Franceslon Forehand (individually and on behalf of putative class members), Naomi Berry, Sadie Bouie, Myles Brown, Ethel Germany, Vivian Johnson, Virginia Jackson, Hollis McClendon, and Jeanette Wynn (collectively “appellants”) appeal the post-trial entry of judgment against them by the district court. Their appeal raises the following issues: (1) whether the district court erred in decertifying the class ten years after the case was filed; (2) whether appellants Berry, Bouie, Jackson, Germany, Brown and McClendon failed to exhaust the EEOC administrative process because they received their right-to-sue letters after the commencement of their Title VII action, prior to the expiration of the statutory 180-day period, and upon counsel’s request; and (3) whether the special master erred in rejecting appellants’ statistical evidence of a pattern and practice of promotion discrimination in favor of appellees’ statistical evidence.1 We affirm in part, vacate in part, and remand for further proceedings.
I. FACTS AND PROCEDURAL HISTORY
Forehand applied for and was denied promotion to Ward Supervisor by the Florida State Hospital at Chattahoochee (“the Hospital”) in 1981. After responding to a vacancy announcement for the Ward Supervisor position, Forehand was interviewed by a screening committee. Of the nine applicants for the position, Forehand received the second highest rating by the screening committee. A white woman was given a somewhat higher rating and was awarded the position. Forehand, who is black, alleges that the Hospital discriminated against her in this promotion decision.
Three days after she was denied the promotion, Forehand filed an administrative complaint with the Equal Employment Opportunity Commission (“EEOC”) in which she alleged discrimination in the hospital’s promotion decision (hereinafter “the individual claim”). Seven months later, Forehand amended her EEOC complaint to state that the discrimination against her was part of a pattern and practice of racially discriminatory recruitment, hiring, job assignments, pro*1565motions, demotions, terminations, lay-offs, reprimands, seniority and affirmative action programs at the Hospital (hereinafter “the pattern and practice claim”). The EEOC investigated Forehand’s individual claim and concluded that the Hospital’s decision to promote the candidate selected was based on non-discriminatory criteria including the fact that she: (1) received the highest interview score;2 (2) possessed more supervisory experience; and (3) had received slightly better performance evaluations.
On April 6, 1983, Forehand filed the present action alleging Title VII employment discrimination. In her final complaint, she was joined by twelve additional plaintiffs: the eight other appellants, Berry, Bouie, Brown, Germany, Johnson, Jackson, McClendon, and Wynn, and four other plaintiffs.3 On July 26, 1985, the district court certified a plaintiff class which included:
All past, present and future black employees of Florida State Hospital, Chattahoochee, Florida, who, after 24 March 1972 have been adversely affected on account of their race by the defendants’ use of their subjective decision-making processes regarding promotions, demotions, reassignments, job performance evaluations, and disciplinary actions.
With the exception of Forehand, none of the appellants received notice of a right to sue from the EEOC within ninety days preceding the commencement of this action.4 Neither Wynn nor Johnson ever possessed a right-to-sue letter during the pendency of this action. Appellants Berry, Bouie, Jackson, Germany, Brown and McClendon filed charges with the EEOC slightly before or after this action commenced.5 They were issued right-to-sue letters by the EEOC after the present action had commenced and before the expiration of the statutory 180-day conciliation period.6 Berry’s charge was dismissed by the EEOC for failure to cooperate with its investigation. The other letters were apparently issued in response to counsel’s request for expedited treatment.
In October 1986, a fifty-five-day bench trial was held before a special master. Almost four years after trial, the special master recommended that the district court enter judgment in favor of appellees. As to the nine appellants, the special master ruled as follows. First, he held that Wynn, Johnson, and Berry7 failed to exhaust their administrative remedies because they failed to receive a right-to-sue letter from the EEOC. Thus, he held that these three appellants failed to satisfy the conditions precedent to suit in federal court. He concluded that these appellants could, however, bypass the conditions precedent under the “single-filing rule”8 whereby they could rely on Fore-*1566hand’s properly exhausted claims (both the individual claim and the pattern and practice claim). The special master held that because Forehand’s claims failed on the merits, so did the claims of these three appellants. Second, the special master held that the other six appellants, Forehand, Bouie, Jackson, Germany, Brown, and McClendon, had satisfied the conditions precedent to suit by receiving right to sue letters, but that their individual claims failed on the merits, i.e., they failed to prove discrimination at trial. Third, the special master held that appellants failed to prove a pattern and practice of discrimination.
Over two years later, the district court entered judgment in favor of appellees. See Forehand, 839 F.Supp. at 812. First, the district court held that the class had been improvidently certified and, therefore, decer-tified the class. Id. at 811-12. Second, it held that only Forehand had satisfied all conditions precedent to suit, but that the other appellants could potentially rely on Forehand’s exhausted claims (both Forehand’s individual claim and her pattern and practice claim) under the single-filing rule. Id. at 818. It then found, however, that the non-filing plaintiffs’ claims were so different from Forehand’s promotion claim that they should not be permitted to rely on Forehand’s charge of discrimination.9 Id. at 820-21. Next, it held that appellants had failed to prove a pattern and practice of discrimination. Id. at 819. These rulings left only Forehand’s individual claim intact. The district court permitted Forehand to supplement the record before deciding her individual claim of disparate treatment. Id. at 821. The court later adopted the special master’s conclusion that Forehand had failed to show that she was discriminatorily denied promotion to the Ward Supervisor position.
II. DISCUSSION
A. Class Decertification
First, appellants challenge the district court’s decision to decertify the class ten years after the case was filed. See Forehand, 839 F.Supp. at 811-12. The district court held that it had improperly certified an “across-the-board” class which included putative plaintiffs who allegedly suffered a variety of discriminatory acts — a class which included all black employees who suffered discrimination at the Hospital. The district court determined that, based on the evidence presented at trial, its previous class definition failed to comply with Rule 23 of the Federal Rules of Civil Procedure. It determined, for example, that its prior Rule 23 numerosity determination “was based on a rough estimate of total black employees and not upon a careful estimate of black employees having specific grievances similar to those of the plaintiffs.” Id. at 811.
Appellants argue that the district court erred in decertifying the class. They contend that the district court was required to revisit class certification to conform the class definition to the evidence presented at trial. They urge that the district court’s decertifi-cation decision was especially inappropriate because the case was filed ten years earlier and had already gone to trial.
A district court may alter or amend its certification order anytime before its decision on the merits. Fed.R.Civ.P. 23(c)(1).
Questions concerning class certification are left to the sound discretion of the district court. Freeman v. Motor Convoy, Inc., 700 F.2d 1339, 1347 (11th Cir.1983). Even after a certification order is entered, the judge remains free to modify it in the light of subsequent developments in the litigation. [General Telephone Co. of Southwest v. Falcon, 457 U.S. 147, 160, 102 S.Ct. 2364, 2372 (1982)].
Cox v. American Cast Iron Pipe Co., 784 F.2d 1546, 1553 (11th Cir.), cert. denied, 479 U.S. 883, 107 S.Ct. 274, 93 L.Ed.2d 250 (1986). Although a class decertification order entered ten years after commencement of the action is unusual and perhaps disfavored, we find no abuse of discretion in this case. The district court’s conclusion that the class failed to conform to the requirements set forth in Falcon, 457 U.S. at 161, 102 S.Ct. *1567at 2372, is correct and supported by the record.10 See Forehand, 839 F.Supp. at 811-12. Accordingly, we affirm the district court’s order decertifying the class for the reasons stated therein. See Id.
B. Exhaustion and Equitable Modification
Berry, Bouie, Jackson, Germany, Brown, and McClendon challenge the district court’s determination that they failed to exhaust their administrative remedies before commencing suit in federal court.11 Before instituting a Title VII action in federal district court, a private plaintiff must file an EEOC complaint against the discriminating party and receive statutory notice from the EEOC of his or her right to sue the respondent named in the charge. Pinkard v. Pullman-Standard, 678 F.2d 1211, 1215 (5th Cir. Unit B 1982), cert. denied, 459 U.S. 1105, 103 S.Ct. 729, 74 L.Ed.2d 954 (1983) (citing, inter alia, 42 U.S.C.A. § 2000e-5(f)(l)).12 Further, if, after the expiration of 180 days, the charge has not been dismissed and no other action has been taken by the EEOC, the EEOC is required to notify the claimant and that claimant may bring suit in district court within 90 days thereafter. 42 U.S.C.A. § 2000e-5(f)(l).13 The court in Pinkard held that the receipt of a right-to-sue letter is not a jurisdictional prerequisite to suit in district court, but rather, is a condition precedent subject to equitable modification. 678 F.2d at 1216. See also Fouche v. Jekyll Island—State Park Authority, 713 F.2d 1518, 1525 *1568(11th Cir.1983) (“[A]ll Title VII procedural requirements to suit are henceforth to be viewed as conditions precedent to suit rather than jurisdictional requirements.”). Because Berry, Bouie, Jackson, Germany, Brown and McClendon filed suit before receiving their right-to-sue letters, they must depend upon equitable modification.
Berry, Bouie, Jackson, Germany, Brown and McClendon argue that the district court erred in holding that they failed to exhaust their administrative remedies by failing to obtain right-to-sue letters from the EEOC prior to suit. Specifically, they argue that this Court in Pinkard established a per se rule in holding that
the receipt of a right-to-sue letter subsequent to the commencement of a Title VII action, but while the action remains pending, satisfies the precondition that a plaintiff obtain statutory notice of the right to sue before filing a civil action under Title VII.
Pinkard, 678 F.2d at 1219.
The district court rejected appellants’ per se interpretation of Pinkard, holding that the exhaustion rule is subject to equitable modification only in appropriate circumstances. Forehand, 839 F.Supp. at 816-17. “That equitable modification was appropriate in Pinkard, however, does not mean that equitable modification is appropriate here. In Pinkard, there was nothing to suggest that plaintiffs in any way frustrated the EEOC’s effort to investigate or conciliate the charges.” Id. at 817. The district court held that the dismissal of Berry’s14 charge for failure to cooperate in the EEOC’s investigation meant that she could not partake in Pinkard’s equitable modification of the exhaustion rule. Further, the court held that Bouie, Jackson, Germany, Brown, and McClendon were not entitled to Pinkard’s equitable modification because they filed suit and requested their notices of right to sue long before the 180-day statutory period had elapsed. Id. at 817. The court found significant that the EEOC only issued the letters after appellants’ attorney sent the following request to the EEOC district director:
Our reason for filing the charge of discrimination with the EEOC was to perfect Title VII jurisdiction in this lawsuit. Because we feel that a conciliation of both class and individual aspects of this charge will not be possible without judicial intervention, WE REQUEST A RIGHT-TO-SUE LETTER.
Id. at 816.
Likewise, appellees urge that plaintiffs have failed to carry their burden of demonstrating an equitable reason why the exhaustion requirement should be relaxed. Like the district court, they argue that appellants deliberately frustrated the EEOC investigation and conciliation process by requesting their right-to-sue letters prior to the expiration of the 180-day period. Because the purpose of Title VII’s exhaustion requirement is to allow the EEOC an opportunity to perform its conciliation function, appellees argue that these appellants are not entitled to equitable modification because they asked the EEOC to forego this function.
In Pinkard, plaintiffs filed their Title VII suit four days after they filed their complaint with the EEOC. After the statutory 180 days had expired (during the pendency of the action), plaintiffs received their right-to-sue letters.
The reason given by the EEOC for issuing the letters was that this lawsuit was pending. Though the reason given was not one of the published conditions for which the EEOC issues right-to-sue letters, see 29 C.F.R. §§ 1601.19, 1601.28, the letters were nevertheless effective.... Whether the EEOC was correct in dismissing the charges for the reason given is immaterial.
Pinkard, 678 F.2d at 1219. The court found it significant that, because more than 180 days had elapsed since they filed charges before the EEOC, plaintiffs had a right under the statute to obtain their letters simply upon request. Id. at 1219. The court concluded that “the receipt of a right-to-sue letter is a condition precedent which, on proper occasion, may be equitably modified.” Id. at 1218-19. It based its decision, at least *1569in part, on the policy underlying EEOC regulations, i.e., permitting the EEOC an opportunity to investigate charges and attempt conciliation between the parties. Id. at 1218.15
The district court in this case distinguished Pinkard on the grounds that Bouie, Jackson, Germany, Brown and McClendon filed suit and requested their right-to-sue letters prior to the expiration of the 180-day statutory period and before the EEOC had an opportunity to investigate and conciliate. Forehand, 839 F.Supp. at 817.
After the district court’s decision in this case, this Circuit decided two more cases that must guide our analysis. In Sims v. Trus Joist MacMillan, 22 F.3d 1059, 1062 (11th Cir.1994), we noted that 42 U.S.C.A. § 2000e — 5(f)(1) does not prohibit the EEOC from issuing a right-to-sue letter prior to the expiration of 180 days. We so noted in the course of holding that the 180-day requirement is a condition precedent to suit subject to equitable modification. In Sims, the EEOC had issued plaintiffs right-to-sue letter less than two weeks after the charge had been filed because “the charge would not be processed within 180 days.” Id. at 1060. Plaintiff had requested a right-to-sue letter at about the same time the EEOC received his charge. The district court dismissed the case, finding that the right-to-sue notice was “requested before the EEOC had an opportunity to investigate the case'.” Id. Reversing the district court’s dismissal, we reasoned:
“The 180-day period was intended to afford victims of employment discrimination a private cause of action where the EEOC does not act, or does not act in a timely fashion. The EEOC’s regulation simply recognizes that the caseload will sometimes be so heavy that it can be determined early on that no action can be taken within 180 days and the issuing of an early right-to-sue letter is a reasonable implementation of the Act.... It is wp to the EEOC to decide how to efficiently administer the Act, and unless its decisions contravene congressional intent we must afford them deference.”
Id. at 1062 (quoting Rolark v. Univ. of Chicago Hospitals, 688 F.Supp. 401, 404 (N.D.Ill.1988)) (emphasis added). We pointed out that an individual’s right to sue is not conditioned upon the EEOC’s performance of its administrative duties. Id. at 1063.
As in this case, the Sims plaintiff requested a right-to-sue letter before the 180-day period had expired and before the EEOC had an opportunity to investigate, and the right-to-sue letter was issued before the expiration of the 180-day period. However, Sims is distinguishable from the instant case because the Sims plaintiff did not file suit until after he received the right-to-sue letter, and because the EEOC stated that it would be unable to process his charge within the 180-day period.
In Cross v. State of Alabama, 49 F.3d 1490 (11th Cir.1995), plaintiffs filed their Title VII suit more than two weeks before they filed their employment discrimination charge with the EEOC. Id. at 1504. Within approximately 60 days, they received their notices of right to sue. Id. As is the case here, the notices stated that “‘[w]ith the issuance of this notice of right-to-sue, the Commission is terminating any further processing of this charge.’” Id.16 The defendants in Cross argued that plaintiffs’ filing of suit prior to filing their charges with the EEOC prejudiced defendants’ right to attempt resolution through the conciliation procedures mandated by Congress. Id. The court held, however, that “the Notices of Right To Sue gave appellees the immediate right to file suit in federal district court.” Id.
It is clear from the foregoing cases that receipt of a right-to-sue letter is not a jurisdictional prerequisite to suit, but rather, is a statutory precondition which is subject to *1570equitable modification.17 Because the issue is one requiring consideration of the equities, we readily conclude that there is no per se rule that receipt of a right-to-sue letter during pendency of the suit always satisfies the exhaustion requirement. We reject plaintiffs’ proposed per se rule.
Thus, we agree with the district court’s general proposition that if a claimant attempts to frustrate investigation or conciliation by the EEOC, equitable modification of the exhaustion rule may be inappropriate. The district court properly applied this rule to Berry; Berry’s failure to cooperate with the EEOC disentitled her to equitable modification. We affirm the district court’s holding that Berry failed to independently exhaust her claims.18
With respect to Bouie, Jackson, Germany, Brown and McClendon, the district court also denied equitable modification, finding that they too had failed to cooperate with the EEOC. The district court’s opinion indicates its reliance upon three factors in making this finding: (1) that these appellants “filed suit before they filed charges with the EEOC,” 839 F.Supp. at 817; (2) that they “requested their notices of right to sue long before the 180-day statutory period had elapsed and long before the EEOC had any opportunity to perform the function assigned to it,” id.; and (3) that “in his letter to the EEOC, plaintiffs’ counsel made clear that he had no interest in permitting the EEOC to first attempt a settlement of his clients’ grievances ... and he perceived the filing of the 1983 EEOC charges as little more than a necessary technicality.” Id.
Because of intervening case law — both Sims and Cross were decided after the district court ruled — we are concerned about the district court’s finding of lack of cooperation on the part of Bouie, Jackson, Germany, Brown and McClendon. With respect to the first factor relied upon by the district court, it would be inconsistent with Cross to base a finding of lack of cooperation simply on the fact that suit was filed before filing the EEOC charge. See also Pinkard, 678 F.2d at 1215 (suit was filed four days after filing the charge). With respect to the second factor relied upon by the district court, it would be inconsistent with Sims, and probably also with Cross, to base a finding of lack of cooperation simply on the fact that a party requested a right-to-sue letter before the 180-day period expired and before the EEOC had an opportunity to perform its function.
Indeed, the EEOC regulations expressly contemplate that a plaintiff may make an early request for a right-to-sue letter, and that the EEOC may issue the letter upon determining that it is probable that it will be unable to complete its administrative processing within the 180-day period. 29 C.F.R. § 1601.28(a)(2). In this case, the EEOC did not certify that it would be unable to complete processing within 180 days.19 However, “Title VII ‘does not condition an individual’s right to sue upon the EEOC’s performance of its administrative duties.’ ” *1571Sims, 22 F.3d at 1063 (quoting Jefferson v. Peerless Pumps Hydrodynamic, Div. of FMC Corp., 456 F.2d 1359, 1361 (9th Cir.1972)); see also Pinkard, 678 F.2d at 1219 (“Whether the EEOC was correct in dismissing the charges for the reason given is immaterial.”). Accordingly, the fact that the EEOC may not have complied with its own regulations is of no moment in determining whether appellants are entitled to equitable modification. A plaintiff should be free to make an early request for a right-to-sue letter upon the assumption that the EEOC will perform as contemplated in the regulations by issuing the letter only if it is probable that it will be unable to complete the administrative processing within 180 days. Both Sims and Pinkard indicate that any deficiency in the EEOC’s performance of its duties should not adversely affect a plaintiffs right to sue.
Thus, the intervening ease law has thrown new light on two of the three factors relied upon by the district court. We are uncertain as to whether the district court would have reached the same finding in light of the intervening case law. Accordingly, we vacate for further proceedings not inconsistent with this opinion. Of course, on remand the burden of proof with respect to equitable modification remains on plaintiffs.
In a footnote, the district court offered what could be construed as an alternative holding:
The special master considered the merits of some, but not all, of the single-filing plaintiffs’ claims. Of those claims that he considered, he found not one to be meritorious. While the single-filing plaintiffs did not include, in the record, the transcripts of their trial testimony, this court cannot— based on the record before it — say that the special master’s findings were clearly erroneous.
Forehand, 839 F.Supp. at 821 n. 6. Bouie, Jackson, Germany, Brown, and McClendon have not provided transcripts of their trial testimony to this Court. Accordingly, based on the limited record, we are unable to determine whether the special master’s findings were clearly erroneous. We recognize that we have the authority to affirm the special master’s conclusion that appellants Bouie, Jackson, Germany, Brown and McClendon failed to prove their case on the merits. See Fed.R.App.P. 10(b)(2) (“If the appellant intends to urge on appeal that a finding or conclusion is unsupported by the evidence or is contrary to the evidence, the appellant shall include in the record a transcript of all evidence relevant to such finding or conclusion.”); United States v. Dallas County Comm’n, 739 F.2d 1529, 1540 (11th Cir.1984) (“In the absence of a complete record, we cannot adequately review the challenged findings ... and must affirm the district court on this issue.”). Although we possess the authority to affirm as to these appellants on this alternative ground because appellants failed to provide us with the appropriate transcripts, we decline to do so for several reasons. First, appellees do not invite us to do so and neither party briefed the issue. Second, the district court’s alternative holding is couched in language that leaves us unsure as to whether it would have relied on this ground alone. Moreover, as to Forehand’s individual claim of intentional discrimination, the district court postponed review of the special master’s recommendation until the parties had an opportunity to supplement the record. 839 F.Supp. at 821. We prefer to permit the district court to exercise its discretion in deciding whether these five appellants should be given a similar opportunity.20
In sum, we affirm with respect to Berry, and vacate and remand with respect to Bouie, Jackson, Germany, Brown and McClendon.
C. Statistical Evidence
Finally, appellants argue that the special master erred in adopting the Hospital’s statistical evidence in lieu of their statistical evidence, which, they contend, proves a pattern and practice of racial discrimination. *1572The statistical evidence in this case was used to determine whether there existed a disparity between the percentage of black employees “eligible” for a promotion and the percentage of blacks promoted. At trial, both parties tendered experts who testified as to the merits of their statistical methodology. The Hospital’s statistical evidence showed no pattern and practice of discrimination, i.e., no disparity between the percentage of blacks eligible for promotion and the percentage promoted.21 Appellants’ statistical evidence showed some evidence of such a pattern and practice, i.e., the percentage of blacks eligible for promotion exceeded the percentage actually promoted.22
The difference between the parties’ statistical methodology derives from the unique factual setting of promotions at the Hospital. Over the ten-year period at issue in this ease, seventy-five percent of “promotions” came from a process that was noncompetitive. Noncompetitive promotions did not require an application and selection process, but rather, were often merely an administrative reclassification of jobs which involved some increase in pay. By contrast, twenty-five percent of promotions resulted from a process that was competitive. These promotions required an application and selection process.23 The percentage of blacks who applied for competitive promotions was greater than the percentage of blacks in the Hospital’s workforce in general.
Ideally, to determine whether there existed a disparity between the percentage of eligible blacks and those promoted, the parties would compare the percentage of blacks seeking competitive promotions to the percentage competitively promoted and, independently, they would compare the percentage of blacks in the workforce to the percentage granted noncompetitive promotions. Appellees did, in fact, undertake such an analysis and found approximate parity in the percentages.
Neither party placed primary reliance on this methodology.24 Instead, both appellants and appellees attempted to compare the percentage of all black employees promoted (including both competitive and noncompetitive promotions) to the percentage in the pool from which promotions were granted. The parties agreed on the data to be used in counting the number and percentage of promotions: they simply calculated the percentage of blacks who were granted both competitive and noncompetitive promotions during the ten-year time span litigated in this case.25 This calculation yielded the percentage of blacks in the “at issue” jobs.
The parties disagreed, however, on the data to be used as the “benchmark,” i.e., the pool from which promotions were granted. The appellants chose as their pool the percentage of blacks who applied for competitive promotions. They call this an “applicant flow” benchmark. To this applicant flow benchmark, appellants compared the percentage of blacks in the “at issue” jobs, i.e., the percentage granted both competitive and noncompetitive promotions. Thus, appellants compared the percentage of blacks who applied for competitive promotions to the *1573percentage of blacks who were granted both competitive and noncompetitive promotions.
This methodology drastically skewed appellants’ results. Because the percentage of blacks was greater in the applicant pool than in the Hospital workforce, the applicant flow benchmark would tend to overstate any un-derrepresentation of blacks in the “at issue” jobs.26 Comparing the percentage of black applicants for competitive positions to the percentage of blacks granted both competitive and noncompetitive promotions created the appearance of disparity — the percentage of blacks in the pool from which promotions were granted would appear to be greater than the percentage of blacks promoted. In short, appellants chose a benchmark which artificially inflated the percentage of blacks in the labor pool from which promotions were drawn. Then, when the percentage of blacks actually promoted fell short of this benchmark, appellants argued that the Hospital engaged in a pattern and practice of discriminatory promotions. This overstatement of any disparity in the selection of blacks was exacerbated by the fact that only twenty-five percent of promotions at the Hospital during the time period at issue were competitive promotions.
By contrast, the Hospital chose as its benchmark the percentage of black employees within each job category at the Hospital.27 We call this pool the “workforce” benchmark.28 By use of the workforce benchmark, the Hospital compared the percentage of blacks in the workforce to the percentage of blacks granted both competitive and noncompetitive promotions. The workforce benchmark is, in fact, based on the pool from which noncompetitive promotions were drawn. Because the percentage of blacks in the Hospital workforce was less than the percentage of black applicants for competitive jobs, the workforce benchmark potentially understated discrimination with respect to competitive promotions. Thus, the workforce benchmark also skewed the statistical results, although not as badly as did the applicant flow benchmark because, inter alia, seventy-five percent of promotions were noncompetitive.
In sum, both appellants’ applicant flow benchmark and appellees’ workforce benchmark skewed the results of their statistical analyses: the applicant flow benchmark overstated any disparity while the workforce benchmark understated any disparity. Because, however, seventy-five percent of promotions were noncompetitive, the distorting effect of the appellants’ applicant flow benchmark was much greater than the distorting effect of the appellees’ workforce benchmark. The latter came closest to the “ideal” analysis and, accordingly, the special master and district court were not clearly erroneous in rejecting the applicant flow benchmark. See Payne, 673 F.2d at 826 (holding that the court’s decision to accept one party’s statistical evidence was not clearly erroneous).
Appellants rejoin that the district court was required, as a matter of law, to accept their applicant flow benchmark. They cite *1574several eases for the proposition that courts generally prefer benchmarks based on applicant flow. For example, in International Brotherhood of Teamsters v. United States, 431 U.S. 324, 342 n. 20, 97 S.Ct. 1843, 1856-57 n. 20, 52 L.Ed.2d 396 (1977), the Court pointed out that “evidence showing that the figures for the general population might not accurately reflect the pool of qualified job applicants” would be relevant to undercutting the statistical evidence at issue in that case. Similarly, in EEOC v. Olson’s Dairy Queens, Inc., 989 F.2d 165, 169 (5th Cir.1993), the court held that the district court had erred by not fully considering an applicant flow analysis. “[T]he ‘most direct route to proof of racial discrimination in hiring is proof of disparity between the percentage of blacks among those applying for a particular position and the percentage of blacks among those hired’_” Id. at 169 (quoting Hester v. Southern Ry., 497 F.2d 1374, 1379 (5th Cir.1974)). See also Payne v. Travenol Laboratories, Inc., 673 F.2d 798, 820-24 (5th Cir.), cert. denied, 459 U.S. 1038, 103 S.Ct. 451, 452, 74 L.Ed.2d 605 (1982) (finding applicant flow data extremely useful in detecting and proving discrimination).
Further, appellants argue that the Supreme Court’s decision in Ward’s Cove Packing Co. v. Atonio, 490 U.S. 642, 109 S.Ct. 2115, 104 L.Ed.2d 733 (1989), is dispositive of the statistical issue in this case. In Ward’s Cove, plaintiffs’ statistics compared the percentage of nonwhite workers in cannery positions to the percentage of nonwhite workers in noneannery positions.29 Plaintiffs sought to make out their prima facie case of Title VII disparate impact by showing that the percentage of nonwhite workers was much higher for cannery positions than for noncan-nery positions. The Court rejected this use of statistics holding that “the pool of cannery workers cannot be used as a surrogate for the class of qualified job applicants [for non-cannery positions] because it contains many persons who have not (and would not) be noncannery job applicants.” Id. at 653, 109 S.Ct. at 2123.
Appellants’ reliance on Ward’s Cove is misplaced and “ ‘fundamentally miscon-ceivefs] the role of statistics in employment discrimination cases.’ ” Id. at 650, 109 S.Ct. at 2121 (quoting Hazelwood School Dist. v. United States, 433 U.S. 299, 308, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977)). The general rule from Ward’s Cove as to the choice of a statistical benchmark is that the “proper comparison [is] between the racial composition of [the at-issue jobs] and the racial composition of the qualified ... labor market.” Id. In this case, the “at issue jobs” were those to which employees were promoted. The composition of the qualified or eligible labor market differed, however, depending on whether the promotion was “competitive” or “non-competitive.” The special master did not err in rejecting appellants’ applicant flow benchmark because it would tend to drastically overstate discrimination. As discussed, appellants’ statistical evidence compared apples and oranges; it compared the applicants for a small percentage of positions to all promotions granted. Although appellees’ choice of benchmark also distorted their results, this distortion was much less than that resulting from appellants’ applicant flow benchmark.
Appellants urge us to adopt a per se rule that applicant flow data are the best measure of the pool from which applicants are selected. We decline. Such a per se rule would be nonsensical. Courts should adopt the benchmark which most accurately reflects the pool of workers from which promotions are granted unless that pool has been skewed by other discriminatory hiring practices.30 Where an application is required for promotion, it will often be appropriate to use a benchmark which calculates the percentage of black applicants. Where, as here, no application is required for most promotions, it makes no sense to compare the percentage of black applicants (for other positions) to the percentage of black noncompetitive appointees.
The usefulness of statistical data in assessing discriminatory practices depends ... on the validity of the basic reference population as the pole star being compared to the work force of the employer ... and *1575that ... [i]n a disparate treatment case, the statistical evidence must be ‘finely tuned’ to compare the employer’s relevant workforce with the qualified populations in the relevant labor market.
Olson’s Dairy Queens, 989 F.2d at 168 (quotation omitted).31
In sum, we affirm the judgment of the district court with respect to the pattern and practice claim.
III. CONCLUSION
Accordingly, for the foregoing reasons, the decision of the district court is
AFFIRMED IN PART, VACATED IN PART, AND REMANDED.