. III. THE SERVICE RETAINS JOINT RESPONSIBILITY FOR THE SELECTION OF COUNTY CHAIRMEN, AND IS THEREFORE LIABLE UNDER TITLE VII FOR DISCRIMINATION IN THOSE SELECTIONS The private petitioners invite this Court to consider whether Title VII permits an employer to delegate its hiring decisions to a third party which invariably acts in a discriminatory manner (85-93 Pet. 49-55). We submit, however, that this question is not properly presented here. The court of appeals concluded that the Service was not "separately responsible" for the selection of county chairment (Pet. App. 412a). It also noted, however (Pet. App. 403a), that "the agreement of the Extension Service and the County Commissioners is required in order to fill the (county chairman) vacancy." The district court explicitly so found: "all appointments are worked out jointly between the Extension Service and the (county) commissioners and no official action can be taken unilaterally by either party" (Pet. App. 77a). No party challenged this finding on appeal, and it is well supported by the record (see, e.g., J.A. 163; C.A. App. 1690). The only question properly presented in this context is thus whether each participant in a joint employment decision must comply with the dictates of Title VII. It requires no extended analysis to demonstrate the fallacy of permitting each participant to avoid liability solely by showing that it shared the decision with another. E.g., Farmer v. ARA Services, Inc., 660 F.2d 1096, 1104 (6th Cir. 1981); Robinson v. Lorillard Corp., 444 F.2d 791, 799 (4th Cir.), cert. dismissed, 404 U.S. 1006 (1971); see Atonio v. Wards Cove Packing Co., 768 F.2d 1120, 1125 (9th Cir. 1985); cf. McDonald v. Sante Fe Trail Transp. Co., 427 U.S. 273, 284-285 (1976). It was accordingly inappropriate for the court of appeals to focus on the statistics regarding the applicants identified by the Service for consideration for promotion, rather than on the statistics with respect to the final choices that were made. /38/ In sum, a remand for reconsideration of the appropriate statistics may be warranted, but this record provides no occasion for consideration of an employer's liability for decisions made without its participation. If the Court nevertheless reaches the issue urged by the private petitioners, we submit, for the reasons explained in our petition, that General Building Contractors Ass'n v. Pennsylvania, 458 U.S. 375 (1982), makes clear that an employer cannot properly be held vicariously liable under Title VII. IV. PRIOR SEGREGATION IN THE 4-H AND EXTENSION HOMEMAKER CLUBS WAS FULLY CURED BY RESPONDENTS' ADOPTION OF A GENUINELY NONDISCRIMINATORY ADMISSIONS POLICY The narrow legal question raised by private petitioners is whether a state actor has failed, as a matter of law, to satisfy its affirmative duty to desegregate a formerly de jure system because it retained a free choice admissions system the result of which is that some all white and all black 4-H and extension homemaker clubs continue to exist (85-93 Pet. i-ii, Question 3). /39/ There is no issue in this Court concerning either the validity of the lower courts' factual conclusion that the 4-H and extension homemaker clubs have been operated in a purely nondiscriminatory manner since at least 1965, or that the respondents have taken no action designed to perpetuate the prior segregation or otherwise to have a segregative effect. /40/ Accordingly, we agree with the court of appeals' conclusion that the "mere existence of all white and all black 4-H and Extension Homemaker Clubs in some racially mixed communities violates neither Title VI nor the equal protection clause of the Fourteenth Amendment" (Pet. App. 424a n.128). /41/ Prior decisions of this Court in the school desegration are demonstrate that a state which has insured that all its practices relating to admissions are truly race-neutral need not assure a particular racial mix /42/ or eliminate personal choice in order to cure its prior segregation. It is well established that the Constitution does not require, either as a matter of substantive right (e.g., Pasadena City Bd. of Educ. v. Spangler, 427 U.S. 424, 433-434 (1976); Spencer v. Kugler, 404 U.S. 1027 (1972)) or as a matter of remedy (Milliken v. Bradley (Milliken I), 418 U.S. 717, 740-741 (1974); Swann v. Bd. of Educ., 402 U.S. 1, 24 (1971)), any particular degree of racial balance in a state's activities. Rather, the Constitution requires that government refrain from any action intentionally designed to exclude or segregate persons on the basis of race. Dayton Bd. of Educ. v. Brinkman (Dayton I), 433 U.S. 406, 413 (1977); Washington v. Davis, 426 U.S. 229, 239 (1976). It is not disputed that the Service engaged in such deliberate segregation after this Court's decision in Brown v. Bd. of Educ. (Brown I), 347 U.S. 483 (1954), and thus "automatically assumes an affirmative duty "to effectuate a transition to a racially nondiscriminatory * * * system.'" Keyes v. School District No. 1, 413 U.S. 189, 200 (1973), quoting Brown v. Bd. of Educ. (Brown II), 349 U.S. 294, 301 (1955). /43/ Such remedial efforts are necessary "to restore the victims of discriminatory conduct to the position they would have occupied in the absence of such conduct" (Milliken I, 418 U.S. at 746) by eliminating the condition that violates the Constitution and the conditions that flow from such a violation. Milliken v. Bradley (Milliken II), 433 U.S. 267, 282 (1977). See Dayton I, 433 U.S. at 417, 420. /44/ Since the condition that offends the Constitution is a race-conscious admissions policy, not the absence of racial balance, the prior constitutional violation with respect to admissions is fully cured by establishing a system of admissions that "exclude(s) no (person) of a racial minority, directly or indirectly, on account of race." Swann, 402 U.S. at 23. /45/ Accordingly, a unitary system of admissions is one "within which no person is to be effectively excluded from any (club) because of race or color." Alexander v. Holmes, 396 U.S. 19, 20 (1969). Accord, Green v. School Board, 391 U.S. 430, 432 (1968), quoting Brown II, 349 U.S. at 300-301 (formerly segregated school systems must "'achieve a system of determining admission to the public schools on a nonracial basis.'"); Brown II, 349 U.S. at 300 ("At stake is the personal interest of the plaintiffs in admission to public schools * * * on a nondiscriminatory basis"). To be sure, in the elementary and secondary school context, achievement of the truly race-neutral admissions scheme necessary to remedy prior segregation requires more than discontinuance of discrimination if the former unconstitutional practices have created a race-based assignment system that remains in place. Simply ending de jure segregation and returning to a neighborhood school system often will not restore the conditions that would have existed absent prior segregation, and thus will not effectively dismantle the dual system. In such school systems, student attendance districts often have been racially gerrymandered, schools may have been purposely constructed in a particular area to achieve artificial racial separation, and the location of segregated schools residential patterns. Swann, 402 U.S. at 25, 28. In these circumstances assigning students to their neighborhood school does not place students -- either those who attended school during the period of de jure segregation or those who will subsequently enter the system -- in the desegregated school environment (or actual school) that they would have experienced absent such prior unconstitutional actions. Accordingly, "(w)hen school authorities present a district court with a 'loaded game board,' affirmative action in the form of remedial altering of attendance zones is proper to achieve truly nondiscriminatory assignments." Id. at 28. To do otherwise would perpetuate the prior de jure segregation by leaving intact conditions that are directly attributable to the unconstitutional actions of state actors. Where the state has no assigned persons during or after the de jure period, however, the process of disestablishing the dual system through achievement of a neutral admissions policy can be more directly accomplished. Where public officials do not assign persons to a particular program, there is no state-controlled attendance pattern, discriminatory or otherwise, to undo or redraw. Thus, unlike elementary and secondary education, affirmative action to assure a genuine and complete termination of all discrimination in activities affecting admissions will not leave in place any discriminatory conditions caused by previous state-imposed segregation. Such a genuinely race-neutral policy will, absent any subsequent conduct that contributes to segregation, fully dismantle the dual admission system because it will restore to the victims of discriminatory conduct (and provide to others) the system mandated by the Constitution, i.e., one in which each person has an equal opportunity to participate in government activities from discrimination and racial separation attributable to state action. This is not to say that a voluntary attendance system is unitary with respect to admissions because the state entity simply announces that it will henceforth conduct admissions without regard to race. Rather, such a formerly segregated government program is unitary only if the state has, in fact, established an admissions "system in which racial discrimination (is) eliminated root and branch." Green, 391 U.S. at 438. Thus, the relevant program must be free from all practices that either indicate that the program is not genuinely open to all free from subtle discrimination or that otherwise create further racial separation. This transition will normally involve diligent oversight and remedial efforts to ensure that the defendant's existing administrative and admissions practices are not tainted by discrimination, and do not send a subtle discriminatory message, that its personnel are sensitized to the need to treat applicants and potential applicants in a wholly nondiscriminatory manner, that its present and future practices do not have the effect of impeding desegregation, and that it is made clear to all that any practices discourage or diminishing racial mixing are truly a thing of the past. Such affirmative action by the state will ensure that a genuinely race-neutral admissions policy has been achieved, and that any subsequent racial imbalance is thus not due to actions attributable to public officials, but to the truly free exercise of choice by private actors unaffected by subtle discriminatory exclusion. /46/ Against the backdrop of recent blatant discrimination, affirmative recruitment activities in the media or otherwise will normally be an important component in this nondiscrimination effort, but the ultimate question in every case is whether the state's policies and practices are conducted in a manner that does not exclude or segregate any person on the basis of race. In this case, as noted, the lower courts found that the respondents had fully cleansed the clubs of any such racially-based activities and that racial imbalance is solely the product of unfettered private action (see note 40, supra). Private petitioners do not maintain that these concurrent findings of fact are erroneous or that the respondents engaged in any action that had a segregative effect. Accordingly, for the reasons stated above, we believe the lower courts' determination that such racial discrimination had been eliminated from the system "root and branch" establishes that respondents had fully performed their duty to redress prior segregation. The Service was not required to attain a particular racial mix in its system, regardless of whether private actors continued to retain the freedom to choose which, if any, club to attend. De facto segregation that is purely the product of private decision making, and thus not attributable to any action by the state, is not proscribed by the Constitution. Dayton I, 433 U.S. at 413. That being so, such racial imbalance need not be eliminated in order to render an admissions system fully desegregated if the government entity has established a genuinely race-neutral admissions pattern. This Court's decision in Spangler makes this clear. There the district court's adoption of a student assignment plan to remedy public school segregation "established a racially neutral system of student assignment in the (school district)." 427 U.S. at 434. After the first year of this plan's implementation, free movement of persons within the district "resulted in some changes in the demographics of (the district's) residential patterns, with resultant shifts in the racial makeup of some of the schools." Id. at 436. "(A)s these shifts were not attributed to any segregative actions on the part of the (school district)" (ibid.), the racial imbalance caused by these voluntary choices of private citizens did not mean that the district had failed to perform its duty to remedy prior segregated student assignments. Consequently, the district court's requirement that the defendants engage in additional remedial admissions practices was impermissible because it enforced a right not secured by the Constitution, i.e., the "'substantive * * * right (to a) particular degree of racial balance or mising.'" Id. at 434, quoting Swann, 402 U.S. at 24, "For having once implemented a racially-neutral attendance pattern in order to remedy the perceived constitutional violations on the part of the defendants, the District Court had fully performed its function of providing the appropriate remedy for previously racially discriminatory attendance patterns." 427 U.S. at 437. Thus, just as the Spangler school board's adherence to the district court's racially neutral student assignment plan fully redressed previously segregated admissions notwithstanding a racial imbalance due to the voluntary decisions of private individuals, so too does the respondents' establishment of a nondiscriminatory system here fully remedy the prior constitutional violation regardless of whether private activities beyond their control result in an absence of racial mixing. To conclude otherwise would be to equate racial balance with effective desegregation; an equation that, as noted, Swann and Spangler expressly rejected. This principle obtains with equal force in situations where private choice, rather than official compulsion, is the means by which attendance is determined. The holding and analysis in Green and its companion cases (Raney v. Bd. of Educ., 391 U.S. 443 (1968); Monroe v. Board of Commissioners, 391 U.S. 450 (1968)) is limited to compulsory public education or other contexts where the government, rather than the participant, traditionally makes the decision concerning where participants in the government program will attend. Thus, Green's rejection of "freedom of choice" student assignment plans simply reflects the familiar rule that a systems which has not yet achieved unitary status may not take steps that foster segregation where there exists a reasonably available alternative that equally or better serves the governmental purpose purportedly being furthered. And, while the "freedom of choice" plan at issue in Green was plainly such a segregation-inducing device, the same is not true where, as here, such private choice is the norm. In public elementary and secondary schools, the decision whether to attend and which school to attend is virtually always made by the school district, not by the student. Accordingly, adoption of a "free choice" or "free transfer" plan which is used or operates to perpetuate the separation of the races is impermissible, at least absent persuasive justification. Utilization of such an unusual and administratively awkward admissions scheme in preference to the traditional, more efficient mandatory assignment system creates the inference that, at a minimum, the defendant school district is not striving in good faith to establish a student attendance system that is truly free from racial considerations. In such circumstances, then, it is the school officials' abdication of their normal responsibility to assign students that causes or at least fosters segregation. Such segregation is thus directly attributable to the state, not to private actors. In short, a non-unitary school district's adoption of a "freedom of choice" plan that predictably, if not designedly, results in greater segregation than the traditional and reasonably available mandatory assignment alternative violates the district's affirmative responsibility to refrain from adopting policies that "serve to perpetuate or reestablish the dual school system" (Columbus Bd. of Educ. v. Penick, 443 U.S. at 460) in preference to less segregative and equally legitimate policies. See Swann, 402 U.S. at 20-21; Columbus, 443 U.S. at 460-462. In our view, it was precisely for this reason that the "free choice" and "free transfer" plans were struck down in Green and Monroe. Noting that normal geographic zoning assignment was a "reasonably available" option that would greatly increase integration, the Green court held that a school board's preference for an attendance scheme such as "freedom of choice" "may indicate a lack of good faith; and at the least it places a heavy burden upon the Board to explain its preference for an apparently less effective method." 391 U.S. at 439. Since the use of a "free choice" plan in these circumstances clearly served no legitimate purpose particularly given the procedural and historical context in which the plan was adopted (id. at 432-433, 440 n.5), the school board in Green was unable to meet this burden. Indeed, "'it is evident that here the Board, by separately busing Negro children across the entire county to the 'Negro' school, and the white children to the 'white' school, is deliberately maintaining a segregated system which would vanish with non-racial geographic zoning.'" 391 U.S. at 442 n.6, quoting Bowman v. County School Board, 382 F.2d 326, 332 (4th Cir. 1967) (concurring opinion). In Monroe, this Court was even more explicit. In that case, the "free transfer" option allowing students to transfer after initial geographic assignment, "permitted the 'considerable number' of white or Negro students in at least two of the zones to return, at the implicit invitation of the Board, to the comfortable security of the old, established discriminatory pattern * * * no attempt has been made to justify the transfer provision as a device designed to meet 'legitimate local problems'; rather, it patently operates as a device to allow resegregation of the races to the extent desegregation would be achieved by geographically drawn zones." 391 U.S. at 459, quoting Goss v. Bd. of Educ., 373 U.S. 683, 689 (1963) (emphasis in original). No such constitutional infirmity attaches to the continuance of free choice in those situations where such choice is a traditional aspect of the government program or activity. In contexts such as public parks, higher education and the 4-H Clubs here, participation or attendance in the state activity is wholly voluntary in all respects. Unlike public schools, free choice as to whether and where to attend is a normal and important feature of the system -- not, as in the public elementary and secondary school context, a newly discovered device that will clearly have a segregative effect. In such cases, a transition to officially compelled assignments would constitute a fundamental alteration of the system. It would increase government regulation of private conduct by denying individuals the liberty to attend, for example, the park that is most convenient or the university best suited to their educational needs. Accordingly, preservation of the participants' traditional "free choice" cannot reasonably be viewed as an act of bad faith or a device that encourages or perpetuates segregation. To the contrary it "serve(s) important and legitimate ends." Dayton Bd. of Educ. v. Brinkman (Dayton II), 443 U.S. 526, 538 (1979). Unlike the public schools in Green and Monroe, therefore, any lingering racial imbalance in the 4-H and extension homemaker clubs here is not attributable to unjustified segregative action by the government respondents, but solely to decisions freely made by private actors. That being so, neither Green nor Monroe suggests that the state has an affirmative duty in such circumstances to eliminate or restrict the discretion of private actors to achieve a greater racial balance. Nor has this Court ever suggested such a rule in any other context. /47/ Indeed, in the same term that Green and Monroe were decided, this Court summarily affirmed, over dissent, a three-judge court's decision that those cases' condemnation of "freedom of choice" plans was inapposite in the higher education context because admissions there are traditionally governed by private choice. Alabama State Teachers' Ass'n v. Alabama Public School & College Authority, 289 F.Supp. 784 (M.D. Ala. 1968), aff'd, 393 U.S. 400 (1969). /48/ Thus, Green did not hold that the continuation of free choice is an inherent defect in a racially imbalanced system, but only that the adoption of such a scheme in the public school context is impermissible if it impedes or frustrates desegregation. Any contrary interpretation of Green "can be supported only by drastic expansion of the constitutional right itself, an expansion without any support in either constitutional principle or precedent." Milliken I, 418 U.S. at 747. Such an understanding of Green would not only read into it a requirement for "mandatory integration" which that opinion expressly rejected (391 U.S. at 437), but also would bring that decision squarely into conflict with this Court's established precedent that no particular degree of racial mixing is constitutionally required. See page 10, supra. Of course, the Service undoubtedly could have sought to increase integration by attempting to close one-race clubs, by denying club members the opportunity to join the club of their choice or by discriminatorily excluding black and white children from clubs where their race predominates through preferential "other-race" goals and quotas. We fail to perceive, however, why such actions are in any way necessary to vindicate the constitutional rights of those persons who unfettered access to government-supported programs is being so restricted. Where a state has unlawfully segregated railroad cars, for example, the state is not constitutionally required to compel blacks to sit in the previously all-white railroad car or to restrict access to the car formerly reserved for blacks. By the same token, so long as the Service ensured equal access to the clubs and did not engage in practices that impeded desegregation, it was not required to restrict persons' normal discretion to join the club of their choice, even if those persons exercised their unfettered choice to join a club in which their race predominates. CONCLUSION For the foregoing reasons, the judgment of the court of appeals should be reversed in part, affirmed in part, and remanded. Respectfully submitted. CHARLES FRIED Solicitor General WM. BRADFORD REYNOLDS Assistant Attorney General CAROLYN B. KUHL Deputy Solicitor General MICHAEL CARVIN Deputy Assistant Attorney General WALTER W. BARNETT LOUISE A. LERNER DAVID B. MARBLESTONE Attorneys JANUARY 1986 /*/ This question does not affect the government's interests; it is accordingly not addressed in this brief. /1/ "Pet. App." refers to the separately bound appendix filed with the petition in No. 85-93. In quoting materials from this appendix, we have corrected typographical errors in the filing; those corrections are indicated by brackets. "Supp. Pet. App." refers to the supplementary appendix bound with the petition in No. 85-428. "J.A." refers to the separately bound appendix filed with this brief. "C.A. App." refers to the 10-volume court of appeals appendix, 10 copies of which have been lodged with this Court. "C.A. Br." refers to the Brief for the United States filed in the court of appeals, 10 copies of which have also been lodged with this Court. /2/ In this brief we will use the term "agents" to refer collectively to employees in all three ranks. /3/ Newly hired agents with advanced degrees, prior relevant experience, or particularly needed skills are paid more than the minimum. Each agent's salary also reflects a contribution by the county in which he is employed, the amount varying from county to county. Pay increases awarded by the county or the state may be in the form of an equal sum to each employee or as a percentage of the salary. Finally, the state and some counties provide for merit pay increased and increases to offset inflation (Pet. App. 109a-115a, 360a-362a). /4/ Before that time, county chairmen were selected jointly by the Service and the board of county commissioners concerned from a list of possible candidates prepared by the Service (C.A. App. 165, 485-488; Pet. App. 101a). /5/ These results were corroborated by other evidence, including an exhibit, based on January 1973 payroll data for 23 counties, that showed 29 black employees earning less than whites in the same county with comparable or lower positions and the same or less tenure and education (C.A. App. 1563-1567). /6/ The court told respondents' counsel (C.A. App. 525): (I)f under the law all those things should have been cranked in there, and if after cranking them in you get a different result, then you win. If they ain't got any business in there or if you've cranked them in and it still doesn't show it, then they win. /7/ Quartile rank is used in determining merit salary increases (see note 32, infra). The government did not use this variable in its regression analyses because the quartile system was itself under attack in this suit as racially discriminatory (see Pet. App. 391a-400a). /8/ The government also introduced statistical and other evidence relating to the selection of county chairmen. The first black chairman was not selected until March 1971, after 151 white chairmen had been selected (J.A. 127; GX 75). Between November 1972, when the first vacancy announcement appeared (see note 4, supra) and October 1981, 72 (93.5%) of the 77 county chairmen selected were white and 5 (6.5%) were black (J.A. 114-124; C.A. App. 919-920). No black was selected for any of the 31 positions filled between November 1972 and July 1, 1976, although 12 (10.3%) of the 116 applicants for those positions were black. Several blacks testified that they had not sought chairmanships because they believed it would be futile to do so (Pet. App. 93a). Of the 5 blacks among the candidates selected for the 46 positions filled between July 1976 and October 1981, 3 were selected for vacancies for which only blacks were in competition, and the sole white applicant with which the other 2 competed was a female; no black has ever been selected in competition with a white male applicant (J.A. 114-124; C.A. App. 1003, 1006; GX 172, at 94-95; see C.A. Br. Table 7, at 10a-11a). /9/ Before analyzing the statistical data, the court explained its approach to the requirements of the Civil Rights Act of 1964 (Pet. App. 121a-122a): Just as it had been found in the area of education that there is no such thing as instant integration, it was soon found in the field of business and industry that there is no such thing as instant (e)quality in employment. Without risking serious disruption of a business by prohi(b)itively costly budgetary alterations and a possible practice of wholesale reverse discrimination it was soon recognized (though not always by the courts) that the adjustments mandated by the law simply could not be made overnight. Thus, the "explanatory evidence" the court viewed as justifying the salary disparities established by the government apparently included the historical fact of discrimination. /10/ The court relied on a list of variables provided by defendants (Pet. App. 133a-134a): (1) Performance of agents measured against the agents' plan of work; (2) The variation in salaries created by across the board state raises with the different percentage of state contributions in each county; (3) The across the board increases in agent salaries by some counties and not in others; (4) The merit raises provided by the state; (5) The merit raises provided for by the counties in which Extension Service personnel have no input; (6) The merit raises provided by the counties with limited or full participation in the merit recommendation by Extension Service personnel; (7) The range in merit salary increases provided by the counties (0 (to) 12% in 1981); (8) Prior and relevant experience; and (9) Variations in salary due to market demands both at time of hire and later for agents with skills in short supply or prior experience. /11/ The district court also held that the plaintiffs failed to establish a prima facie case of racial discrimination in the selection of county chairmen, and that "in any event the defendants have effectively rebutted plaintiffs' case by showing the inaccuracy and insignificance of plaintiffs' proof" (Pet. App. 100a). The court found that only 77 county chairman positions had been filled since the institution of statewide vacancy announcements in 1972, the year Title VII was made applicable to public employers, and that blacks had applied for only 18 of those positions (Pet. App. 78a, 85a-86a). Considering those 18 positions (and thus including positions for which only blacks applied but excluding positions for which only whites applied) the court found the selection rate for blacks acceptable (Pet. App. 79a-80a, 86a), and held that the Service's selection procedures for county chairman have, since 1972, been applied in a nondiscriminatory manner (Pet. App. 100a-101a). The court rejected plaintiffs' claim that blacks had been deterred from applying for chairmanships (Pet. App. 99a), and also rejected all individual claims of discrimination in promotions to county chairman (Pet. App. 227a-317a). /12/ The court's initial analysis (Pet. App. 413a-414a), includes all positions for which black candidates applied, whether or not white candidates also competed, but excludes positions for which only white candidates applied. The court's more extended analysis (id. at 418a-421a) includes "applicant flow data" for all positions filled during the years 1968-1981. In the latter analysis, the court's references to "applicants" or "applications" during the period 1968-1971, when there were no vacancy announcement and application procedures, evidently refers to the list of possible candidates prepared by Service officials, from which they later made recommendations to the counties (see note 4, supra); the court's references to white or black "approvals" refers to the recommendations made by the Service to the counties (see chart preceding Pet. App. 420a). /13/ He also dissented from the majority's rejection of the 4-H and extension homemaker club claims (Pet. App. 425a, 469a-481a). /14/ As Judge Phillips recognized the complaint, filed in November 1971, included claims based on the Constitution and Title VI; the latter has been applicable to the states since 1964. A three-year limitation period applies to those claims. Cox v. Stanton, 529 F.2d 47, 49-50 (4th Cir. 1975). /15/ Judge Phillips disagreed with the panel majority's reading of Evans and Hazelwood as applicable to plaintiffs' salary claims (Pet. App. 462a-467a). Those cases, he explained, do not permit an employer to "continue practices now violative (of Title VII) simply because at one time they were not" (Pet. App. 465a). In his view, the Evans-Hazelwood principle "simply has no logical application" in cases involving "pay and other 'condition of employment' claims, as opposed to hiring and other work-force composition claims" (Pet. App. 466a). /16/ Section 703(h) of Title VII, 42 U.S.C. 2000e-2(h), validates only "bona fide" seniority systems. If an employee can show that the seniority system was adopted with a discriminatory intent, Section 703(h) affords the employer no protection. Pullman-Standard v. Swint, 456 U.S. 273, 276-277 (1982). This Court has emphasized that Section 703(h) "makes no distinction between seniority systems adopted before its effective date and those adopted after its effective date." American Tobacco Co. v. Patterson, 456 U.S. 63, 76 (1982). /17/ There appears to be a conflict among the circuits as to whether an employee can avoid a statute of limitations defense by establishing he was denied promotion pursuant to a continuing practice of discriminatory promotion denials, or whether he must file within the statutory period after he is himself denied promotion even where such a practice is alleged. Compare, e.g., Acha v. Beame, 570 F.2d 57, 65 (2d Cir. 1978) (continuously maintained promotion policy may be subject of complaint until statutory time "after the last occurrence of an instance of that policy," citing cases), and Patterson v. American Tobacco Co., 586 F.2d 300, 304 (4th Cir. 1978) (Evans is inapplicable where a discriminatory promotion system is maintained; the discrimination continues from day to day and a specific violation occurs whenever a promotion is made), with, e.g., Dumas v. Town of Mount Vernon, 612 F.2d 974, 977-978 (5th Cir. 1980) (suit must be filed within statutory time after employee should have perceived discrimination was occurring). However, we are unaware of any case in which a court has permitted an employee to rely on the current effect of a discriminatory promotion policy that was terminated in a time-barred period. Such a complaint, we submit, would clearly be untenable under Evans. Moreover, we have serious doubts about the validity of any theory that would permit an employee who does not sue in a representative capacity to recover when the denial of promotion that affected him is wholly pre-Act or time-barred, even if the discriminatory policy continues so as to affect other employees. /18/ See note 16, supra. This Court addressed the question of continuing pre-Act salary disparities in the context of a suit under the Equal Pay Act, 29 U.S.C. 206; Corning Glass Works v. Brennan, 417 U.S. 188 (1974). That decision is not strictly in point here, however, as a violation of the Equal Pay Act is established simply by showing the payment of lower wages to women than to men performing the same work; the date when the disparity originated, and the reasons underlying the disparity, are largely irrelevant. In contrast, the plaintiff in a discriminatory treatment case under Title VII must establish not only the disparity in wages, but also the employer's intent to discriminate. See International Brotherhood of Teamsters v. United States, 431 U.S. at 335 n.15. Thus, a current disparity in salaries, without more, cannot be the basis for this Title VII claim of discriminatory treatment. Instead, it is necessary to examine the basis for that disparity to determine whether there has been actionable intentional discrimination subject to a timely challenge. Here, although the Service's decision to pay black employees less than whites for the same work was taken before Title VII became applicable to public employers, there is no dispute that the Service's compensation scheme remained intentionally discriminatory. See pages 7, 9, supra; Pet. App. 439a-440a; J.A. 129-130. /19/ Further fact finding will be necessary to establish the recovery due any individual employee. As both courts below emphasized, the effects of the original discriminatory salary practices are part of a complex matrix of pre- and post-Act salary decisions, including merit raises, cost of living increases, and county-to-county variations in salary increases (Pet. App. 109a-116a, 360a-362a). The extent to which these decisions carry forward the effects of the original discriminatory practices, and the extent to which any such decision is actionable in this suit by any employee, must be resolved first by the district court. /20/ Our discussion focuses on the use of one type of statistical evidence: multiple regression analyses. A party may offer other types of statistics, such as cohort studies and multiple pool tests, either instead of or in addition to regression analyses. We do not suggest that regression analyses are the only, or even the best, statistical tools for use in disparate treatment cases. /21/ In addition, of course, the multiple regression analysis must have statistical reliability. "(A) regression not only estimates the effects of the variables involved in the model but also measures the certainty or accuracy of such estimates. In addition, it provides overall measures of how well the model fits the data as a whole." Fisher, supra, 80 Colum. L. Rev. at 716. The statistical significance of the petitioners' regression analyses was not questioned by the courts below and is not an issue in this case. /22/ See, e.g., Valentino v. U.S. Postal Service, 674 F.2d 56, 70-71 (D.C. Cir. 1982); Eastland v. Tennessee Valley Authority, 704 F.2d 613, 624-625 (11th Cir. 1983); Wilkins v. University of Houston, 654 F.2d 388, 401-405 (5th Cir. 1981); cf. St. Marie v. Eastern R.R. Ass'n, 650 F.2d 395, 400 (2d Cir. 1981). /23/ Some cases have indicated that an employer cannot challenge a plaintiff's statistical evidence without making a showing that the factor the employer claims should have been included in the plaintiff's analysis would in fact have eliminated the racial effect. Segar v. Smith, 738 F.2d 1249, 1267-1270 (D.C. Cir. 1984), cert. denied, No. 84-1200 (May 20, 1985); Trout v. Lehman, 702 F.2d 1094, 1102 (D.C. Cir. 1983), rev'd on other grounds, 465 U.S. 1056 (1984); cf. Griffin v. Carlin, 755 F.2d 1516, 1526-1528 (11th Cir. 1985). These cases rely on Teamsters, 431 U.S. at 360-362, for the relevant pattern of proof in class disparate treatment cases. But the pattern of proof in Teamsters does not differ significantly from that stated in Burdine. Indeed Teamsters explicitly acknowledges that the employer may attack the plaintiff's case by showing that plaintiff's proof is "inaccurate or insignificant" as well as by "provid(ing) a nondiscriminatory explanation for the apparently discriminatory result" (431 U.S. at 360-361 & n.46). /24/ In this hypothetical, if the plaintiff was alleging that job attendance records had been kept in a discriminatory fashion, plaintiff's regression analysis need not include this "tainted" variable, so long as plaintiff offered some proof that attendance records were tainted by racial discrimination. See note 34, infra. /25/ We submit that petitioners established a prima facie case; that the Service did not successfully produce probative rebutting evidence; and that the petitioners therefore sustained their burden of persuasion. In accordance with United States Postal Service Board of Governors v. Aikens, 460 U.S. 711 (1983), we now focus on the evidence as a whole. /26/ See pages 32-33, infra. /27/ Contrary to the district court's suggestion (Pet. App. 136a-139a), the inclusion of county chairmen in the data base did not distort the results of these regressions. Job title was included as a variable in the critical regressions, and therefore the salaries of county chairmen were only compared with those of other county chairmen -- and the salary claims related only to the salaries of agents. /28/ The regressions for 1981 showed a smaller disparity which lacked statistical significance (C.A. App. 1578). The lack of a significant racial disparity in 1981 affects, at most, the relief to which plaintiffs are entitled; it does not affect the Service's liability for the post-Act violation of Title VII established by the 1974 and 1975 regressions, whether or not the pattern and practice of salary discrimination demonstrated for those years continued at the time of trial in 1981. /29/ The majority stated (Pet. App. 390a): The across-the-board and percentage pay increases granted by a county, as well as counties which simply pay higher salaries, also are bound to flaw the regression analyses since they were not taken account of. It is also apparent that percentage pay raises from whatever source will magnify any existing disparity. This discussion relates to items (2), (3) and (7) on the defendants' list of omitted variables (see note 10, supra), and may also relate to items (4) and (6). /30/ Although it is not necessary to reach the issue in order to decide this case, it is arguable that petitioners should be excused from including this variable in their regression analyses because until the brink of trial the Service did not identify inter-county salary variance as a major factor influencing salary. The multiple regression analyses introduced at trial included variables the service identified during discovery as the primary factors affecting salary (see page 5, supra). Respondents criticized the omission of other variables from these regressions, variables that they had listed for the first time a few days before trial in response to our motion for summary judgment on the salary issue (Def. Response filed Nov. 27, 1981). The court states that the regressions sufficed to establish a prima facie case of salary discrimination, and indicated that petitioners would prevail unless the Service established that the introduction of other variables into the analyses would eliminate or reduce the disparities shown (see note 6, supra). Trial proceeded on that basis, and the Service failed to meet its burden. Nevertheless, the district court entered judgment for the Service on the salary claim, stating that the plaintiffs "had probably made out a prima facie case," but that the Service had rebutted that case by meeting is "burden of articulating plausible reasons for its actions * * * which the court found convincing" (Pet. App. 190a). This post hoc reallocation of the burdens of proof raises serious questions of procedural fairness in this case. Cf. Keyes v. School District No. 1, 413 U.S. 189, 209 (1973). /31/ The Service's method of evaluating job performance is its quartile system under which, each year, the respective district chairmen place each agent in one of four quartiles (C.A. App. 283-284). Quartile rank is used as the basis for awarding merit raises, with agents in quartile I receiving the highest salary increases (C.A. App. 1382-1387). There are no numerical ratings or rankings within a quartile (Pet. App. 113a-114a), so the only feasible measure of job performance is quartile itself. The validity of the quartile system under Title VII, although not raised in the petitions for certiorari, was a separate issue in the courts below. The court of appeals held that quartile ranks were assigned in a nondiscriminatory manner (Pet. App. 391a-397a). According to the court of appeals majority, this finding meant that there was no discrimination in awarding merit salary increases. In reaching that conclusion, the majority did not mention the Service's 1975 regression which included quartile rank and thereby increased the race effect. Moreover, as Judge Phillips pointed out in his dissent (Pet. App. 452a n.135), even if the quartile system itself was valid, it by no means follows automatically that the process for determining salaries (including merit increases) was nondiscriminatory. /32/ However, inclusion of quartile rank significantly decreased the disparity for 1981 ($310 to $142) (C.A. App. 1713-1714, 1716). As was true of the United States' regressions for 1981, the Service's regressions for that year, both with and without quartile rank, showed racial effects that were not statistically significant (C.A. App. 1691-1692, 1700, 1701). /33/ Quartile rank is pertinent to other items on the Service's list of variables that allegedly should have been included in petitioners' regressions: "merit raises provided by the state," and "merit raises provided by the counties with * * * participation * * * by Extension Service personnel" (Pet. App. 133a). Merit increases -- whether provided by the state or a county -- are, according to respondents, based on quartile rank (Pet. App. 112a-115a (state); id. at 110a-112a (county)). Item (5) on the Services's list is "merit raises provided for by the counties in which Extension Service personnel have no input." There is no factual basis for the district court's finding (Pet. App. 133a) that merit raises have been made in that manner, as the court of appeals seems to have acknowledged (Pet. App. 361a n.3). /34/ Job performance is, of course, normally related to salary and thus should be controlled for in a regression analysis, at least where, as here, the employer considers job performance in making salary determinations. Here, however, the only feasible measure of job performance -- quartile rank -- was itself alleged to be discriminatory. In our view a plaintiff should not be excused from including an otherwise important factor in his multiple regression analysis by merely alleging that the factor is itself discriminatory. A plaintiff must in addition make at least some showing, through statistics or otherwise, that the factor he excludes from his analysis is in fact discriminatory. If a plaintiff is not required to negate the significance of directly relevant, omitted variables, this effectively creates a presumption that the variables that were excluded because they may be "tainted" by discrimination are in fact being applied in a discriminatory manner. Such a burden of proof simply assumes the conclusion that the defendant has discriminated with respect to factors affecting salary. If a plaintiff is to be excused from including numerous variables that are most directly related to salary because those variables are susceptible to discriminatory manipulation, he must also be required to provide supplementary evidence to support the claim that those variables have been so manipulated. In this case petitioners did seek to prove that the quartile system was applied in a discriminatory manner, by offering evidence separate from the multiple regression analyses. Pet. App. 115a-116a. The correctness of the court of appeals' determination that quartile ranks were assigned in a nondiscriminatory manner is not presented for decision by this Court. And there also is no need to determine whether the showing made was sufficient to excuse petitioners from including quartile rank in their regression analyses because, as stated in the text, the respondents included quartile rank in their regressions. /35/ These items are "(p)rior and relevant experience" (item 8)) and "(v)ariations in salary due to market demands both at time of hire and later for agents with skills in short supply or prior experience" (item (9)) (Pet. App. 133a-134a). The latter item is, as the district court recognized (Pet. App. 134a), "difficult if not impossible to measure and quantify." /36/ This case is, in our view, distinguishable from Segar v. Smith, 738 F.2d 1249 (D.C. Cir. 1984), cert. denied, No. 84-1200 (May 20, 1985) on this basis. In Segar, which involved the assignment of new employees to different job levels (GS-7 or GS-9), the federal agency had a written standard as to prior experience qualifying an employee for placement at the higher level (see 738 F.2d at 1275). Under those circumstances, as we pointed out in our petition for certiorari in that case, prior experience is one of the factors normally expected to explain salary differentials and therefore should be included in plaintiffs' regression analyses as part of their prima facie case. /37/ The United States placed into evidence Service personnel files including applications which list the employee's pre-hire work history (GX-63). Analysis of those applications shows that black agents averaged more years of pre-hire experience than did white agents (C.A. Br. Table 6, at 9a). Many black witnesses described their prior experience, and much of it was directly related to extension work (see, e.g., Tr. 34, 627-629, 807, 924). /38/ See note 12, supra. The court did refer briefly to the ultimate success rates of black and white applicants, and determined, "(u)sing the district court's method of calculation," that the relative rates did not raise an inference of disparate treatment (Pet. App. 416a), but the court used incorrect figures in calculating those rates. The court's analysis was also flawed by its inclusion of positions for which there were no white applicants while excluding positions for which there were no black applicants; of the 5 blacks selected, only 2 were in competition with a white applicant, in each case a white female (see note 8, supra). In addition the court failed to take account of the fact that although the Service in the vast majority of cases recommended only a single candidate to the county, it never recommended a black candidate without also recommending a white candidate, if one was available. In each instance where both a white and black candidate were recommended, the white candidate was always selected even though the stated reason for recommending more than one candidate was that they were equally qualified (C.A. App. 914-915). /39/ The precise extent of racial mixing within the 4-H and extension clubs is unclear. Between 1972 and 1980, the total number of integrated (at least one member of a different race) 4-H Clubs in mixed communities grew from 586 to 1442, while the number of single race units in mixed communities remained fairly constant (892 in 1972; 880 in 1980) (C.A. App. 1807, 1140). Of all participants in 4-H Clubs in the state, 15.7 percent belong to these 880 one-race clubs (GX 11). As of 1972, the last year for which statistics are contained in the record, only 22 out of 1866 homemaker clubs were integrated (J.A. 103). /40/ The court of appeals found that "(t)he record is totally devoid of any proof of discrimination with respect to services provided by any 4-H or Extension Homemaker Club" (Pet. App. 424a n.128). The district court likewise found no evidence of any discrimination in services or membership (Pet. App. 165a-185a) and found that any racial imbalance existing among the clubs resulted exclusively from the wholly voluntary and unfettered choice of private individuals (Pet. App. 172a). On this basis, the district court distinguished two other cases which found segregated extension-related clubs had not been dismantled (Pet. App. 175a). Specifically, the district court noted "that the Extension Service has had a policy that all voluntary clubs be organized without regard to race and that each club certify that its membership is open to all persons regardless of race; that it instructs its agents to encourage the formation of new clubs without regard to race; that it publishes its policies in the media; that all of its club work and functions above the local community level are being conducted on a fully integrated basis; that its 4-H camps are fully integrated and have been for over ten years and that no person has been denied membership in any club on account of race" (Pet. App. 181a). The district court found that "(i)n working with the clubs all (Extension) agents work across racial lines" (Pet. App. 169a n.53). Although the United States and private petitioners introduced evidence that the Service had assigned agents to clubs on a racial basis until 1974 (J.A. 138-139; Tr. 1078-1079, 1994-1995, 2019-2020, 2025), they did not present the district court with a proposed finding of fact on this subject. We do not understand petitioners to argue that the court's findings were clearly erroneous, or that the Service has taken any action after the de jure period that has "serve(d) to perpetuate or re-establish the dual * * * system" Columbus Bd. of Educ. v. Penick, 443 U.S. 449, 460 (1979). No evidence relating to the racial composition of clubs formed after the de jure period was introduced at trial, because neither the United States nor private petitioners drew any distinction between clubs formed before or after this period. /41/ The discrimination prohibited by Title VI is coextensive with that proscribed by the Constitution; our discussion will therefore address this question in terms of constitutional requirements. Guardians v. Civil Service Commission, 463 U.S. 583 (1983). The United States Department of Agriculture has promulgated regulations pursuant to Title VI which require that federal aid recipients that have "previously discriminated against persons on the ground of race, color or national origin, * * * must take affirmative action to overcome the effects of prior discrimination." 7 C.F.R. 15.3(b)(6)(i). As discussed in the text infra, where, as here, admissions are traditionally the product of private voluntary choice, an entity covered by Title VI has fulfilled this requirement to take affirmative action to overcome the effects of prior segregation in admissions when it has established a genuinely nondiscriminatory admissions system. Since the regulation simply embodies this statutory and constitutional requirement, the lower courts' conclusion that the Service has created such a nondiscriminatory system in the 4-H and extension homemakers clubs establishes that it is in compliance with these regulations. The Department of Agriculture sent affirmative action "guidelines" in 1972 and 1973 to the Directors of all Extension Services expressing the Department's views concerning steps those Services should take with respect to desegregating 4-H and extension homemaker clubs. These guidelines, on their face, specifically exempt the respondent Extension Service from coverage because of the pendency of this litigation (Pet. App. 173a n.55; C.A. App. 1905, 1914-1958). It is also undisputed that these guidelines are, in any event, nonbinding. The stated purpose of these guidelines is to set forth standards by which Extension Services will "assure that all Extension 4-H and other youth programs * * * are provided to all clientele on a nondiscriminatory basis without regard to race, color, or national origin" (C.A. App. 1941; see also C.A. App. 1923, 1864-1865, 1888). To accomplish this goal of nondiscrimination, the guidelines state that formerly segregated Extension Services should establish "that all reasonable efforts have been made to recruit individuals from all racial groups represented; and reasons for individuals not desiring to become members of such clubs in interracial settings are not based upon race, color or national origin" (C.A. App. 1905 (emphasis in original)). Respondents did not institute one of the recruitment efforts specified by the Department as part of their affirmative effort to ensure nondiscrimination: personal visits by Extension staff members to potential minority participants in the clubs (C.A. App. 1910). As a general enforcement policy, the guidelines' "personal visit" standard is an appropriate and beneficial recruitment measure to assure both the Department and the covered Extension Service that all persons are aware that clubs are truly nondiscriminatory, so that the "reasons for individuals not desiring to become members of such clubs in interracial settings are not based upon race." However, this component of the Department's affirmative action program is not an absolute prerequisite to finding that the clubs are truly nondiscriminatory. The district court's findings of nondiscrimination and that any racial imbalance is solely attributable to the free choice of persons unfettered by subtle racial barriers are not clearly erroneous. For this reason, the United States did not appeal the district court's decision on this issue. /42/ Although private petitioners do not expressly state that any particular racial mix is required, their assertion that the state's efforts have "failed" (85-93 Pet. i-ii) can only be premised on the assumption that some degree of racial mixture is necessary for "success". /43/ It is also undisputed that it was the state's deliberate action that caused intentional segregation in the clubs and thus constituted "state action" governed by the Fourteenth Amendment. See Moose Lodge No. 107 v. Irvis, 407 U.S. 163 (1972). /44/ Both Milliken cases, like other cases cited above, concerned limitations on the remedial discretion of federal courts in segregation cases, and not what a state actor such as respondent must do in order to cure prior segregation. These cases, nevertheless, established the outermost bounds of desegregation remedial efforts because they involved the question of whether a district court has exceeded its authority by fashioning a remedy which goes beyond the "nature and extent of the * * * violation." Milliken I, 418 U.S. at 744. In defining the limitations on judicial remedial authority, however, these cases do not establish that every defendant must take every step a court could possibly order to cure past discrimination. For example, a district court would be authorized to require intensive affirmative recruitment efforts to cure prior discrimination. However, a state defendant that has not performed such recruitment in all particulars will nevertheless have complied with the Constitution if it arrives at the same end: nondiscriminatory admissions. (See note 41, supra.) /45/ Of course, a system does not become unitary in all respects simply by curing its prior discriminatory admissions. Rather, it must remedy all aspects of the system that have been affected by such discrimination. In the public school context this would include faculty, facilities, and the other so-called "Green factors." See Green v. School Board, 391 U.S. 430, 435 (1968). Our discussion is expressed in terms of admissions only because we do not understand the private petitioners to contend that the Service has not fulfilled its affirmative duty with respect to the formation of clubs or any other aspect of the clubs' programs. /46/ The 4-H and extension homemaker clubs are not actually a part of the Extension Service, so the precise form of respondents' affirmative remedial duty is somewhat different, although no less compelling, than a state actor that must cure its own internal practices. With respect to admissions, the respondents are obliged to take affirmative action to ensure nondiscrimination against club members by the private voluntary leaders of those clubs. For example, with respect to the sole instance of discrimination by a voluntary worker in the record, the Service promptly removed the worker and admitted the black child to the 4-H Club (Pet. App. 167a n.51). Of course, the Service must also ensure that its own actions related to these clubs -- formation of clubs, recruitment of leaders, provision of services -- are nondiscriminatory and are not used as devices to perpetuate segregation. /47/ For example, this Court issued a series of per curiam decisions following Brown which invalidated segregation in various state activities without any suggestion that the relevant public officials must do more than henceforth insure a nondiscriminatory admissions policy, i.e., provide free choice. See Muir v. Louisville Park Theatrical Ass'n, 347 U.S. 971 (1954) (parks); Mayor of Baltimore v. Dawson, 350 U.S. 877 (1955) (beaches and bathhouses); Holmes v. City of Atlanta, 350 U.S. 879 (1955) (golf courses); Gayle v. Browder, 352 U.S. 903 (1956) (public transportation). /48/ Of course, a higher education system must take affirmative remedial action not only with respect to admissions, but also, for example, faculty, facilities, and lingering funding disparities before it will become unitary in all respects.