SEYMOUR, Circuit Judge.
Janice Pitre brought this action individually and on behalf of a class of women employees against Western Electric Company, alleging gender-based discrimination in violation of 42 U.S.C. §§ 2000e et seq. (1982) (Title VII). The district court resolved the liability issues in favor of Pitre and the class, and formulated a remedy. Western Electric appeals the determination that it violated Title VII by discriminating against Pitre and the class. Both parties appeal the remedy. We affirm the district court’s liability findings, and reverse and remand for reconsideration of the remedy in light of this opinion.
[1265]*1265I.
Western Electric’s service and installation plants in Merriam and Lenexa, Kansas, employed nonsalaried, salaried, and management personnel during the relevant time period. The salaried employees were divided into five levels, M-10, M-20, M-30, M-40, and M-50. These jobs and the lowest management position, section chief, are at issue here. Typically, employees began at the lower levels and progressively advanced. Section chiefs, however, came not only from the M-50 salaried level but also from nonsalaried and occasionally lower salaried levels. In 1975, Pitre was demoted from her job as a section chief to an M-50 classification.
After exhausting administrative remedies, Pitre brought this action alleging that her demotion violated Title VII. In 1977, she amended her complaint to allege class-wide discrimination against all female employees in the five salary grades and the section chief position in assignment, promotion, downgrading, and layoff. The district court certified the class and determined that the relevant period had begun December 11, 1974, 180 days before Pitre initially filed a complaint with the Equal Employment Opportunity Commission. Pi-tre’s theory of liability was essentially that Western Electric’s all-male management personnel held discriminatory attitudes toward women that resulted in a clustering of women in the lower salary grades. The company responded that the distribution of men and women resulted from historical practice prior to the relevant period and/or nondiscriminatory company policies.
After a full trial, the district court held in favor of the class and Pitre individually. Ree., vol. I, doc. # 126 (District Court Memorandum and Order on Liability) (hereinafter Pitre I). Although the court found that economic conditions necessitated demotions during the mid-1970s, it also found that Western Electric’s demotion and layoff policy discriminated against women. This discrimination continued into the late 1970s when the company began to promote employees.
The parties agreed to a back pay award for Pitre, but they were unable to resolve the class relief or prejudgment interest issues. Consequently, the court imposed a formula for determining interest and a class back pay and front pay award. Rec., vol. I, doc. # 134 (District Court Memorandum and Order on Remedy) (hereinafter Pitre II). The court also enjoined Western Electric from continuing to discriminate. The company appeals the court’s liability findings. Both parties appeal the class remedy.
II.
Western Electric begins by advancing three arguments why the district court’s decision can and should be reversed as a matter of law. First, the company claims that plenary review is appropriate because the district court’s conclusions were not based on credibility determinations. Second, it asserts that because much of the evidence offered by Pitre involved discrimination prior to the period at issue in the case, the court’s findings on the ultimate issue are erroneous as a matter of law regardless of the standard of review. Third, the company argues that because the class did not show significant statistical disparity between male and female demotion and promotion practice, the class claim must fail as a matter of law. We address these contentions separately.
A. The Standard of Review
The analytical framework for trying a Title VII case is now well established. First, the plaintiff must present a prima facie case of discrimination. McDonnell Douglas Corp. v. Green, 411 U.S. 792, 802, 93 S.Ct. 1817, 1824, 36 L.Ed.2d 668 (1973). This burden is satisfied by presenting a scenario that on its face suggests the defendant more likely than not discriminated against the plaintiff. See Texas Dep’t of Community Affairs v. Burdine, 450 U.S. 248, 253, 101 S.Ct. 1089, 1093-94, 67 L.Ed.2d 207 (1981). “The burden of establishing a prima facie case of disparate treatment is not onerous.” Id. If this initial burden is fulfilled, the defendant must set forth non[1266]*1266discriminatory justifications for its allegedly discriminatory practices. Id. at 254-55, 101 S.Ct. at 1094-95. The burden of production placed on the defendant at this stage can be fulfilled by presenting nondiscriminatory reasons for its actions that are specific enough to provide the plaintiff with “a full and fair opportunity to demonstrate pretext.” Id. at 256, 101 S.Ct. at 1095. “General assertions of good faith or of hiring only the best applicants ... are insufficient.” Payne v. Travenol Laboratories, Inc., 673 F.2d 798, 817 (5th Cir.), cert. denied, 459 U.S. 1038, 103 S.Ct. 452, 74 L.Ed.2d 605 (1982). The plaintiff then bears the ultimate burden of proving a statutory violation. This burden can be satisfied either “directly by persuading the court that a discriminatory reason more likely motivated the employer or indirectly by showing that the employer’s proffered explanation is unworthy of credence.” Burdine, 450 U.S. at 256, 101 S.Ct. at 1095. Once the district court makes the ultimate determination as to whether the employer violated the statute, its legal conclusions in regard to the first two stages of evaluating the evidence become irrelevant on appeal. United States Postal Service Bd. of Governors v. Aikens, 460 U.S. 711, 715, 103 S.Ct. 1478, 1481-82, 75 L.Ed.2d 403 (1983).
Neither party questions the district court’s application of these legal standards to the facts of this case. Rather, Western Electric contends that in applying this law to the facts, the district court was not required to make credibility determinations. It maintains that the application of legal standards to undisputed documentary and statistical evidence results in mixed questions of fact and law. Such questions, according to Western Electric, are entitled to significantly less deference than purely factual determinations. Western Electric’s argument, however, is unequivocally undermined by the language of the cases it cites.
The issue in a disparate treatment case such as this one is whether the employer intended to discriminate against the employee. Aikens, 460 U.S. at 715, 103 S.Ct. at 1481-82. “In short, the district court must decide which party’s explanation of the employer’s motivation it believes.” Id. at 716, 103 S.Ct. at 1482. This decision is “both sensitive and difficult,” id., and is therefore treated as a “pure question of fact, subject to Rule 52(a)’s clearly erroneous-standard. It is not a question of law and not a mixed question of law and fact.” Pullman-Standard v. Swint, 456 U.S. 273, 288, 102 S.Ct. 1781, 1789-90, 72 L.Ed.2d 66 (1982). “Where there are two permissible views of the evidence, the factfinder’s choice between them cannot be clearly erroneous.... This is so even when the district court’s findings do not rest on credibility determinations, but are based instead on physical or documentary evidence or inferences from other facts.” Anderson v. City of Bessemer City, 470 U.S. 564, 574, 105 S.Ct. 1504, 1511-12, 84 L.Ed.2d 518 (1985).1 The Court thus concluded that “[i]f the district court’s account of the evidence is plausible in light of the record viewed in its entirety, the court of appeals may not reverse it even though convinced that had it been sitting as the trier of fact, it would have weighed the evidence differently.” Id. at 573-74, 105 S.Ct. at 1511.
B. Evidence of Discrimination Prior to the Period in Question
Western Electric argues that the district court must have relied heavily upon evidence of discrimination that occurred prior to the relevant period, because only from this evidence could discrimination have been proven. It argues that under United Air Lines, Inc. v. Evans, 431 U.S. 553, 97 [1267]*1267S.Ct. 1885, 52 L.Ed.2d 571 (1977), such evidence can serve only as “background” information, and that the district court’s nearly exclusive reliance on such evidence to find a pattern and practice of discrimination within the relevant period requires us to reverse.
Evans has recently been explained in Bazemore v. Friday, 478 U.S. 385, 106 S.Ct. 3000, 3007 n. 6, 92 L.Ed.2d 315 (1986), where the Court noted that there simply had been no evidence of a current Title VII violation in Evans. The Court made clear in Bazemore that evidence of pre-Act discrimination is “quite probative” to establish that current conduct is discriminatory. Id. 106 S.Ct. at 3010 n. 13. Particularly when a company’s decision-making process has not changed, evidence of prior discrimination “might in some circumstances support the inference that such discrimination continued.” Id. at 3010 (quoting Hazelwood School Dist. v. United States, 433 U.S. 299, 309 n. 15, 97 S.Ct. 2736, 2742-43 n. 15, 53 L.Ed.2d 768 (1977)). Because the decision-making process at Western Electric had undergone virtually no change before this case began, the district court properly considered past discrimination as evidence of current discriminatory intent.
C. Statistical Evidence of Class-Wide Discrimination
Western Electric contends that class-wide discrimination cannot, as a matter of law, be established here in the absence of significant statistical evidence. It argues that the district court found all of the statistics presented by Pitre uninformative except for those dealing with the percentage of women demoted from each job classification. The only statistics on which the court relied, the company contends, were legally irrelevant and meaningless because, under widely accepted modes of statistical analysis, the percentages are too insignificant to rule out mere chance as a basis for the slight disparity between the men and women demoted.
The Title VII violation in this case arose from the allegedly disparate treatment of women rather than the disparate impact of a particular policy.
“ ‘Disparate treatment’ ... is the most easily understood type of discrimination. The employer simply treats some people less favorably than others because of their race, color, religion, sex, or national origin. Proof of discriminatory motive is critical, although it can in some situations be inferred from the mere fact of differences in treatment.”
International Bhd. of Teamsters v. United States, 431 U.S. 324, 335 n. 15, 97 S.Ct. 1843, 1854-55 n. 15, 52 L.Ed.2d 396 (1977). In order to prove a pattern or practice of discrimination, a plaintiff must show more than accidental or sporadic incidents of discrimination; she must show that “discrimination was the company’s standard operating procedure — the regular rather than the unusual practice.” Id. at 336 & n. 16, 97 S.Ct. at 1855 & n. 16; accord Bazemore, 106 S.Ct. at 3008.
Although statistics may be useful to show differences in treatment and to establish a pattern and practice, they are clearly not required, especially when the sample size is too small to produce meaningful results. See Teamsters, 431 U.S. at 339-40, 97 S.Ct. at 1856-57; see also Bazemore, 106 S.Ct. at 3009 (“A plaintiff in a Title VII suit need not prove discrimination with scientific certainty; rather, his or her burden is to prove discrimination by a preponderance of the evidence.”); Hazelwood School Dist. v. United States, 433 U.S. 299, 311 n. 17, 97 S.Ct. 2736, 2743 n. 17, 53 L.Ed.2d 768 (1977) (“precise calculations of statistical significance [are not] necessary in employing statistical proof”); Catlett v. Missouri Highway & Transp. Comm’n, 828 F.2d 1260, 1265 (8th Cir.1987). “If the statistical disparity is insufficient alone to establish a prima facie case, the plaintiff may get over his or her initial hurdle by combining statistics with historical, individual, or circumstantial evidence.” Payne, 673 F.2d at 817; see also Catlett, 828 F.2d at 1265; EEOC v. American Nat’l Bank, 652 F.2d 1176, 1189 (4th Cir.1981), cert. denied, 459 U.S. 923, 103 S.Ct. 235, 74 L.Ed.2d 186 (1982); Fitzgerald v. Sirloin Stockade, Inc., 624 F.2d 945, 954-55 (10th Cir.1980). [1268]*1268Although the small sample sizes in this case impaired the effective use of statistics, Pitre presented ample direct and circumstantial evidence of discrimination.
III.
Western Electric also attacks as clearly erroneous the district court’s ultimate findings of statutory violations in its demotion and promotion practices.
A. Statistical Evidence
Western Electric first argues long and hard that the district court’s findings of discrimination are clearly erroneous because “no legally significant statistical disparity” was or could have been shown from the statistics on which the court relied. Although we have already determined that statistics are unnecessary to prove a disparate treatment case, reliance by the court on irrelevant statistical evidence (like reliance on any irrelevant evidence) could render all or part of its findings clearly erroneous. Thus, we must examine the court’s treatment of the statistical evidence.
When a plaintiff presents statistics, their relevance should not be determined through the rigid application of statistical analysis alone. Cf. Payne, 678 F.2d at 817. “[A]ll of the evidence, statistical and non-statistical, tending to establish a prima fa-cie case should [be] assessed on a cumulative basis.” American Nat’l Bank, 652 F.2d at 1189. “[A]s long as the court may fairly conclude, in light of all the evidence, that it is more likely than not that impermissible discrimination exists, the plaintiff is entitled to prevail.” Bazemore, 106 S.Ct. at 3009; Catlett, 828 F.2d at 1265.
The district court in this case specifically asserted that it did not rely exclusively on statistical evidence for any part of its decision. Rather, it noted that it recognized the limited value of the statistics provided by the parties. The court’s opinion makes clear that it did rely heavily on the general discriminatory atmosphere at Western Electric and the anecdotal descriptions of individual incidents of discrimination.
The court did, however, rely to some extent on percentage statistics that showed that “women suffered a disproportionate number of downgradings in each M-level category.”2 Pitre I at 27. Western Electric argues that these statistics are legally irrelevant and therefore the court’s reliance upon them to any extent renders its finding on the company’s demotion policy clearly erroneous. In defining legal relevance, the company contends that statistically significant results for social scientists must exceed “two or three standard deviations.” It therefore urges us to conclude that “legally significant statistics” must be [1269]*1269at least that precise. Answering and Reply Brief of Defendant/Appellant Western Electric Co., Inc. at 6. The company’s reply brief shows the standard deviations for the demotions for each job classification and demonstrates that each is less than “two or three.” Id. at 7a.
At least one other circuit has addressed an argument similar to Western Electric’s. See American Nat’l Bank, 652 F.2d at 1191. In that case, the court pointed out that statistics that are insignificant to the social scientist may well be relevant to a court. Id. at 1192. As the Fourth Circuit recognized, while social scientists search for certainty, the trier of fact in a Title VII case need only find that discrimination is more likely than not. Id. “Another factor suggesting great caution in making fine-tuned use of standard deviation analysis in these cases is that its reliability diminishes in ways probably not susceptible of precise handling by courts as the binomial distribution sample diminishes in size.” Id. at 1193 n. 12. The samples in this case were clearly too small to make standard deviations relevant. The higher employment classifications contained so few women that a management decision that affected all of them (as the demotions did for the M-50 classification) or none of them would not have satisfied a social scientist.
When dealing with such small samples, courts could simply have forbidden the introduction of statistics. Instead, they have chosen to allow their introduction, but required that they be interpreted carefully and in conjunction with all other relevant evidence. See id. This court has applied a common sense approach. In general, a finding of discrimination based generally on a showing of discriminatory acts and attitudes is not clearly erroneous because the district court also considered statistics that a social scientist would find insignificant. See Fitzgerald, 624 F.2d at 954-55. From our review of the record and the district court’s opinion, we conclude that these principles were properly applied. The court’s finding of liability in Western Electric’s demotion policy was not clearly erroneous solely because it relied in part on the statistics in question.
B. Specific Evidence of Discrimination
Western Electric disputes the specific findings of actual discrimination, arguing that its reasons for downgrading and making the subsequent promotions were legitimate and nondiscriminatory. In some situations, the company argues that certain facts relied upon by the court either do not indicate discrimination or that other undisputed facts exist which rule out the possibility of discrimination. In other situations, it argues that the evidence on the record simply does not lead to the conclusion the district court reached.
Before responding to Western Electric’s points, we must articulate the very relevant evidence of discrimination that the company attempts to wish away. The district court articulated a long history of discrimination against the female salaried employees at the Lenexa and Merriam facilities, resulting in the clustering of women in the lower-paying salary grades.3 There is substantial evidence in the record that management decisions relating to hir[1270]*1270ing, assignment, job performance, and promotion were based on subjective criteria. Job openings were not posted, and advancement depended on the recommendation of supervisors, who were overwhelmingly male during the relevant period. In addition, no grievance or appeal mechanism existed to challenge the promotion decisions.
Pitre also offered evidence that Western Electric’s management personnel, who made promotion and demotion decisions throughout the relevant period, held discriminatory attitudes toward women. Even some members of management admitted that discriminatory attitudes towards women existed until at least the late 1960s. Specifically, some thought women could not handle certain technical jobs. Cf. Thorne v. City of El Segundo, 726 F.2d 459, 468 (9th Cir.1983) (“A refusal to hire a woman because of a sex-stereotyped view of her physical abilities is the kind of invidious discrimination that violates Title VII.”), cert. denied, 469 U.S. 979, 105 S.Ct. 380, 83 L.Ed.2d 315 (1984). Pitre presented numerous examples in which these attitudes were acted upon either through beneficial treatment for men in promotion and demotion decisions or through harassment of women on the job. Some women testified to discriminatory application of the company’s overtime, temporary upgrade, and personal leave policies. One woman testified that she was told a promotion would entail night work when in fact the man who received the position was not required to work evenings. Others testified regarding sexual harassment. “[A] plaintiff may establish a violation of Title VII by proving that discrimination based on sex has created a hostile or abusive work environment.” Meritor Sav. Bank, FSB v. Vinson, 477 U.S. 57, 106 S.Ct. 2399, 2405-06, 91 L.Ed.2d 49 (1986). Evidence of harassment, therefore, helps create an inference of discrimination in Western Electric’s promotion and demotion practices. Cf. McDonnell Douglas, 411 U.S. at 804, 93 S.Ct. at 1825 (employer’s treatment of employee during term of employment useful evidence in determining whether nondiscriminatory justification is pretextual).
Finally, Pitre points out that over fifteen years after the passage of Title VII, women were still receiving far less than half of the promotions to upper level jobs which did not require special skill or education.4 Although the adoption of Title VII did not require the immediate equalization of male/female disparity in upper level jobs, it is implicit in the district court’s findings that in the decade before the slowdown requiring downgradings, Western Electric [1271]*1271had the opportunity to promote more than two women to the position of section chief or to integrate the sexes more adequately in upper level positions.
Notwithstanding the above evidence, Western Electric contests several of the district court’s specific factual findings. The company argues that because it followed its written Corporate Instruction consistently in demoting employees, the district court could not conclude that it used the Instructions as a pretext for discrimination. The company contends that in choosing the time-in-grade rather than time-in-employment standard, as was its option under the Instruction, it could not have been discriminating because choosing either seniority system would have resulted in a slightly greater number of female demotions than male.
Western Electric apparently believes that any policy facially based on seniority cannot, if consistently applied, violate Title VII because of section 703(h), 42 U.S.C. 2000e-2(h). This exception, however, protects bona fide seniority systems adopted with no intent to discriminate pursuant to collective bargaining agreements. See American Tobacco Co. v. Patterson, 456 U.S. 63, 102 S.Ct. 1534, 71 L.Ed.2d 748 (1982) (“Seniority provisions are of ‘overriding importance’ in collective bargaining and they ‘are universally included in these contracts.’ ”). As the district court noted, the salaried employees in this case had no say in the Corporate Instruction on which the demotion policy was based. In addition, the seniority plan was applied far less formally and consistently than a collectively bargained plan would have been. Cf. Payne, 673 F.2d at 827 (informal and erratic application falls far short of bona finde seniority system). In such a situation, the employer’s claim that seniority considerations justified apparently discriminatory demotion practices may well be considered pretextual. Id. The employer is entitled to reorganize its company as necessary to suit the current needs of its business; however, it “is not free to exercise that right in a manner that prevents a qualified member of a protected class from retaining a position that he or she was previously performing satisfactorily.” Hawkins v. Anheuser-Busch, Inc., 697 F.2d 810, 814 (8th Cir.1983).
Moreover, there were deviations from use of time-in-grade in the demotion process. Western Electric attempts to explain these away as instances where other factors exempted the particular employee from demotion. The Corporate Instruction, it points out, required consideration of numerous factors other than seniority including “ ‘ability, performance, potentiality, ... and the needs of the business.’ ” Brief of Defendant/Appellant at 26. The district court found that these other highly subjective factors also served as pretexts for discrimination. The potential for discrimination in a situation where numerous subjective factors are allegedly considered was explained in Grano v. Department of Dev. of City of Columbus, 699 F.2d 836 (6th Cir.1983):
“Courts have frequently noted that subjective evaluation processes intended to recognize merit provide ready mechanisms for discrimination. Moreover, the legitimacy of the articulated reason for the employment decision is subject to particularly close scrutiny where the evaluation is subjective and the evaluators themselves are not members of the protected minority.”
Id. at 837 (citations omitted); see also Segar v. Smith, 738 F.2d 1249, 1290 (D.C.Cir.1984), cert. denied, 471 U.S. 1115, 105 S.Ct. 2357, 86 L.Ed.2d 258 (1985); Burrus v. United Tel. Co., 683 F.2d 339, 342 (10th Cir.), cert. denied, 459 U.S. 1071, 103 S.Ct. 491, 74 L.Ed.2d 633 (1982); Payne, 673 F.2d at 826; Stewart v. General Motors Corp., 542 F.2d 445, 450 (7th Cir.1976), cert. denied, 433 U.S. 919, 97 S.Ct. 2995, 53 L.Ed.2d 1105 (1977). “While some subjectivity is inevitable in filling [upper level] jobs ..., the total lack of objective standards ... could only reinforce the prejudices, unconscious or not, which Congress in Title VII sought to eradicate as a basis for employment.” Stewart, 542 F.2d at 450. Thus, in this circuit, "the rejection of an otherwise qualified individual on the basis of subjective considerations entitles [1272]*1272the plaintiff ‘to the benefit of an inference of discrimination.’ ” Burrus, 683 F.2d at 342 (quoting Bauer v. Bailar, 647 F.2d 1037, 1046 (10th Cir.1981)).
We recognize that when management considers individuals for upper level positions, subjective factors must play some role. Their use does not per se constitute discrimination. In this case, however, those making the subjective decisions were men, at least some of whom held discriminatory attitudes and who had participated in past discrimination. Cf. Payne, 673 F.2d at 826. The record also reflects specific discriminatory incidents. When we consider that without the subjective explanations, Western Electric cannot account for the inconsistent application of a seniority policy that was itself found to be a pretext for discrimination, we conclude that the district court properly considered the company’s reliance on subjective evaluation methods to be additional evidence of discrimination.
Western Electric similarly argues that its policy of upgrading those employees previously downgraded before promoting new employees was consistently applied and actually benefited women since more of them had been downgraded. When we focus on upper level positions where the discrimination predominantly occurred, however, the company’s argument loses force. Pitre, who was demoted from section chief, was never repromoted to her old job. Western Electric explains this failure by noting that its policy of reupgrading was not applied to management personnel. Since the M-50 level, the highest nonmanagement position, was predominantly male when the downgrading began, adopting a reupgrade policy for that level clearly disadvantaged women. The district court could have reasonably presumed that a decision to use the policy at the M-50 level but not at the section chief level was arbitrary. That the only deviations from the reupgrade policy at the M-50 level were for two men adds credence to the district court’s conclusions. Considering all the evidence in the record, we conclude that the district court’s finding of discrimination in the company’s reupgrade policy was not clearly erroneous.
Western Electric raises two related points concerning the relative treatment of men and women. It argues that because similarly situated male employees were treated as Pitre was, her treatment could not have been discriminatory or retaliatory,5 and because more women than men were promoted during the relevant period, its promotion policy could not have been discriminatory. However, an employer is not immunized from liability simply because some males received detriments before or contemporaneously with a Title VII plaintiff or because other protected classes received benefits instead of a Title VII plaintiff. See United States v. Lee Way Motor Freight, Inc., 625 F.2d 918, 950 (10th Cir.1979). “It is clear beyond cavil that the obligation imposed by Title VII is to provide an equal opportunity for [all] ..., without regard to whether members of the applicant’s [sex] are already proportionately represented in the work force.” Furnco Constr. Co. v. Waters, 438 U.S. 567, 579, 98 S.Ct. 2943, 2951, 57 L.Ed.2d 957 (1978). Similarly, that a member of a protected class was hired or promoted in place of a Title VII plaintiff has repeatedly been held insufficient to insulate the employer from liability. Cockrham v. South Central Bell Telephone Co., 695 F.2d 143, 145 (5th Cir.1983); Peters v. Lieuallen, 693 F.2d 966, 970 (9th Cir.1982); DeLesstine v. Fort Wayne State Hosp., 682 F.2d 130, 132-33 (7th Cir.), cert. denied, 459 U.S. [1273]*12731017, 103 S.Ct. 378, 74 L.Ed.2d 511 (1982); Lee Way Motor Freight, 625 F.2d at 950. A statistical showing that a higher percentage of female applicants are hired than male has also been held unavailing. Catlett, 828 F.2d at 1266. “[Irrespective of the form taken by the discriminatory practice, an employer’s treatment of other members of the plaintiffs’ group can be ‘of little comfort to the victims of ... discrimination.’ ” Connecticut v. Teal, 457 U.S. 440, 455, 102 S.Ct. 2525, 2535, 73 L.Ed.2d 130 (1982) (quoting Teamsters, 431 U.S. at 342, 97 S.Ct. at 1858). Thus, a showing that some similarly situated males were demoted at the same time as Pitre, or a showing that more women than men were actually promoted, does not preclude the district court from finding a statutory violation.
Although favorable treatment of members of a protected class tends to undermine a plaintiff’s Title VII claim, the district court must weigh this evidence against the relevant evidence tending to show discrimination. Evidence that most of the promotions during the relevant period were given to women does not establish that other promotion decisions were not made on a discriminatory basis, particularly when most of the promotions were in the lower M levels that were comprised predominantly of women. The Sweeney court addressed an argument similar to that advanced here as follows:
“This information [of favorable treatment of women] obviously bears on the question of discrimination, but it does not render the district court’s contention clearly erroneous_ Nor does the promotion of one woman of admittedly outstanding credentials to [a top position] and the denial of such a promotion to two men preclude a finding that another woman was denied the same position because of sex. One familiar aspect of sex discrimination is the practice, whether conscious or unconscious, of subjecting women to higher standards of evaluation than are applied to their male counterparts. The district court could have concluded ... that [plaintiff] would have been promoted had she been evaluated against the standard that was applied generally to men.”
Sweeney v. Board of Trustees of Keene State College, 604 F.2d 106, 114 (1st Cir.), cert. denied, 444 U.S. 1045, 100 S.Ct. 733, 62 L.Ed.2d 731 (1979). The district court pointed out that several of Pitre’s performance ratings were modified after their submission by her supevisor without the supervisor’s knowledge, and at least one clearly less qualified person was promoted ahead of Pitre.6 Considering this and other evidence, the evidence of facially nondiscriminatory promotion practices simply cannot render the court’s ultimate decision clearly erroneous.
Similarly, the district court’s finding that Western Electric demoted Pitre because she was a woman was not clearly erroneous. She presented evidence that her performance was in the middle of the pack of section chiefs. The district court, therefore, could have relied on Western Electric’s failure to present evidence of a systematic comparison of all the section chiefs to infer that Pitre was demoted before less qualified males who were never considered for demotion. Even if we assume that Western Electric has proven that the males demoted were similar to Pitre, the company did not show that all those not demoted were superior to her. We therefore cannot hold the district court's finding clearly erroneous.7
[1274]*1274IV.
REMEDIES
Both parties disagree with the class remedies imposed by the district court, which included some back pay, and some front pay, and an injunction.8 Title VII remedies are equitable awards intended to advance the dual statutory goals of eliminating the effects of past discrimination and preventing future discrimination. Albemarle Paper Co. v. Moody, 422 U.S. 405, 418, 95 S.Ct. 2362, 2372, 45 L.Ed.2d 280 (1975). The purpose of the statute is also “to make persons whole for injuries suffered on account of unlawful employment discrimination.” Id. Typically, although not invariably, this make-whole relief takes the form of back pay awards. Id. at 415-16, 95 S.Ct. at 2370-71. The district court may, however, employ whatever remedies are necessary to achieve the public policy interests at issue. See, e.g., Franks v. Bowman Transp. Co., 424 U.S. 747, 762-68, 96 S.Ct. 1251, 1262-63, 47 L.Ed.2d 444 (1976) (awarding retroactive seniority relief); Albemarle, 422 U.S. at 415-18, 95 S.Ct. at 2370-72 (back pay); Fitzgerald, 624 F.2d at 957 (front pay); United States v. City of Chicago, 549 F.2d 415, 436-37 (7th Cir.1977) (mandatory hiring quotas). In general, courts should strive to award “the most complete relief possible.” Albemarle, 422 U.S. at 421, 95 S.Ct. at 2373; see Franks, 424 U.S. at 764 n. 21, 96 S.Ct. at 1264-65 n. 21 (reducing the existing earnings gap caused by past discrimination part of affording complete relief).
In some situations, the court can evaluate each victim of discrimination individually and adopt a remedy that makes the injured person as well off as she would have been absent discrimination. E.g., Teamsters, 431 U.S. at 371-72, 97 S.Ct. at 1872-73. Where the discrimination has continued for an extended time and the challenged employment practices throughout the relevant period turned on subjective criteria, however, any attempt to predict where any individual victim would be in the employment process absent discrimination is mere guess-work. In such cases, some courts have concluded that a group remedy can best fulfill the purposes of the statute. See Catlett, 828 F.2d at 1267; Segar, 738 F.2d at 1289-91; Love v. The Pullman Co., 569 F.2d 1074, 1077 (10th Cir.1978); Stewart, 542 F.2d at 452; Pettway v. American Cast Iron Pipe Co., 494 F.2d 211, 261-62 (5th Cir.1974).
In this case, both parties agreed that a group remedy was appropriate and that the back pay award should be divided pro rata among the class. Pitre II, at 3. The district court concurred:
“The Supreme Court established the usual procedure for determining class-wide recovery in Teamsters v. United States, 431 U.S. 324, [97 S.Ct. 1843, 52 L.Ed.2d 396] (1976):
Initially, the court will have to make a substantial number of individual determinations in deciding which of the minority employees were actual victims of the company’s discriminatory practices. After the victims have been identified, the court must, as nearly as possible, ‘recreate the conditions and relationships that would have been had there been no’ unlawful discrimination.
Id., [431 U.S.] at 371-72 [97 S.Ct. at 1872-73] (quoting Franks v. Bowman Transportation Co., 424 U.S. 747, 769 [96 S.Ct. 1251, 1266, 47 L.Ed.2d 444] (1976)). In the case at hand, however, it is virtually impossible to determine which class members would have been promoted had there been no discrimination. There was no posting of promotion opportunities, women were given misinformation regarding job duties and discour[1275]*1275aged from seeking promotions, and promotions were made on solely subjective criteria. Hence, the Teamsters approach is not practical, and a formula approach to back pay is appropriate.”
Id. at 2-3.
A. The District Court’s Award
The district court adopted a method of calculating back pay that was similar to the one used by Western Electric’s statistical expert. Rather than merely look at the static distribution of men and women at any one time, the court focused on the dynamics of the work force during the relevant period. It calculated the percentage of women in a given classification and determined for each relevant year whether they received their proportionate share of the promotions given to the next highest level. If the percentage of women promoted was lower than the percentage in the next lower classification, back pay was awarded from that year through the relevant period. The back pay amount was calculated by comparing the salaries of the men promoted to those of the women who were not. This amount was distributed among the women who remained at the lower classification. Back pay for discrimination in demotions was calculated using a similar percentage comparison of women in each job classification.
The district court recognized that this method of calculating back pay ignored the effect of past discrimination on the current position of women in the company. The court attempted to address the problem by ordering front pay. Pitre II, at 6-8. Since fifty percent of Western Electric’s work force was female, the court concluded that front pay was appropriate for women in any level for which the next highest level consisted of more than fifty percent men. This award would continue, the court held, until women were adequately represented throughout the company. The district court declared that it preferred the front pay remedy rather than a more complete back pay award because the former would provide added incentive for Western Electric to move women into high level positions as rapidly as possible. Id. at 8.
Finally, the court issued an injunction barring further discrimination or retaliation against the class and ordering Western Electric to notify all prospective female promotees about the availability of future job opportunities. Id.
B. Western Electric’s Objection to the Remedy
Western Electric’s basic objection to the remedy is that the women receiving the greatest compensation were not actual victims of discrimination and therefore the remedy violates the principles announced in Firefighters Local Union No. 1784 v. Stotts, 467 U.S. 561, 579-82 n. 15, 104 S.Ct. 2576, 2589-90 n. 15, 81 L.Ed.2d 483 (1984).9 Stotts, an affirmative action case, contains broad language indicating that section 706(g) of Title VII, 42 U.S.C. § 2000e-5(g), permits the grant of a remedy only to actual victims of illegal discrimination. This dicta in Stotts has since been limited in Local 28 of the Sheet Metal Workers’ Int’l Ass’n v. EEOC, 478 U.S. 421, 106 S.Ct. 3019, 92 L.Ed.2d 344 (1986). A majority of the Court agreed that, in the context of affirmative action, section 706(g) does not deprive courts of all power to fashion a Title VII remedy which benefits nonvictims of the discrimination. See id. 106 S.Ct. at 3049-50 (Brennan, J., joined by Marshall, J., Blackmun, J., and Stevens, J.); id. at 3054 (Powell, J., concurring); id. at 3062 (White, J., dissenting). Justice Brennan noted in particular that many circuit courts subsequent to Stotts have declined to read the decision broadly. Id. at 3050 n. 47 (citing cases); cf. Catlett, 828 F.2d at 1267-68.
We need not decide how far a district court may go in ordering a group remedy to be shared by the class, instead of holding hearings to determine the individual victims of the employer’s discrimination, [1276]*1276because Western Electric agreed to the procedure utilized below. Cf. Thompson v. Sawyer, 678 F.2d at 286-87. In these circumstances, we will instead apply the logic of the group remedy cases. In so doing, we will examine the particular remedy imposed to ensure that it does not frustrate the purposes of the statute. Albemarle, 422 U.S. at 415-16, 421-22, 95 S.Ct. at 2373-74.
We recognize that a Title VII remedy will often fail to completely “make whole” each victim of actual discrimination. Cf Franks, 424 U.S. at 776 (truly complete relief would involve downgrading those who would not have their positions but for the discrimination). In the case of a group remedy, moreover, precise proof of the discrimination giving rise to each award of back pay or other remedy will not exist. See, e.g., Segar, 738 F.2d at 1290-91. We must scrutinize the group remedy carefully to ensure that the purposes of the statute are fulfilled, id. at 1293, and extend deference to the district court’s “keener appreciation of those facts and circumstances peculiar to particular cases,” Albemarle, 422 U.S. at 421-22, 95 S.Ct. at 2373-74. In other words, although the district court’s discretion is bounded by the dictates of considered judgment, id. at 415-16, 95 S.Ct. at 2370-71, we do not review Title VII remedy judgments de novo, see, e.g., Salone v. United States, 645 F.2d 875, 878 (10th Cir.), cert. denied, 454 U.S. 894, 102 S.Ct. 390, 70 L.Ed.2d 208 (1981); EEOC v. Safeway Stores, Inc., 634 F.2d 1273, 1282 (10th Cir.1980), cert. denied, 451 U.S. 986, 101 S.Ct. 2321, 68 L.Ed.2d 844 (1981); Lee Way Motor Freight, 625 F.2d at 932.
Western Electric argues that the court’s formula for determining back pay and front pay should be reversed because the record does not reflect discrimination sufficient to justify the award. However, the district court found that the company had violated Title VII. The court was then required by statute to devise a remedy that would best serve the public policy interests at issue. Western Electric’s dissection of parts of the proposed remedy and its argument that little or no proof of discrimination supported a particular back pay award is of no moment, given its concession below that a pro rata group remedy should be fashioned. The company had its chance at the liability stage to argue that discrimination did not occur. We have found sufficient evidence of discrimination to support the district court’s liability determination. Thus, the only justification for overturning the remedy on appeal is that it fails to serve the statute’s purposes.
Even if we interpret Western Electric’s argument as an attack on the remedy’s effectiveness instead of as just another argument against liability, we find it no more persuasive. Western Electric does not offer any alternative method of formulating a group remedy that would serve the statutory purposes as well or better than that ordered by the district court. Courts endorsing a group remedy have said that Title VII “should not be read as requiring effective denial of backpay to the large numbers of [employees] whom [the employer’s] discrimination has injured in order to account for the risk that a small number of undeserving individuals might” be overcompensated. Segar, 738 F.2d at 1291. In group remedy cases, “[i]f effective relief for the victims of discrimination necessarily entails the risk that a few nonvictims might also benefit from the relief, then the employer, as a proven discriminator, must bear that risk.” Id. (citing Stewart, 542 F.2d at 452-53). Since we have long preferred to leave the issue of fashioning the remedy to the district court, Fitzgerald, 624 F.2d at 956, the lack of a better alternative weighs strongly in favor of its method.
In short, we conclude that the trial court’s decision to grant back pay, front pay, and an injunction was a proper exercise of its equitable discretion.
C. The Class’ Objections to the Remedy
The class’ arguments against the district court’s remedy focus on its patent unfairness in awarding the majority of the back and front pay awards to a relatively small portion of the class. The women receiving [1277]*1277the greatest rewards, according to plaintiff, were typically those who suffered the least because of the discrimination and who offered the least assistance in proving the class case. The crux of plaintiffs argument is expressed in the following passage: “The Court conceded the impossibility of identifying with any precision those women who would have been elevated to the M-40 level, yet it chose to reimburse only those who had, to the exclusion — and detriment— of the other candidates.” Brief of Plaintiff/Appellee at 62. Plaintiff also argues that the district court’s attempt to use front pay as a hedge against the inadequate back pay award fails because it too would go only to women already in the higher job levels. In addition, the court restricted the availability of front pay to those women who remain in Western Electric’s employ.
Specifically, plaintiff describes the net effect of the district court’s remedial order as follows:
“1. Nineteen class members who were laid off after December 11, 1974, received no back pay whatever for those terminations.
“2. Back pay for demoted class members began with the month of demotion and ended when they were returned to former levels. No compensation was ordered for the pay differentials existing from 1973 on by reason of the company’s past discriminatory promotion policy. The remedy gives its imprimatur to discriminatory effects as they existed during the actionable period.
“3. Back pay for promotions denied to women from 1976 was ordered only for some class members who were M-20’s in 1976 and those class members who were M-40’s before 1981. No other women received any compensation for discriminatory promotion practices. One M-20 was totally excluded from this recovery (J. Deutman).
“4. In lieu of back pay from 1973 to 1980 to all class members suffering pay differentials stemming from historic promotion denials and the effects of the downgrading, the Court ordered front pay to some of those M-40’s who received back pay in paragraph 3 above. No other class members received front pay.
“5. Thirteen females received virtually all the money awarded. Of this number, only three were willing, despite their fear of retaliation, to participate in the preparation of the class aspects of the case and to testify at trial. The vast majority of the witnesses who assisted and testified received either nothing or only a paltry sum. The Remedy Order rewarded those whose fear of retaliation kept them from coming forward.
“The inequity in the Court’s remedy is apparent: 24 class members received no compensation at all (including 16 who were laid off, five who were not downgraded at all, and three who resigned during the actionable period); 24 women shared $40,328.00; and the remaining 13 females received an aggregate sum of $468,224.00."
Brief of Plaintiff/Appellee at 58-59.
In reviewing the district court’s remedy, we adopt the three guideposts that other circuits have followed in similar cases involving group remedies. These are as follows:
“(1) unrealistic exactitude is not required; (2) ambiguities in what an employee or group of employees would have earned but for discrimination should be resolved against the discriminating employer; (3) the district court, far closer to the facts of the case than we can ever be, must be granted wide discretion in resolving ambiguities.”
Stewart, 542 F.2d at 452; Pettway, 494 F.2d at 260-61. Where a group has suffered discrimination through subjective management decisions, “a precise calculation of the pecuniary effects of discrimination” is impossible. Stewart, 542 F.2d at 452. Thus, “[t]he method of calculating a class-wide backpay award must not be rigid.” Pettway, 494 F.2d at 260. “The key is to avoid both granting a windfall to the class at the employer’s expense and the unfair exclusion of claimants by defining the class or the determinants of the [1278]*1278amount too narrowly.” Id., 494 F.2d at 262 n. 152.
Although a district court’s remedy should be disturbed only when it frustrates the statutory purposes, we must reverse in this case. See Albemarle, 422 U.S. at 421-22, 95 S.Ct. at 2373-74. The district court properly attempted to balance the employer’s and the employees’ interests, but it erred by failing sufficiently to consider the effects of past discrimination.
First, we believe the court unintentionally disregarded earnings lost due to the lingering effects of past discrimination. “[B]ecause promotions at [Western Electric] are cumulative, the effects of discrimination in promotions are also cumulative. Denial of promotion to one grade affects the [employee’s] eligibility for later promotions to higher grades.” Segar, 738 F.2d at 1291; accord Thompson v. Sawyer, 678 F.2d 257, 290 (D.C.Cir.1982).
“When discrimination continues over time, as it did at GPO, the harms it causes are compounded. An employee denied a raise in one year will fall further behind if raises in subsequent years are a function of prior salaries. Likewise, an employee denied a promotion in a given year may be frozen out of additional promotional opportunities, unless the missed rung on the promotion ladder is somehow replaced. When the effects of illegal discrimination are compounded, basing the plaintiffs’ compensation only on events within the two-year accrual period seriously shortchanges their recovery for the harm actually suffered during the accrual period itself.”
Id. at 290. Although the district court specifically recognized this proposition, Pitre II, at 7, it concluded:
“In computing back pay, the court could have considered events occuring beyond the limitations period. We could have gone back to 1965 (when Title VII became effective) and determined the number of promotions discriminatorily denied women between 1965 and 1974 (the starting date for back pay calculations). See Thompson v. Boyle, 499 F.Supp. 1147 (D.D.C.1980), modified sub nom. Thompson v. Sawyer, 678 F.2d 257 (D.C.Cir.1982) (citing Salone v. United States, 645 F.2d 875 (10th Cir.), cert. denied, 454 U.S. 894 [102 S.Ct. 390, 70 L.Ed.2d 208] (1981)). This method would have resulted in the number of women that should have been at each level by 1974 in the absence of discrimination. Because a greater number of women would have been at the higher levels, a greater back pay recovery for the class would result for the years 1974 through 1981....
“In recognition of the inequity discussed in Thompson v. Sawyer, and in order to fully compensate the class, we have decided that front pay is appropriate in this case.... We have chosen to allow front pay in this manner rather than calculating back pay by taking into account events occurring back to 1965. This method better serves the purposes of Title VII. Front pay in this case will have the effect of moving women into the higher levels rather than merely compensating them for discrimination in the past. Back pay for only a few of those years of discrimination, without restoring women to their rightful places, would not fully compensate the plaintiff class.”
Id. at 6-8 (emphasis added.)
It is apparent from the quoted statements that the district court believed front pay would in part replace back pay. However, front pay is intended to compensate victims of discrimination for the continuing future effects of discrimination until the victim can be made whole. Thus, in Lee Way Motor Freight, 625 F.2d at 932, we stated:
“ ‘[B]ack pay in Title VII cases may be supplemented, in the district court’s discretion, by an award equal to the estimated present value of lost earnings that are reasonably likely to occur between the date of judgment and the time when the employee can assume his new position.’ ”
Id. (quoting White v. Carolina Paperboard Corp., 564 F.2d 1073, 1091 (4th Cir.[1279]*12791977) and adding emphasis). Front pay is a substitute for immediate promotion which cannot occur because there are no positions currently available. Cf. Bruno v. Western Elec. Co., 829 F.2d 957, 966 (10th Cir.1987) (front pay is merely a substitute for reinstatement when reinstatement is not feasible). Front pay is thus a supplement to back pay, not a substitute for it, particularly where as here the front pay is granted to only a limited number of employees.10
Second, the district court appears to have insufficiently considered the effects of past discrimination in determining how to distribute the amount of back pay it deemed appropriate. Thus the court limited the distribution of back pay to the women in the job classification who were considered in the determination of the amount. See supra at 1275. Even if the court had valid reasons for considering only the women in the M-40 grade as eligible for promotion to the M-50 grade,11 for example, in a group remedy context it should not have limited the pro-rata distribution of the back pay to those women already in the M-40 classification because, but for the discrimination, other women would have progressed to the M-40 level.
It is not our province to attempt to devise a remedy at this level. Rather, we remand to allow the district court to resolve these problems. On remand, the court is free to consider any method of determining the appropriate amount of back and front pay to which the class is entitled. Some courts have ordered parties to negotiate a suitable approach to determining back pay. See Pettway, 494 F.2d at 262 n. 152 (citing cases). Other courts have expressed a preference for group comparisons. See Love, 569 F.2d at 1077; Stewart, 542 F.2d at 453; Pettway, 494 F.2d at 262-63. For example, the salary of representative groups of women could be compared with groups of men of similar education and experience or length of service. The differences could be distributed pro rata among the members of the particular groups. By not using the women’s current position in the company, the effects of past discrimination may be more fully incorporated into the formula. In principle, the district court’s method of calculating the promotions that should have gone to women with those that actually did go to women could again be used. The court must, however, either attempt to account for the effect of past discrimination on the position of women at the beginning of the relevant period or explain why such consideration is unnecessary. Cf. Albemarle, 422 U.S. at 417, 95 S.Ct. at 2371.12
Whatever method is ultimately chosen to determine the amount of back pay due to the class, the district court might well consider the plaintiff class’ expressed desire as to how the award should be distributed among the class members. In Title VII suits, the class victims are often in the best position to evaluate the actual damage caused to particular individuals by discrimination in employment practices. We stress, however, that the ultimate fashion[1280]*1280ing of relief is left to the district court’s good judgment, guided by the factors set out in Albemarle, 422 U.S. at 416-21, 95 S.Ct. at 2370-73.13
V.
The decision of the district court is affirmed as to liability and reversed as to remedy. The case is remanded for the formulation of a new remedy consistent with this opinion.