Carlotta Rossini and Jane Zukofsky, on Behalf of Themselves and All Persons Similarly Situated v. Ogilvy & Mather, Incorporated

MESKILL, Circuit Judge: Carlotta Rossini and Jane Zukofsky appeal from a judgment of the United States District Court for the Southern District of New York, Gagliardi, J., that dismissed their individual and class complaints of sex discrimination in employment, after a six week bench trial. The complaints were brought under Title VII of the Civil Rights Act of 1964, as amended and presently codified at 42 U.S.C. §§ 2000e to 2000e-5 (1982). Rossini and Zukofsky raise numer *594 ous claims of error ranging from early discovery and class certification rulings to the final disposition of the case by the district court. Because several of these claims have merit, we reverse and remand for further proceedings. BACKGROUND The facts and procedural history of this case are generally set forth in two published opinions of the district court, Rossini v. Ogilvy & Mather, Inc., 597 F.Supp. 1120 (S.D.N.Y.1984) (first opinion), and Rossini v. Ogilvy & Mather, Inc., 615 F.Supp. 1520 (S.D.N.Y.1985) (second opinion). We presume the reader’s familiarity with both opinions and set out here only a brief sketch of the case’s background. Additional details will be added as necessary during our discussion of the issues raised on appeal. The two individual plaintiffs-appellants are professional employees of the defendant-appellee, Ogilvy & Mather, Incorporated (O & M), 1 a large advertising agency with an office in New York City. They first filed suit in 1978 alleging sex discrimination. After an initial denial of class certification, a class was eventually certified in 1979 with only Zukofsky as class representative. Judge Brieant, to whom the case had originally been assigned, recused himself in 1981. Trial before Judge Gagliardi began in early October 1983 and ended in mid-November of that year. Judge Gagliardi’s first opinion, issued in November 1984, decertified the class as to several issues and then disposed of one of the two remaining class claims and disposed of all of Rossini’s and most of Zukofsky’s individual claims on the merits. Judge Gagliardi reserved decision on the class and Zukofsky salary discrimination claims pending the receipt of additional evidence pursuant to his reversal of a magistrate’s discovery order. That order had earlier been affirmed by Judge Brieant. 597 F.Supp. at 1164-68, 1174. Judge Gagliardi received the additional evidence together with a stipulation of the parties waiving any additional hearing. On August 27, 1985, Judge Gagliardi issued his second opinion, in which he rejected the class and Zukofsky salary discrimination claims on the merits. The plaintiffs appeal from the final judgment filed August 30, 1985. DISCUSSION Rossini and Zukofsky argue that the district court erred in (1) denying class representative status to Rossini and excluding all officers from the class; (2) partially decertifying the class represented by Zukofsky; (3) issuing three discovery orders; (4) refusing to. admit certain statistical evidence and evaluating the statistical evidence that was admitted; (5) disposing of the individual retaliation claims; and (6) disposing of Zukofsky’s salary discrimination claim. These and several other subsidiary arguments will be addressed below. 1. Denial of Class Representative Status to Rossini The adequacy of a class representative, like other class certification questions, is a matter committed to the sound discretion of the district court. Malchman v. Davis, 761 F.2d 893, 899 (2d Cir.1985), cert. denied, — U.S. —, 106 S.Ct. 1798, 90 L.Ed.2d 343 (1986); see also Society for Good Will to Retarded Children, Inc. v. Cuomo, 737 F.2d 1239, 1243 (2d Cir.1984). Accordingly, we will reverse a class representative decision only where the court has abused its discretion. Such an abuse occurred here. Judge Brieant first denied class representative status to Rossini, who is a vice president of O & M, in a memorandum and order issued on October 18, 1978. Rossini v. Ogilvy & Mather, Inc., 80 F.R.D. 131 (S.D.N.Y.1978). Judge Brieant found that the combination of Rossini’s “desire to gain *595 ... credentials” as a political activist and her agency status as an 0 & M officer created conflicts of interest which prevented her from serving as a representative of a class that included non-officers. Id. at 135-36. In an affidavit in support of a motion to reconsider this decision, Rossini’s attorney stated that “Vice Presidents of Ms. Rossini’s level have virtually no managerial authority, and ... no actual or apparent authority ... to bind the Corporation.” J.App. at 573. The attorney asked for a hearing or an opportunity to file evidence in order to demonstrate that the conflict of interest finding was erroneous. Id. at 573-74. Judge Brieant neither granted nor denied the request for a hearing but described his decision on Rossini’s status as “conditional,” subject to revision based on facts to be disclosed during discovery. Id. at 620-23. In a subsequent memorandum and order, Judge Brieant reaffirmed his denial of class representative status to Rossini. Rossini v. Ogilvy & Mather, Inc., 20 Empl. Prac.Dec. (CCH) if 30,045 (S.D.N.Y. Apr. 18, 1979). He limited his earlier decision, however, stating that it had not been based on Rossini’s political activism but rather on the fact that Rossini as a Vice President of Ogilvy encounters an unavoidable conflict between her role as agent for the corporation and her role as fiduciary for a class maintaining litigation against that corporation. Rossini as a present officer of Ogilvy has an inherent and continuing conflict with regard to managing class litigation against that corporation on behalf of non-officers or non-director employees. Id. at 11,251. In accordance with his decision on Rossini’s status, Judge Brieant also excluded all officers from the class. He certified a class consisting of “all females (excluding corporate officers) who are, have been, or will be, or were since May 30, 1975, employed by defendant Ogilvy & Mather, New York, as managers or professionals and who have been, are, continue to be, or would be affected by the discriminatory practices of defendant.” J.App. at 729. The class was certified with only Zukofsky as its representative. During trial before Judge Gagliardi, motions to reverse these decisions by allowing Rossini to serve as representative of the class and including most officers in the class were explicitly forecast in opening argument, Tr. 13. 2 The motion as to Rossini was finally denied near the end of the trial. Tr. 3559. The denial of class representative status to Rossini based solely on her purported status as a corporate agent of O & M is irreconcilable with the facts found by the trial court. Frances Devereux, personnel director at O & M, testified that the title of vice president was “basically a [sic] honorarium” which carried with it no real change in job function. Tr. 25. O & M president Kenneth Roman testified that election as an “officer” of O & M brought with it no change in function, no policy making authority and no authority to bind the company in any way. Tr. 825-27. After hearing such testimony, Judge Gagliardi found that 150 O & M employees — or about one-quarter of the official, professional and managerial staff — had officer titles, the lowest of which was vice president, and that receipt of such titles brought no change in duties, salary or authority. 597 F.Supp. at 1126-28, 1167. Rossini’s vice-presidential title was, therefore, an insufficient basis for concluding that she was not an adequate representative of a class of female professionals and managers at O & M. The decision to recognize or deny class representative status should be based on substance, not mere form. See Sheehan v. Purolator, Inc., 103 F.R.D. 641, 651 n. 10 (E.D.N.Y.1984) (actual agency, not mere corporate title, determined class representative status). *596 O & M argues on appeal that Rossini was denied representative status because of her minor role in personnel decisionmaking. This argument is not persuasive. O & M uses an “account group” structure in which a group of employees services a client under the leadership of an “account group director.” Thus, like Rossini, many members of the class certified by Judge Brieant were supervisors of other employees and were involved in making recommendations for promotion of their underlings. As we have noted, one segment of the class was “managers,” a title necessarily implying such supervisory responsibilities. Furthermore, the argument that Rossini’s actual personnel role was the basis for the denial of representative status is refuted by Judge Brieant’s plainly worded memoranda on the subject. In his first memorandum, Judge Brieant presumed that Rossini’s officer title made her a corporate agent and that this agency entailed a responsibility for “compliance by the corporation with proper personnel policies.” 80 F.R.D. at 135. The second memorandum made clear that it was this presumed agency which created the supposed conflict of interest. 20 Empl.Prac.Dec. at 11,251. Thus, the finding that Rossini was a corporate agent was clearly erroneous. Judge Gagliardi’s refusal to grant representative status to Rossini after the error in Judge Brieant’s “conditional” decision had become clear was an abuse of discretion. The court’s exclusion of officers from the class, apparently based on its clearly erroneous finding as to the agency status of O & M officers, was another abuse of discretion. While it would have been reasonable, as suggested at trial by the plaintiffs, see Tr. 13, to exclude only that narrow segment of the O & M officers who actually had the sort of agency powers that concerned the court, it was not reasonable to exclude an entire group of professional employees simply because they possessed what all of the evidence indicated were essentially honorary titles. The erroneous decision to deny class representative status to Rossini and to exclude her and other non-controlling officers from the class is a basis for reversal if it was prejudicial. We conclude that it did cause prejudice to the class, as more fully discussed in parts 2(b) and 3(a), infra. 2. Decertification of the Class Represented by Zukofsky O & M moved the court in August 1982 to decertify the class represented by Zukofsky, relying on the recent decision of the Supreme Court in General Telephone Company of the Southwest v. Falcon, 457 U.S. 147, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982). Judge Gagliardi did not act on the motion until after trial. Then, in his first opinion, he analyzed the class claims under the standards set forth in Falcon and granted the decertification motion as to the claims of discrimination in transfer, promotion and training at O & M. 597 F.Supp. at 1129-39, 1172. Zukofsky, as an individual, asserted that she was a victim of a discriminatory denial of transfers from one O & M department to another. Judge Gagliardi’s decertification of the class transfer discrimination claim was based on his conclusion that the plaintiffs had failed to prove the existence of any class of employees suffering from such discrimination. 597 F.Supp. at 1134. See 7 C. Wright and A. Miller, Federal Practice and Procedure § 1760 (1972) (hereinafter “Wright”). This conclusion was supported by the evidence and was not an abuse of discretion. The decertifications as to discrimination in promotion and training, however, were predicated, under Falcon and Fed.R.Civ.P. 23(b)(3), on the notion that Zukofsky’s individual claims were insufficiently similar to those made on behalf of the class. 597 F.Supp. at 1135-37, 1139. These two decertification decisions require further analysis. a. Falcon In Falcon, the Court held that it was error for the district court to presume, *597 “[w]ithout any specific presentation identifying the questions of law or fact that were common to the claims of [the plaintiff] and of the members of the class he sought to represent,” that the Mexican-American plaintiffs claim of discrimination in promotion was “typical” of the claims of other Mexiean-Americans who had not been hired or promoted by the same employer. 457 U.S. at 158-59, 102 S.Ct. at 2371. The Court's rationale was that such a presumption ignored the requirements of Fed.R.Civ.P. 23(a), under which a putative class representative must demonstrate “ ‘numerosity, commonality, typicality, and adequacy of representation,! ” the purpose of those requirements being to “ ‘limit the class claims to those fairly encompassed by the named plaintiffs claims.’ ” Id. at 156, 102 S.Ct. at 2370 (quoting General Telephone Co. of the Northwest v. EEOC, 446 U.S. 318, 330, 100 S.Ct. 1698, 1706, 64 L.Ed.2d 319 (1980)). The Court focused on the requirement of typicality, noting that it and commonality “tend to merge.” Id. at 157 & n. 13, 102 S.Ct. at 2370 & n. 13. Rather than a presumption of typicality, the Court directed that there be “a rigorous analysis” to determine whether Rule 23(a) has been satisfied. Id. at 161, 102 S.Ct. at 2372. The Falcon Court did not hold that a plaintiff asserting one particular type of employment discrimination claim can never represent a class of employees asserting other types. The Court did decide that Rule 23(a) may not be satisfied merely by showing that the named plaintiff is a member of an identifiable class of persons allegedly subject to class discrimination. However, the Court observed in a footnote: [i]f [an employer] used a biased testing procedure to evaluate both applicants for employment and incumbent employees, a class action on behalf of every applicant or employee who might have been prejudiced by the test clearly would satisfy the commonality and typicality requirements of Rule 23(a). Significant proof that an employer operated under a general policy of discrimination conceivably could justify a class of both applicants and employees if the discrimination manifested itself in hiring and promotion practices in the same general fashion, such as through entirely subjective decisionmaking processes. Id. at 159 n. 15, 102 S.Ct. at 2371 n. 15. (emphasis added). In a separate concurrence, in which no other member of the Court joined, Chief Justice Burger implied that the hiring and promotion claims might properly have been asserted by the same class representative in Falcon if it had been shown that the same people were involved in making both hiring and promotion decisions. Id. at 162 n. *, 102 S.Ct. at 2373 n. * (Burger, C.J., concurring in part and dissenting in part). In the wake of Falcon, courts have been generally strict in their application of the Rule 23(a) criteria. See, e.g., Roby v. St. Louis Southwestern Railway Co., 775 F.2d 959, 962 (8th Cir.1985) (party alleging one type of discriminatory discharge could not represent class suffering from other type of discriminatory discharge); Walker v. Jim Dandy Co., 747 F.2d 1360, 1364-65 (11th Cir.1984) (applicants for supervisory position could not represent class of applicants for lower level positions; applicants asserting discrimination in “hiring” could not represent class alleging discrimination in “recruitment”); see generally 7 Wright § 1771 at 368 & n. 58.6 (Supp.1985). In Holsey v. Armour & Co., 743 F.2d 199, 216 (4th Cir.1984), the Fourth Circuit went so far as to incorporate Chief Justice Burger’s concurrence into the rule of Falcon, holding that employee plaintiffs asserting promotion discrimination claims could not represent a class of outside applicants for the same jobs, despite their showing that the same standards, application procedure and “entirely subjective criteria” were used as to both groups, absent an additional showing that the same people had made both hiring and promotion decisions. Some courts have recognized, as we do, that the primary thrust of Falcon was that satisfaction of Rule 23(a) requirements may not be presumed. Weiss v. York Hospital, 745 F.2d 786, 810 & n. 37 (3d Cir. *598 1984) (holding that physician who actually applied for staff privileges at hospital could represent class including those dissuaded from applying by allegedly discriminatory policy in restraint of trade), cert. denied, — U.S.—, 105 S.Ct. 1777, 84 L.Ed.2d 836 (1985); Sheehan v. Purolator, Inc., 103 F.R.D. at 648. See also Paxton v. Union National Bank, 688 F.2d 552, 562 (8th Cir.1982) (burden of showing typicality not onerous, but requires “something more than general conclusory allegations”), cert. denied, 460 U.S. 1083, 103 S.Ct. 1772, 76 L.Ed.2d 345 (1983). The Supreme Court, itself, seems to share this view. See Cooper v. Federal Reserve Bank of Richmond, 467 U.S. 867, 877, 104 S.Ct. 2794, 2801, 81 L.Ed.2d 718 (1984) (“Falcon thus holds that the existence of a valid individual claim does not necessarily warrant the conclusion that the individual plaintiff may successfully maintain a class action.”). Zukofsky did not ask the district court to presume that she was a proper representative of a class asserting claims of discrimination in promotion and training. Rather, she attempted to prove that O & M had denied women opportunities to advance themselves, either as she sought to do, by transfer, or as other class members sought, by training or promotion, through its use of a subjective evaluation system without formal job descriptions, objective experience or education requirements, posting of job openings or listing of employees seeking better positions. Moreover, evidence was offered to show that many of the decisions affecting employees’ opportunities for transfer, training and promotion were made by the same, central group of people within O & M. Thus, Zukofsky’s evidence, if believed, would indicate that O & M discriminated “in the same general fashion” against Zukofsky in her quest for transfer and against other class members seeking training or promotion. Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. at 2371 n. 15; see also Brown v. Eckerd Drugs, Inc., 564 F.Supp. 1440, 1446 (W.D.N.C.1983) (commonality found to exist where discrimination in individual demotion and discharge claims was “manifested in the same fashion” as discrimination in class promotion and transfer claims). No more was necessary to satisfy the typicality requirement of Rule 23(a). b. Rule 23(b)(3) Judge Gagliardi presumed that Judge Brieant had certified the class here pursuant to Fed.R.Civ.P. 23(b)(3) because Judge Brieant had required that notice be sent to class members advising them of their ability to opt out of the class. 597 F.Supp. at 1132; see Fed.R.Civ.P. 23(c)(2) (requiring notice of opportunity to opt out of classes certified under Rule 23(b)(3)). While this presumption was not necessarily correct, see Plummer v. Chemical Bank, 668 F.2d 654, 657 n. 2 (2d Cir.1982) (noting split in circuits over whether opting out is permissible under Rule 23(b)(2)), it is not challenged on appeal. Appellants note that the parties had agreed that the class should be certified, if at all, under subsection 23(b)(2) instead, Br. of Plaintiffs-Appellants at 11 n. 11, but they do not argue that the application of subsection 23(b)(3) was erroneous. Accordingly, we will deem any such claim waived and will analyze Judge Gagliardi’s application to the class promotion and training claims of the 23(b)(3) requirement that “the questions of law or fact common to the members of the class predominate over any questions affecting only individual members.” First, we note that Zukofsky’s satisfaction of the typicality requirement of Rule 23(a), as set forth above and in Falcon, goes a long way toward satisfying the Rule 23(b)(3) requirement of commonality. See Falcon, 457 U.S. at 157 n. 13, 102 S.Ct. at 2370 n. 13 (commonality and typicality tend to merge). Zukofsky’s individual claims of discriminatory placement and discriminatory denial of a transfer into another department, like the class claims of discrimination in training and promotion, require a showing of discriminatory intent on the part of O & M. If such intent were proved through evidence on the class claims, that *599 would assist Zukofsky in proving her individual claim. See International Brotherhood of Teamsters v. United States, 431 U.S. 324, 362, 97 S.Ct. 1843, 1868, 52 L.Ed.2d 396 (1977) (proof of pattern or practice of discrimination supports inference that “any particular employment decision, during the period in which the discriminatory policy was in force, was made in pursuit of that policy”). Much of the plaintiffs’ evidence at trial consisted of statistics designed to prove this intent. Thus, discriminatory intent clearly was a major issue of law and fact shared in common by the individual and class plaintiffs. In addition to the common intent issue, the class training claim and Zukofsky’s placement claim are factually similar. As the district court noted, the class plaintiffs were arguing that opportunities for training depended directly on an employee’s assignment to a particular task and that certain tasks were not assigned to women. See 597 F.Supp. at 1139. The court refused to recognize Zukofsky as a representative as to such claims, however, because she had not suffered such a discriminatory assignment of tasks within her department. Id. The parallel that the court failed to see was that Zukofsky was claiming that the assignment to her department was itself discriminatory. Both Zukofsky and the class alleged that there was a discriminatory system in which a small group of company officials assigned them to departments or tasks within departments dealing with what was viewed by the assigning officials as “women’s work.” See id. at 1137-38. Common issues of law and fact would necessarily be involved in proving the existence and functioning of that system. Commonality was not destroyed merely because Zukofsky’s individual claim also required proof of some facts that differed from those of the class claim. See Holsey v. Armour & Co., 743 F.2d at 216-17. The court also overlooked the common thread running between Zukofsky’s individual transfer discrimination claim and the class claim of discrimination in promotion. The court distinguished the two claims, describing the class claim as centering on proof of discriminatory evaluation of performance in a single position and Zukofsky’s claim as centering on proof of discriminatory barriers preventing movement from one position to another. 597 F.Supp. at 1136. Those barriers, however, consisted in large part of allegedly discriminatory evaluations of Zukofsky’s performance. See 597 F.Supp. at 1170-71. She and the class were both alleging that they had to “perform” at a higher level than men in order to advance their careers. The distinction that Zukofsky was performing on creativity tests while class members were performing on day-to-day work assignments is of little importance in light of the allegation that it was the same standard-less, subjective evaluation system that operated in both cases. Zukofsky and the class also alleged that they jointly confronted another barrier to transfer and promotion: 0 & M’s failure to publish a list of openings or to maintain lists of employees interested in particular assignments within their own or other departments. In sum, these common issues of law and fact predominated over those that separated Zukofsky from the class. Thus, it was an abuse of discretion to decertify the class represented by Zukofsky as to the promotion and training claims either under Falcon or under Rule 23(b)(3). However, even if that decision were correct with Zukofsky as the representative, we would still reverse it in light of our holding that it was error to prevent Rossini from representing the class. Unlike Zukofsky, Rossini was alleging individual claims of discrimination in both training and promotion. Proof of her claims, therefore, aided the class case directly, rather than by analogy as was true with the claims raised by Zukofsky. Rossini as class representative shared essentially all issues of law and fact going to the claims of training and promotion in common with the class. Had she not been eliminated as class representative, the litigation over the certification of these claims would have been much more easily resolved. Indeed, the motion to de *600 certify under Falcon might not have been made. This additional litigation burden was one aspect of the prejudice suffered by the class as a result of the denial of representative status to Rossini. 3. Discovery Orders The plaintiffs complain of three orders issued by the court during the discovery phase of this case: (a) the order denying production of the personnel files of employees promoted to officer status at O & M; (b) an order that permitted the plaintiffs’ attorneys, but not the individual plaintiffs themselves, to review the personnel files that were produced; and (c) an order that limited communication between the parties and members of the class. We will discuss each in turn. a. Officer Files The order denying production of officer files was issued by a magistrate and affirmed by Judge Brieant. 3 It was finally reversed by Judge Gagliardi in his first opinion. 597 F.Supp. 1167-68. 4 The issue on appeal is not the magistrate’s order but the propriety of the limitations Judge Gagliardi placed on the plaintiffs’ use of the officer file evidence, once he had ordered its production. These limitations might be characterized more properly as rulings on the admission of evidence rather than on the scope of discovery. The abuse of discretion standard controls our review in either event. See Long Island Lighting Co. v. Barbash, 779 F.2d 793, 795 (2d Cir.1985) (discovery order); In re Martin-Trigona, 760 F.2d 1334, 1344 (2d Cir.1985) (evidentiary ruling). Seeking to ensure that “neither side [could] take advantage of the initial erroneous discovery ruling,” Judge Gagliardi strictly limited the parties’ use of information gained as a result of this ruling’s reversal. 597 F.Supp. at 1167. He ordered that the new data be analyzed in the same way as prior data, that no new variables be used, that the new data not be used to alter prior data as to non-officers, and that the studies based on the new data pertain solely to the class salary discrimination claim. Id. at 1167. His rationale for the latter restriction was that “there is an insufficient showing that the officer data significantly could affect the evidence submitted at trial with regard to the other issues in this case.” Id. at 1167-68. When both parties subsequently attempted to make wider use of the new data, Judge Gagliardi chastized them and strictly enforced his limits on the uses to which the data could be put. 615 F.Supp. at 1521 & n. 2. Despite the plaintiffs’ continued efforts over a period of years to overturn it, the discovery order was still operating during the trial, preventing plaintiffs from making full use of any relevant anecdotal or statistical evidence possibly contained in the officer files. While the trial court’s desire to limit the damage caused by its tardy decision to reverse the order is understandable, the court’s order, in effect, punished the plaintiffs for not having succeeded earlier in winning the reversal. The limits imposed on use of the evidence cannot be justified, as the trial court sought to do, by the speculative rationale that the evidence, as yet unseen, would have no significant effect on the various *601 issues at stake in the case other than the class salary claim. The court abused its discretion in limiting the use of the officer file evidence on that basis. On remand, the plaintiffs must be allowed to make full, appropriate use of the officer file evidence. Of course, the district court may exercise its discretion in dealing with any specific objections to admission of the evidence as to particular claims. b. Access to Files by Individual Plaintiffs The plaintiffs next complain of an order, issued by Judge Brieant, which permitted only their lawyers to have direct access to the personnel files produced by 0 & M in the course of discovery. This order, dated June 6, 1980, carries Judge Brieant’s handwritten notation that it was issued “on consent after hearing counsel this date.” J.App. at 899. Judge Gagliardi specifically found, over plaintiffs’ protests to the contrary at trial, that this order was, in fact, entered on consent. 597 F.Supp. at 1146 n. 30. He further held that Rossini’s failure to move to amend the order prior to trial precluded her from challenging it. Id. While Judge Gagliardi was wrong about plaintiffs’ failure to move to amend the order, see J.App. at 1331, 1333, we have no basis for holding that his finding that they consented to the order was clearly erroneous. On its face, the order appears to have been a proper attempt to balance the plaintiffs’ desire for full disclosure of relevant information against the defendant’s desire to preserve the privacy of its employees. Cf. Gray v. Board of Higher Education, 692 F.2d 901, 907-08 (2d Cir.1982) (advocating balancing approach regarding the disclosure of plaintiff’s tenure file). The order did not prevent Rossini’s attorney from offering into evidence any relevant information the files might contain, cf. Lynn v. Regents of the University of California, 656 F.2d 1337, 1345-48 (9th Cir.1981) (denial of access to tenure file to plaintiff and her attorney violated due process), cert. denied, 459 U.S. 823, 103 S.Ct. 53, 74 L.Ed.2d 59 (1982), based on indications from Rossini as to which promotions or assignments she believed she had been discriminatorily denied. Judge Gagliardi’s finding of consent to this order was not clearly erroneous. Therefore, we cannot say that his refusal to reverse the order was an abuse of discretion. c. Order Limiting Communication Between Parties and Class Members In response to a complaint from the plaintiffs about a questionnaire circulated by O & M among its female employees soon after the commencement of this action, and to a counter-complaint from O & M about a public meeting at which Rossini had allegedly discussed the pending suit, Judge Brieant in July 1978 issued an oral order requiring prior court approval of virtually all oral and written communications between the parties and potential members of the class. As authority for the order, Judge Brieant relied on the Manual for Complex Litigation, cf. Manual for Complex Litigation Second § 30.24 (1985), reprinted in 1 J. Moore, Moore’s Federal Practice Part 2 (2d ed. 1986), and on Weight Watchers of Philadelphia, Inc. v. Weight Watchers International, Inc., 455 F.2d 770 (2d Cir.1972), a case which held only that such orders are not immediately appealable, id. at 772-75. More than two years later, the plaintiffs moved to vacate the order. Judge Brieant instead limited it, on January 5, 1981, to written communications only. At that time, the plaintiffs’ attorney stated that she had no objection to the requirement that any written communications with the class receive prior court approval. J.App. at 976. More than a year thereafter, the plaintiffs moved to set aside and vacate the remainder of the order pursuant to the Supreme Court’s then recent decision in Gulf Oil Co. v. Bernard, 452 U.S. 89, 101 S.Ct. 2193, 68 L.Ed.2d 693 (1981). The motion was apparently never acted on. In Gulf Oil, the Court struck down an order similar to the one at issue here as an abuse of discretion, stating that *602 an order limiting communications between parties and potential class members should be based on a clear record and specific findings that reflect a weighing of the need for a limitation and the potential interference with the rights of the parties. Only such a determination can ensure that the court is furthering, rather than hindering, the policies embodied in the Federal Rules of Civil Procedure, especially Rule 23. In addition, such a weighing — identifying the potential abuses being addressed — should result in a carefully drawn order that limits speech as little as possible, consistent with the rights of the parties under the circumstances. Id. at 101-02, 101 S.Ct. at 2200 (footnotes omitted). The Court pointedly added that absent a clear record and specific findings of need, the Manual for Complex Litigation could not authorize such an order. Id. at 104, 101 S.Ct. at 2201. Unlike the order at issue in Gulf Oil, see id. at 93, 101 S.Ct. at 2196, the order here was based on specific findings of fact. The findings were sufficient to distinguish this case from Gulf Oil. Over the course of two hearings, Judge Brieant found that both parties had engaged in improper conduct which posed the dangers of “possible chilling of the rights of the potential class members or ... seeming to pressure any of them unduly to opt out of the class ... or ... creating confusion,” J.App. at 123, and “threatenfed] the essence of fairness and due process” in class actions, id. at 126. Judge Brieant stated that he was imposing the restriction on future communications between the parties and the class in order to guard against potential abuses of the class action process. Having placed these findings of fact and conclusions of law on the record, the judge then offered the parties an opportunity to request additional findings. Neither party at that time requested that the court’s order be more narrowly drawn. The order issued here may have been broader than necessary to curb the abuses found by the court, but we will not hold that its issuance was an abuse of discretion absent a complaint at the time of issuance about the order’s breadth. The court was making a reasonable effort to respond to real and potential problems and might well have drawn the order more narrowly if the plaintiffs had asked it to do so. 4. Admission and Evaluation of Evidence The plaintiffs challenge several of the district court’s evidentiary rulings and determinations. Chief among the challenges is the claim that the court erroneously refused to admit certain statistical studies offered as rebuttal evidence during trial. In addition, plaintiffs argue that the court gave undue weight to the presence or absence of variables measuring military experience, that it improperly required plaintiffs to supplement their statistical case with anecdotal evidence and that it erred in failing to apply both discriminatory treatment and disparate impact analyses to the evidence. We discuss each argument in turn. a. Green Report Much of the evidence offered by both parties in this case consisted of statistical studies of the O & M workforce. In order to measure certain employees’ experience, plaintiffs’ statistical expert chose to “construct” dates of graduation for those employees based on age, so that the date of graduation from high school was assumed to be 18, from junior college or technical school, 20, from college, 22 and from graduate school, 24. Some time before trial, an expert report for the defense generally criticized the practice of constructing data. At trial, O & M objected to the introduction of statistical reports containing the constructed data, arguing that the constructed dates were frequently wrong and that the results of studies incorporating the data were, therefore, unreliable. In response to these objections, the plaintiffs introduced what has been characterized in the parties’ briefs and the opinions below as the “green report,” a study which com *603 pared the original results obtained with constructed data to new results obtained with actual graduation data in an áttempt to prove that the claim of unreliability was false. In accordance with the practice being followed by the parties, the plaintiffs at the time of submitting the report supplied O & M with the computer tapes on which it was based. They did not indicate until later, however, that the actual graduation dates used in the report had not come from the computer data base transmitted to the defense during discovery. The dates had been gleaned instead directly from personnel files supplied by 0 & M. After initially admitting the green report into evidence, the trial court reversed itself. It ruled in its first opinion that 0 & M’s objection to the report would be sustained. 597 F.Supp. at 1163. The court reasoned that allowing admission of the report during trial would have been unduly burdensome on the defense, especially in light of the pretrial objection to construction of data, because such admission would require the defense to obtain and verify information outside the data base. The court also relied on Fed.R.Civ.P. 16(e), which allows the modification of pretrial orders only to avoid manifest injustice. Both grounds for exclusion of the green report are untenable. Because of the exclusion’s apparently prejudicial effect on the outcome of the case, we reverse the district court’s decision on this issue. Although the information used in the green report was not in the plaintiffs’ original computer data base, that alone was not enough to render the report’s admission overly burdensome. The information, after all, came directly from defendant’s own files. Culling graduation dates from the 144 files at issue in order to verify the plaintiffs’ figures would not have been a very great task. The defense had obviously already done some of the work, or work similar to it, in order to substantiate its argument that the constructed dates originally used by the plaintiffs were inaccurate. Furthermore, the green report was not in violation of the pretrial order. In that order, plaintiffs specifically reserved their right to introduce rebuttal exhibits. J.App. at 1976. Certainly, the fact that the new tables set forth in the green report were styled “supplemental” did not affect their essential nature as rebuttal evidence. We do not hold that a party may gain admission of any kind or amount of evidence via a boilerplate reservation of rights and a claim that the evidence serves to rebut some insignificant objection. Here, however, the exclusion of the green report had a crippling, if not fatal, effect on plaintiffs’ largely statistical case. Having excluded this evidence, which might have negated the argument that the construction of graduation dates rendered plaintiffs’ studies unreliable, the trial court then proceeded to describe the discrepancies between the actual and estimated graduation dates as “egregious” and to find that the “errors in the constructed graduation data seriously undermine[d] the value” of several of plaintiffs’ key statistical tables. 615 F.Supp. at 1525; see also id. at 1530, 1542. Although the district court found other flaws in plaintiffs’ statistical studies as well, we think it is sufficiently likely that the erroneous exclusion of the green report had an effect on the outcome to require a reversal on this ground. b. Military Experience Plaintiffs next argue that the trial court accorded undue weight to the incorporation in 0 & M’s statistical studies of variables designed to register an employee’s military experience. They argue that “military experience” is a proxy for “male” and that there was no evidence to indicate that military experience was actually considered in O & M job decisions. . The trial court’s decision to credit the O & M studies which incorporated military experience variables came after extensive testimony from experts on both sides of the issue. Although, as factfinders, we might not have reached the same conclu *604 sion, we cannot say that the trial court’s reliance on these studies was clearly erroneous. See Coser v. Moore, 739 F.2d 746, 751 (2d Cir.1984). We hold, however, that it was an abuse of discretion to reject one of plaintiffs’ statistical tables, “Second Table C,” on the speculative basis that the table’s results might “possibly” have been different if military experience had been included. 615 F.Supp. at 1545. No evidence was offered during trial to support that speculation. See id. & n. 35. While the court could reasonably find that military experience was a permissible variable, it had no basis for holding, in effect, that such a variable was required in order for a statistical study to be valid. See Bazemore v. Friday, — U.S. —, 106 S.Ct. 3000, 3007-11, 92 L.Ed.2d 315 (1986) (Brennan, /., for a unanimous Court, concurring in part) (regression analysis excluding some variables may still be valid evidence of employment discrimination); Eastland v. TVA, 704 F.2d 613, 623 (11th Cir.1983) (variables in statistical models should be those likely to have actual effect in decisionmaking process under study), cert. denied, 465 U.S. 1066, 104 S.Ct. 1415, 79 L.Ed.2d 741 (1984). 5 c. Anecdotal Evidence Plaintiffs argue that the trial court improperly held that this employment discrimination case could not be proved on the basis of statistical evidence alone. They misread the trial court’s opinion. The court actually stated that “strong statistical proof ... is critical [in the instant case], since there is no anecdotal evidence supporting the figures.” 615 F.Supp. at 1545 (footnote omitted). The court was not deviating from the rule, stated in International Brotherhood of Teamsters v. United States, 431 U.S. 324, 339, 97 S.Ct. 1843, 1856, 52 L.Ed.2d 396 (1977), that employment discrimination may be established by statistical proof. See also United States Postal Service Board of Governors v. Aikens, 460 U.S. 711, 714 n. 3, 103 S.Ct. 1478, 1481 n. 3, 75 L.Ed.2d 403 (1983) (discriminatory intent may be proved by circumstantial evidence). It was merely noting the importance, also recognized in Teamsters, of using anecdotal evidence, when available, to bring “the cold numbers convincingly to life.” 431 U.S. at 339, 97 S.Ct. at 1856; see also Coser, 739 F.2d at 751-52 (lack of anecdotal evidence a major weakness in employment discrimination case); Sobel v. Yeshiva University, 566 F.Supp. 1166, 1186 (S.D. N.Y.1983) (same). “[S]tatistics are not irrefutable; they come in infinite variety and, like any other kind of evidence, they may be rebutted. In short, their usefulness depends on all of the surrounding facts and circumstances.” Teamsters, 431 U.S. at 340, 97 S.Ct. at 1856. In evaluating all of the evidence in a discrimination case, a district court may properly consider the quality of any anecdotal evidence or the absence of such evidence. See, e.g., EEOC v. Sears, Roebuck & Co., 628 F.Supp. 1264, 1278 n. 2 (N.D.Ill.1986) (describing evaluation of statistical and non-statistical evidence in discrimination case). There was no error in the district court’s evaluation of the small amount of anecdotal evidence of discrimination offered by the plaintiffs, see 615 F.Supp. at 1545 n. 37. d. Disparate Impact Analysis The plaintiffs argue that the district court erred in evaluating the evidence in this case only as it related to a claim of discriminatory treatment, rather than also considering their claim as one of disparate *605 impact. An important difference between the two approaches is that discriminatory motive need not be proved to make out a case of unlawful discrimination under the disparate impact analysis, while such motive is an essential element of a discriminatory treatment case. Teamsters, 431 U.S. at 335 n. 15, 97 S.Ct. at 1854 n. 15. The district court found that plaintiffs had tried the case as one of discriminatory treatment and that fairness to O & M prevented the court from considering the disparate impact theory after trial. 597 F.Supp. at 1158 & n. 41. The court noted, however, that 0 & M had successfully defended a disparate impact claim. Id. at 1158 n. 41. The district court was correct in finding that the main thrust of plaintiffs’ case, as tried, was a claim of discriminatory treatment. Because plaintiffs chose to try the case in that manner, it would have been unfair to evaluate the evidence under the disparate impact theory after trial. Ste. Marie v. Eastern Railroad Association, 650 F.2d 395, 399 n. 2 (2d Cir.1981). Moreover, the disparate impact analysis does not appear appropriate to the facts of this case. A claimant asserting disparate impact must show that there are “policies or practices that are neutral on their face and in intent but that nonetheless discriminate in effect against a particular group.” Teamsters, 431 U.S. at 349, 97 S.Ct. at 1861. Thus, “[t]he disparate impact theory has been used mainly in the context of quantifiable or objectively verifiable selection criteria which are mechanically applied and have consequences roughly equivalent to results obtaining under systematic discrimination.” Zahorik v. Cornell University, 729 F.2d 85, 95 (2d Cir.1984); see Furnco Construction Corp. v. Waters, 438 U.S. 567, 575 & n. 7, 98 S.Ct. 2943, 2948 & n. 7, 57 L.Ed.2d 957 (1978) (approving use of disparate treatment rather than disparate impact analysis, distinguishing cases). While other courts have extended the disparate impact analysis to subjective selection criteria, like those allegedly used by O & M, see, e.g., Griffin v. Carlin, 755 F.2d 1516, 1522-25 (11th Cir.1985), we stated recently that “[ejvidence of systematic exclusion by the mechanical application of facially neutral criteria is necessary” in order to prove a disparate impact claim. Zahorik, 729 F.2d at 96. See also Paxton v. Union National Bank, 688 F.2d 552, 556 n. 3 (8th Cir.1982); EEOC v. Sears, Roebuck & Co., 628 F.Supp. at 1281-85; Sheehan v. Purolator, Inc., 103 F.R.D. 641, 646-47 (E.D.N.Y.1984). Such evidence is present where, as in Connecticut v. Teal, 457 U.S. 440, 448, 102 S.Ct. 2525, 2531, 73 L.Ed.2d 130 (1982), a test given to applicants has a disparate impact on an identifiable group. See also Dothard v. Rawlinson, 433 U.S. 321, 328-31, 97 S.Ct. 2720, 2726-28, 53 L.Ed.2d 786 (1977) (disparate impact of height and weight requirements); Griggs v. Duke Power Co., 401 U.S. 424, 429-33, 91 S.Ct. 849, 852-55, 28 L.Ed.2d 158 (1971) (aptitude and intelligence tests, educational requirements). The core of plaintiffs’ case here, however, is that there are no such objective criteria in the O & M employee evaluation process. In now asserting a disparate impact claim, plaintiffs are trying to have it both ways. They argue, on the one hand, that the employee evaluation system at O & M is entirely subjective and, therefore, inherently suspect. On the other hand, in seeking to state a claim of disparate impact, they argue that the same subjective system is facially neutral. Because it would have been logically impossible to prove both propositions, it is not surprising that, as the district court found, the plaintiffs chose one of the two propositions and attempted to prove it at trial. Under the circumstances, the court did not err in treating this as only a discriminatory treatment case. 5. Retaliation Claims The plaintiffs’ retaliation claims require little discussion. The plaintiffs argue that the district court failed to examine their claims that 0 & M retaliated against them for filing complaints with the EEOC and bringing this suit. This argument is meritless. The court did consider the evidence offered by both sides on the retalia *606 tion issue and found for O & M as to both plaintiffs. 597 F.Supp. at 1156-57, 1172. These findings were not clearly erroneous and will, therefore, not be reversed. 6. Dismissal of Zukofsky’s Salary Claim Finally, Zukofsky argues that the district court erred in dismissing her salary discrimination claim without making any findings of fact or conclusions of law on it. This claim has merit. In its first opinion, the district court reserved decision on Zukofsky’s salary claim, pending receipt of the officer file evidence. 597 F.Supp. at 1169. The court promised that the claim would be analyzed in conjunction with that new evidence. Id. In its second opinion, however, the court analyzed the new evidence only as it related to the class salary claim and then dismissed that claim, adding that “Plaintiff Zukofsky’s individual salary discrimination claim is also dismissed in its entirety.” 615 F.Supp. at 1546. No findings of fact or conclusions of law were specifically directed to Zukofsky’s claim. Fed.R.Civ.P. 52(a) requires the court in a bench trial to “find the facts specially and state separately its conclusions of law.” While such findings are not necessary to our jurisdiction, they are an important aid in reviewing a case. SEC v. Frank, 388 F.2d 486, 493 (2d Cir.1968). Accordingly, it has been held that “[i]t is an error for the trial court to fail to make findings of fact, even though no requests for such findings were made.” Squirtco v. Seven-Up Co., 628 F.2d 1086, 1092 (8th Cir.1980). 0 & M argues that Zukofsky’s failure to move the court for additional findings under Rule 52(b) prevents her from complaining of the lack of findings now. O & M relies on Evans v. Suntreat Growers & Shippers, Inc., 531 F.2d 568, 570 (Temp. Emerg.Ct.App.1976), where a party which had failed to propose findings was barred from complaining on appeal that the findings made werp not specific enough. Here, however, Zukofsky submitted the functional equivalent of proposed findings on her individual salary discrimination claim in a post-trial brief. The trial judge had specifically requested that findings be submitted in post-trial memoranda rather than in a formal list of proposed findings. Therefore, even if we were to find Evans persuasive, it would be inapposite. We hold that it was error for the district court to omit specific findings on Zukofsky’s claim and direct that such findings be made on remand. CONCLUSION We hold that the district court erred in (1) denying class representative status to Rossini and excluding non-controlling officers from the class; (2) decertifying the class as to claims of discrimination in training and promotion; (3) limiting the plaintiffs’ use of officer file evidence; (4) refusing to admit the “green report;” (5) rejecting plaintiffs’ “Second Table C” on the basis of its lack of military experience variables; and (6) dismissing Zukofsky’s individual salary discrimination claim without making specific findings. We have reviewed plaintiffs’ other claims of error and find them to be without merit. However, we vacate the district court’s dismissals of the individual discrimination claims in order that Rossini and Zukofsky on remand may obtain the benefit of any relevant evidence admitted pursuant to this opinion and of the presumption of discriminatory intent that will arise if the class discrimination claims succeed. Griffin v. Carlin, 755 F.2d at 1529. See Teamsters, 431 U.S. at 362, 97 S.Ct. at 1868; Duke v. University of Texas at El Paso, 729 F.2d 994, 997 (5th Cir.), cert. denied, — U.S. —, 105 S.Ct. 386, 83 L.Ed.2d 320 (1984); Sledge v. J.P. Stevens & Co., 585 F.2d 625, 637 (4th Cir.1978), cert. denied, 440 U.S. 981, 99 S.Ct. 1789, 60 L.Ed.2d 241 (1979). We are confident that on remand the trial judge will provide the parties with the opportunity for additional hearings, presentation and consideration of evidence necessary for a full and fair hearing of this case. Once all of the properly admissible evidence is in, it will be up to the district court *607 to apply settled standards in addressing the sensitive and difficult question of whether the plaintiffs were the victims of intentional discrimination on the basis of sex. Remanded for further proceedings consistent with this opinion. Costs on appeal shall be divided between appellants and appellee, with each side bearing fifty percent of the total expense. Footnotes [1] Defendant Ogilvy & Mather, Inc. was dissolved as a separate corporate entity in December 1980, at which time it became Ogilvy & Mather Advertising, a division of Ogilvy & Mather International, Inc. [2] Transcript references are to the five bound transcript volumes submitted by the parties on appeal. Their pagination differs from the trial transcript. [3] Although the basis for the initial denial is not clear on the record, see J.App. at 908, we presume that if officers had properly been included in the class, production of officer files would not have been denied. Thus, the "long and tortured” litigation over this issue, 597 F.Supp. at 1164, represents another aspect of the prejudice to the plaintiffs from the denial of representative status to Rossini and the concurrent exclusion of officers from the class. [4] Judge Gagliardi reaffirmed the order insofar as it applied to Rossini’s individual case "[i]n view of O & M’s strong showing of non-discriminatory reasons" for the promotion and assignment decisions challenged by Rossini, the number of officer files to which Rossini’s attorney had already been given access and the court’s speculation that statistical evidence derived from further disclosure of officer files would not further her case. 597 F.Supp. at 1152 n. 35. This reaffirmance and the other limitations placed on use of the officer file evidence were improper, as we explain infra. [5] Another reason given by the district court for downgrading plaintiffs' statistical studies in general and Second Table C in particular was that they relied in part on pre-statute of limitations data. 615 F.Supp. at 1522-24, 1545. On remand, the court should take note of Bazemore v. Friday, — U.S. —, 106 S.Ct. 3000, 3005-09, 3009 & n. 13, 92 L.Ed.2d 315 (1986) (Brennan, J., for a unanimous Court, concurring in part) (affirmative duty to eliminate pre-Title VII salary disparities based on race; evidence of pre-Act discrimination “quite probative" on issue of post-Act disparities).