In this appeal Southern Bell Telephone and Telegraph Company (Southern Bell) challenges the district court's conclusions, as embodied in an interlocutory injunctive decree, that it had violated Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e et seq., by engaging in discriminatory promotion and pay practices in respect of a broadly-defined class of female employees in its various facilities in North Carolina and in respect of four individual plaintiffs whom it permitted to act as representatives of the class. Southern Bell's chief contentions on appeal go to the court's determination under Fed.R.Civ.P. 23(c)(1) that the action might be maintained as a class action. A closely related contention is that considerations of res judicata and of comity precluded maintenance of the class action during the pendency of a government pattern and practice discrimination suit against American Telephone & Telegraph Co. (AT&T) and its associated Bell System companies, including Southern Bell, in the United States District Court for the Eastern District of Pennsylvania where consent decrees affecting class claims in the instant case had been entered.

Finding reversible error in the class action certification and in the determination of liability on the individual pay claims but no reversible error in the determination of liability on the individual promotion claims, we accordingly reverse in part and affirm in part.

* In January 1973 the Department of Justice, the Department of Labor and the Equal Employment Opportunity Commission (EEOC), proceeding pursuant to §§ 706(f)(1), 42 U.S.C. § 2000e-5(f)(1), and 707(e), 42 U.S.C. § 2000e-6(e), of Title VII, filed a complaint against AT&T and its associated Bell System companies, including Southern Bell, in the United States District Court for the Eastern District of Pennsylvania, charging discrimination against women and minorities in violation of Title VII. Shortly thereafter, a consent decree prepared by the parties was approved by the district court. This 1973 decree established goals, timetables, and priorities for the placement of women and minorities in certain jobs and provided for payment of monetary awards to certain employees. The decree included a model affirmative action plan that established fifteen hierarchical job classifications throughout the Bell System's management and nonmanagement ranks. The decree further provided for employee information programs, a promotion pay plan, a plan for assessment of job qualifications, and a provision for assessment of opportunities provided to female college graduate employees. By its terms this consent decree remained in effect for six years, until January 17, 1979, and the Pennsylvania district court retained jurisdiction over the provisions of the agreement in order to supervise and enforce its terms.

In May 1974 a second consent decree was entered by the Pennsylvania district court. It established a method for setting uniform entry salaries for the various levels of management and provided for monetary awards to certain employees. This judgment remained in effect until May 1979 with jurisdiction retained for supervision. Employees accepting monetary awards under either consent decree were required to execute releases of any claims of alleged violations of Title VII that occurred before the date of the consent judgment.

While the Pennsylvania proceeding was still pending under the two consent decrees, Stastny, a management employee of Southern Bell at its Charlotte facility, commenced this action as an individual plaintiff in January 1975, alleging discrimination against her in promotions, pay, and other conditions of employment. In April of 1975 she amended her complaint to allege class claims. In late 1975 and in 1976 the other three named plaintiffs in this action, Andrews, Springs and Rogers, commenced individual actions alleging claims similar to Stastny's. Each of these three additional plaintiffs was also employed at Southern Bell's Charlotte facility. All four named plaintiffs had filed timely charges with the EEOC, the earliest being Stastny's of December 18, 1970. The four cases were consolidated for discovery and trial, and on June 22, 1976, over Southern Bell's objection, the district court determined under Fed.R.Civ.P. 23(c)(1) that the action might be maintained as a class action with Stastny the representative of a class defined as including two subclasses:1

a. All females employed in North Carolina in management positions since August 20, 1970, who have been classified, restricted, discriminated against, or otherwise deprived of employment opportunities or status because of their sex, and

b. All females (including female craft employees of the defendant) who, since August 20, 1970, have been denied employment in management positions in North Carolina on account of their sex.

The action then proceeded to trial with this alignment of individual and class claims and claimants. At trial plaintiffs presented testimony of the named plaintiffs and the husband of one of them concerning their treatment as individual employees; testimony of the former general manager of Southern Bell in North Carolina concerning the company's employment policies and practices; and statistical evidence, including opinion evidence from a statistical expert. Plaintiff's statistical evidence was based upon data drawn from all of Southern Bell's North Carolina facilities, totalled in a way that reflected statewide effects of its challenged practices. This suggested that on a statewide basis women in management were concentrated in its lower levels; that although statewide the company filled 87% of its management vacancies by promotion from nonmanagement, the percentage of women promoted from the statewide nonmanagement ranks was less than their percentage in those ranks; that the percentage of women promoted within the management ranks was lower than their percentages in the ranks from which promotions were made; that annual appraisals of employees for advancement purposes were more likely to be made by men than by women because a greater percentage of supervisors on a statewide basis were male; and that within certain salary classes male average salaries exceeded to a significant degree the average salaries paid to women.

Southern Bell presented in rebuttal and by way of defense its own statistical evidence and the testimony of an expert, evidence concerning the Pennsylvania consent decrees, and testimony by a former commissioner of the EEOC and by employees of AT&T and Southern Bell. This evidence suggested that the statistical disparities revealed by plaintiffs' evidence were either statistically insignificant or resulted from the operation of factors other than the sex of the employee, for example, level of education.

With all the evidence in, the individual plaintiffs moved that Andrews, Springs and Rogers be added as representatives of the class as defined in the court's June 22, 1976 order. Southern Bell resisted this motion, and renewed its original objection to the district court's determination that the action could be maintained as a class action. The district court allowed the addition of these three plaintiffs as representatives and in the process tacitly rejected Southern Bell's renewed objection to the class action's maintenance.

Turning then to the merits, the district court made extensive findings of fact and concluded that a pattern or practice of discrimination against the defined class of women employees justifying class relief had been proved. The court also made extensive findings on the question of discrimination against the four named plaintiffs as individuals. On the basis of these findings, the court concluded that since September 19, 1970 (ninety days prior to the filing of Stastny's first charge with the EEOC), Southern Bell had discriminated against the defined class by denying women employees entry into management,2 promotion within management, equal pay within the same salary class, and cross-training and developmental job assignments; that the annual appraisal system operated in a discriminatory manner against women; and with respect to the four named plaintiffs, that each had been deprived of management opportunities because of her gender. The court also rejected defendant's argument that the action was barred by laches.

Based on these conclusions, the court entered a "partial judgment" that granted immediate injunctive relief to the defined class;3 adjudged the named plaintiffs and other members of the class entitled to remedial front and back pay awards to be individually determined by a master in referred proceedings; and made an interim award of costs and witness and attorney fees. Expressly determining that the Pennsylvania decrees were entitled to no generally preclusive effect in respect of the class claims, the court nevertheless avowedly shaped its judgment to avoid direct conflicts with and to give defendant "the benefit" of those decrees. Southern Bell was generally enjoined "to eliminate all employment practices" violative of Title VII, and specifically, to post notices of management vacancies and to appraise each woman management employee annually.4 The relevant date for computation of back pay awards was determined to be December 18, 1968, two years before the filing of Stastny's EEOC complaint. In its order of reference filed in conjunction with the partial judgment, the court specified the process for determining class member claims in respect of unequal pay and denied promotions. Proof that a claimant's pay was less than the average male salary in salary classes in which males averaged higher salaries than females on a statewide basis would constitute prima facie proof of entitlement to back pay for the salary difference. Proof that a claimant had been employed on or after September 19, 1970, and had since December 18, 1968, been denied a management vacancy that was filled by a male employee would constitute prima facie proof of entitlement to the position, hence to a remedial monetary award for its denial. As to vacancies occurring prior to January 18, 1973,5 the burden would be upon Southern Bell to "satisfy the master by clear and convincing evidence" that a claimant "was not qualified or less qualified than the male" who got the position in order to overcome the prima facie case. As to vacancies occurring after January 18, 1973, the claimant would have the burden (in addition to that for establishing the prima facie case?) "to show that it is more likely than not that she was more qualified than the person who filled the vacancy." Referring expressly to the standards for determining entitlement to remedial awards announced in White v. Carolina Paperboard Corp., 564 F.2d 1073 (4th Cir. 1977), the order of reference admonished that "the essence of the job is to try to figure out under the facts of each case on principles of equity and fairness where each claimant would be and would have been but for discrimination based on sex." In a further accommodation to the Pennsylvania decrees, the court directed that class members who had accepted monetary awards under those decrees were barred from further recovery for any violations occurring prior to the date of the releases they had signed, except that class members who had signed releases after the determination of class action status in this suit (presumably the original determination of August 11, 1975) were barred only if the release were shown to be knowingly executed.

The effect of all portions of the "partial judgment" and its accompanying order of reference were, by stipulation, stayed pending resolution of this appeal, which then followed.

As indicated, Southern Bell's attack on the propriety of allowing this action to be maintained as a class action has had two essential prongs throughout: first, an alleged preclusive effect of the Pennsylvania consent decrees deriving either from res judicata or policy-based comity concerns; second, the asserted failure of the plaintiffs to establish the justification for class action status under the technical requirements of Fed.R.Civ.P. 23(a) and (b).

The serious question raised by the first ground of attack the effect of the Pennsylvania decrees has apparently now been resolved against Southern Bell's contention by the Supreme Court's recent decision in General Telephone Company of the Northwest v. EEOC, --- U.S. ----, 100 S. Ct. 1698, 64 L. Ed. 2d 319 (1980).6 We need not address the point, however, because of our conclusion that in any event, the plaintiffs failed to establish under Rule 23(a) and (b) a sufficient basis for maintenance of the class action as certified by the district court. Because we consider that the promotion-based claims and the pay-based claims presented somewhat different problems for the maintenance of a class action, we discuss these separately and in the order given. We preface the separate discussions, however, with some general observations prompted by recent judicial developments that apply to both sets of class claims.

Whatever its ultimate implications for the maintenance of Title VII class actions, the Supreme Court's admonition in East Texas Motor Freight Systems, Inc. v. Rodriguez, 431 U.S. 395, 405, 97 S. Ct. 1891, 1898, 52 L. Ed. 2d 453 (1977), that "careful attention to the requirements of Fed.Rule Civ.Proc. 23 remains nonetheless indispensable (in Title VII actions)," was plainly intended to insure for these cases the same kind and degree of practical and conceptual analysis looking to class action allowance that the rule (hereafter Rule 23) prescribes without differentiating by subject matter for class actions in general. The most obvious implication, given the course of developments leading up to the admonition, was that the broad remedial purposes of Title VII and the undoubted utility and fitness of the class action device for many Title VII actions do not relieve the obligation imposed by subsections (a) and (b) of Rule 23 to inquire into the specific fitness, on its own facts, of each Title VII case for which class action status is sought.7 In the Title VII context, as any other, this inquiry must be carefully adapted to the controlling principles of substantive and procedural law that give content to and order proof of the particular claims and defenses asserted. Only so will the protections that these class action criteria are designed to provide for putative and actual class members on the one hand and the party opposing the class on the other be realized.

In Title VII cases this means adapting the closely related inquiries into commonality of issues, typicality of claims, and adequacy of representation8 to the specific factual claims as shaped by either or both of the two substantive theories of liability disparate treatment and disparate impact that have emerged in Title VII doctrine.9 Central to both theories of liability where class-wide sex (as other) discrimination is alleged is the existence of an identifiable employment pattern, practice or policy that demonstrably affects all members of a class in substantially, if not completely, comparable ways. In disparate treatment theory, this is the "pattern or practice," followed as an employer's "regular or standard operating procedure," which so demonstrably treats women in relatively unfavorable ways that it justifies a rebuttable inference that it proceeds from an intention to treat them differently simply because of their sex. See Teamsters v. United States, 431 U.S. 324, 336, 97 S. Ct. 1843, 1855, 52 L. Ed. 2d 396 (1977). In disparate impact theory, it is the practice or policy which, though demonstrably neutral or even benign in any sex-related intent, nevertheless bears with disproportionately adverse impact and without any justification of business necessity upon women. See Griggs v. Duke Power Co., 401 U.S. 424, 91 S. Ct. 849, 28 L. Ed. 2d 158 (1971).

It is in relation to the "pattern or practice" element of both substantive theories that the commonality criteria for class action maintenance become most critical in Title VII litigation. Indeed at this point the class action and merit inquiries essentially coincide. For to answer the procedural questions whether there is a sufficiently homogeneous class vis-ea-vis an identified practice to permit binding or benefitting it by class judgment; whether the issues presented in respect of the putative class members' claims have sufficient commonality to permit their resolution on an unindividualized basis; whether the individual claim of a representative plaintiff is sufficiently "typical" to permit its use as the prototype for resolution of the common claims of the class; and whether, given the nature of the representative's claim, adequate representation of class members' claims can be assured from the representative's efforts in effect requires answering the substantive question whether, under either of the available theories, there exists the requisite "pattern or practice" sufficiently and comparably affecting an identifiable class of protected employees.10

Intertwining of class action inquiry with merits inquiry is of course not confined to Title VII actions, but the degree of coincidence is perhaps unique in Title VII cases such as the instant one where there exist critical substantive issues of the very existence and scope of any practice having sufficiently "common" effects on an alleged class. A consequence where this degree of coincidence exists is that fair determination of the propriety of class action may be exceedingly difficult without conducting an inquiry roughly comparable to that required to resolve the "pattern or practice" issue on the merits.

Yet another feature of most Title VII class action litigation increases the difficulty of making a satisfactory determination of the class action question in the pre-trial stages. This results from the heavy reliance necessarily placed in such cases, including the instant one, upon statistical evidence, and from recent developments affecting the appropriate use of that evidence. Cases such as Hazelwood School District v. United States, 433 U.S. 299, 97 S. Ct. 2736, 53 L. Ed. 2d 768 (1977), and in this Circuit, EEOC v. Radiator Specialty Co., 610 F.2d 178 (4th Cir. 1979), and Roman v. ESB, Inc., 550 F.2d 1343 (4th Cir. 1976), have now emphasized the necessity that statistical proof designed to show disparities must be fairly adapted to the actual labor pools, work categories and like groupings of persons within which a discriminatory employment practice is alleged to have occurred. And cases such as Radiator Specialty have shown that the fitness of particular statistical data for the intended purpose may not be apparent in pre-trial stages; that there may be critical questions going precisely to the appropriate statistical data base labor pools for hiring and promotion practices, work categories for pay practices, etc. within which statistical disparities may fairly be assessed. As is obvious, the resolution of these "evidentiary" questions will bear importantly upon the class action determination. For under the substantive and procedural doctrine that has emerged, identification of the appropriate data base for statistical proof purposes geographical and qualification related labor pools, work categories, etc. may have to precede or at least coincide with determination whether and to what extent there exists a set of protected employees in whose behalf a class judgment may fairly be entered (and, incidentally, of those persons who may appropriately represent any class found proper). See Hill v. Western Electric Co., 596 F.2d 99 (4th Cir. 1979).

The resulting difficulty of making a fair determination of class action status in advance of trial on the merits in these cases produces a dilemma for trial courts that is well known and for which no happy general solution has yet been, or is likely to be, found. In paraphrase of Judge Godbold's wry summation, it arises from the composite command of rule and decision to "find out . . . early, (but not) too much." Huff v. N. D. Cass Co., 485 F.2d 710, 714 (5th Cir. 1973) (en banc). On the one hand for perfectly good and obvious reasons Rule 23(c)(1) admonishes that the class action determination shall be made "as soon as practicable." On the other hand for equally good and obvious reasons trial courts have been cautioned not to act too precipitately, on insufficient records, in making the determination. E. g., Belcher v. Bassett Furniture Industries, Inc., 588 F.2d 904, 905-06 (4th Cir. 1978). The solution of conducting a generalized preliminary inquiry a mini-trial into the merits of the class claims to determine the propriety of class action treatment has been expressly rejected. Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 177-78, 94 S. Ct. 2140, 2152, 40 L. Ed. 2d 732 (1974). On the other hand, a deliberate deferral of the determination until full trial on the merits will necessarily have resolved the coincident issues is fraught with serious problems of judicial economy, and of fairness to both sides.11

Rule 23 itself suggests the best means for handling the problem on a case by case basis. Immediately after directing class action determination "as soon as practicable," Rule 23(c)(1) provides that the determination "may be conditional, and may be altered or amended before the decision on the merits" presumably down to the very point of decision. The tentativeness that this implies for all class action determinations is simply highlighted for the reasons just noted in the Title VII action where the very existence and scope of an employment practice having a "common" reach is itself in dispute. See Harriss v. Pan American World Airways, Inc., 74 F.R.D. 24, 36-37 (N.D.Cal.1977). Ideally, by careful structuring of the inquiry, with pre-trial procedures focused precisely upon the commonality factors as they relate to the existence, the nature, and the common reach of the employment pattern or practice that must eventually be shown, it may well be possible to make the final class action determination before proceeding to trial on the merits.12 For reasons above explored, this may not, however, be possible despite a court's best efforts. It may then be proper to make a formal determination to proceed on express conditions that focus attention on specific class action issues that may be answered during the trial on the merits but before it has run its full course. Whether such a pre-trial determination is expressly made conditional or not, the court must be prepared under Rule 23(c)(1) to alter or amend it if the course of trial on the merits reveals the impropriety of class action maintenance. And this of course necessarily implies that a reviewing court must be prepared to find error in a trial court's failure to alter or amend to "decertify" the class action if on the whole record, including the evidence adduced on trial, it appears to the reviewing court that this was the proper course.13

Because we consider that these problems were peculiarly in play in the instant case, we have made these general observations as background for our conclusion now to be amplified that though its pre-trial certifications may have been justified, the district court erred14 in failing to withdraw certification of the class action at the conclusion of the trial on the merits, if not sooner. At that point and possibly earlier it was manifest that the representative plaintiff had not shown, as was her burden, that the action met the requirements of Rule 23(a) and (b) in respect of any identifiable class or subclass. In consequence, certification of the class action should then have been withdrawn and the class allegations stricken.

In the district court Southern Bell attacked maintenance of the promotion-related class claims15 when they were first added by amendment to Stastny's complaint and specifically renewed its attack before and at the conclusion of trial. Though put primarily in terms of Stastny's unfitness (and later that of the other named plaintiffs) to act as representatives because of conflicting interests in respect of some class members and the lack of sufficient typicality in the representatives' claims, the contention ran also to the lack of sufficient commonality of issues, hence to the inappropriateness of defining a statewide class of women employees. As indicated, we think that the correctness of this position on the composite "commonality criteria" was manifest at least upon the conclusion of the presentation of evidence and should have been acted upon whether in direct response to the attack or sua sponte.

With nothing before it at the outset but the original pleadings and motion papers, and even later in the pre-trial stage following discovery, the district court may have been completely justified in making its two preliminary determinations that the action might proceed as a class action in respect of the promotion-related claims. At those stages, for aught that appeared, it might be possible on the basis of the existing or later-amended class allegations, for the representative plaintiff to show under Rule 23(a) and (b) a sufficient nexus, see Wells v. Ramsay, Scarlett & Co., 506 F.2d 436 (5th Cir. 1975), between the statewide class as then tentatively defined and a statewide practice or pattern respecting the alleged violations. If this did not eventuate, the possibility existed, and may further have justified the preliminary certifications, that sufficient commonality might be shown in respect of subclasses that could then be defined to reflect the actual proof. By finding the class action appropriate under Rule 23(b)(2) at these stages, the court can be taken properly to have recorded its tentative determination that Southern Bell might be shown to have "acted on grounds generally applicable to the class," Rule 23(b)(2), as then defined, i. e., a statewide class. Within the class allegations then before the court, this might have been accomplished either by establishment of a disparate impact or a disparate treatment type of pattern or practice. See note 10 supra.

But the tentative quality with which these pre-trial class action determinations should have been viewed is plain from the record. Substantial questions of great relevance to any ultimate determination were then open and uncertain. As anticipated in our general observations, the uncertainty related essentially to the commonality criteria. Was there an identifiable pattern or practice affecting a definable class in common ways? If so, how wide was its sufficiently "common" reach? Judge Schwarzer has perceptively identified in greater detail the relevant questions as viewed from the pre-trial vantage point in similar litigation:

(i) What is the nature of the unlawful employment practice charged is it one that peculiarly affects only one or a few employees or is it genuinely one having a class-wide impact.

(ii) How uniform or diverse are the relevant employment practices of the employer, considering matters such as: size of the work force; number of plants and installations involved; extent of diversity of employment conditions, occupations and work activities; degree of geographic dispersion of the employees and of intra-company employee transfers and interchanges; degree of decentralization of administration and supervision as opposed to the degree of local autonomy.

(iii) How uniform or diverse is the membership of the class, in terms of the likelihood that the members' treatment will involve common questions.

(iv) What is the nature of the employer's management organization as it relates to the degree of centralization and uniformity of relevant employment and personnel policies and practices.

(v) What is the length of the time span covered by the allegations, as it relates to the degree of probability that similar conditions prevailed throughout the period.

Harriss v. Pan American World Airways, Inc., 74 F.R.D. at 41 (footnote omitted).

This, we think, is an accurate analysis of the specific questions that must in logic be considered in assessing the commonality criteria in Title VII cases. The significance and genuine disputability of different ones of these will of course vary from case to case. We do not mean to suggest that they should be adopted as a rigid rule-of-law checklist for all cases. But the relevance of the general lines of inquiry suggested is compelling. Such an inquiry requires development of an adequate record the sooner the better but, in any event, at some point before the certification question can properly be considered closed. The burden of producing such a record lies of course with the party who has the burden of establishing the propriety of class action maintenance the class representative.16 Failure to develop an adequate record for the purpose must have the normal litigation consequence for a party with the burden of presentation on a disputed matter resolution against his position.

Here, we conclude that the record finally laid before the court was deficient in critical ways that prevented any informed decision on certain of these crucial commonality questions. As a result, the commonality determinations implicit in the district court's certification could only have been made on the basis of speculation or larger leaps of inference than we think are permitted for Title VII cases under the teaching of Rodriguez. Indeed, for reasons to appear, we consider that the most supportable inferences from the record directly contradict the commonality determination as made.

The crucial deficiencies in the record that undercut the district court's certification of a statewide class trace essentially to an apparent failure to appreciate the significance of the dispersion of Southern Bell's work force hence of the putative class members throughout a great number of geographically separated facilities in North Carolina.17 The significance of this factor was anticipated in our general observations in Part II.A., and is of course well established in Title VII decisional law. See, e. g., Hill v. Western Electric Co., 596 F.2d at 101-02; Taylor v. Safeway Stores, Inc., 524 F.2d 263, 269-70 (10th Cir. 1975); Doninger v. Pacific Northwest Bell, Inc., 564 F.2d 1304, 1311 (9th Cir. 1977). Briefly put, it bears crucially upon the likelihood that there is truly an employment pattern or practice, particularly one of intentional disparate treatment, that sufficiently affects "in common" the class member employees in all the facilities. That there is such a pattern or practice may of course be sufficiently shown in a multi-facility setting. E. g., Patterson v. American Tobacco Co., 535 F.2d 257 (4th Cir. 1976) (two facilities). It might conceivably be shown by necessary inference from evidence of a sufficient number of instances of similar discriminatory treatment experienced by individuals throughout a multi-facility enterprise. It might conceivably be shown by statistical data showing comparable disparities in treatment experienced by protected employees at a sufficient number of the separate facilities to justify the inference that it reflects a system-wide policy or practice.18 Or, of course, it may be shown by direct evidence or by admission that there is in fact such a system-wide practice. E. g., Wetzel v. Liberty Mutual Insurance Co., 508 F.2d 239 (3d Cir. 1975) (clear nationwide application of benefit plan). Despite the plaintiffs' bald assertion in their brief that the challenged practices were shown to be statewide, a careful review of the evidentiary record reveals this to be unsupportable except by speculation in unwarranted assumptions.

The named plaintiffs' evidence in respect of the treatment experienced by each as individuals related solely to employment decisions made in and affecting persons in the Charlotte facility where all were employed. Though statewide statistical data was introduced to show statewide disparities in filling management positions, there was no showing of the extent to which, if at all, the overall disparities were paralleled in the separate facilities or even a statistically reliable sample of them. There was no showing by direct evidence that a statewide practice or policy, either objectively or subjectively administered, existed.19

Inattention to the significance of the multi-facility dispersion of the putative class left the record essentially silent on two critical factors for class action determination: the locus of autonomy in making the challenged employment decisions, see Harriss, 74 F.R.D. at 41, and intertwined with this, whether the management positions in question were filled from an essentially statewide work force labor pool; essentially from the work-force of each separate facility or from some intermediate geographical or group-of-facility pools. Compare, e. g., Hill v. Western Electric Co., 596 F.2d 99 (4th Cir. 1979), with Patterson v. American Tobacco Co., 535 F.2d 257 (4th Cir. 1976).

The evidence that bore at least indirectly upon the autonomy question seemed to point to a substantial degree of and perhaps almost complete local autonomy in the separate facilities. Indeed, the district court specifically found, in assessing the challenged appraisal system used by Southern Bell that "promotion and pay decisions are, and have been since at least 1965, entirely in the hands of one's superiors primarily one's immediate boss and are conferred upon employees based largely upon the boss's opinion of the subordinate's performance." While such evidence of subjectivity in employment decisions may well serve to bolster proof on the merits of individual claims of disparate treatment, and, as a "badge of discrimination," Brown v. Gaston County Dyeing Machine Co., 457 F.2d 1377, 1383 (4th Cir. 1972), to bolster statistical proof of class-wide discrimination in the very facility where the autonomy is exercised, it cuts against any inference for class action commonality purposes that local facility practices were imposed or enforced state-wide with respect to a statewide class.

Similarly, from the lack of any indication in the record that the pool of employees from which the management positions were filled was essentially a statewide one (i. e., that transfers between facilities to fill these positions was the ordinary, as distinguished from possible, practice), the opposite and more reasonable ordinary inference seems compelled, that there were in effect as many labor pools as there were separate and essentially autonomous facilities.

In the face of those two strong suggestions on the final record of local autonomy and facility-confined labor pools, the plaintiffs' evidence of disparities hence of the scope of any practice with "common reach" for class action purposes remained steadfastly dependent upon statewide statistical data unrelated to any separate facilities, including Charlotte itself, employment site of the named plaintiffs. Because the very number of separate facilities did not appear, the number of class members employed at each over time of course did not, and neither, obviously, did the number of vacancies at each over time. This made it impossible, since not even extrapolation from a sample of facility-related data was possible, to infer that such statewide disparities as were statistically shown were similarly reflected throughout the separate facilities, and from this to infer an effective pattern under centralized control that sufficiently affected all class members "in common." Rather, for all that appeared with the record fully made, such discrimination as may have been inferable from the statewide statistics could have been confined to relatively few of Southern Bell's unspecifically numerous, presumably autonomous, facilities located throughout North Carolina.20 By the same token, the absence of any critical data related to the separate facilities made it also impossible fairly to define any facility-related subclasses as to which common practices may have been inferable.21

For substantially the same reasons that class action treatment of the promotion-related claims was inappropriate,22 so was class action treatment of the unequal pay claims. Here too the statistical data was not broken out by separate facilities. Rather, it was designed to show that on a statewide basis women employees in management positions were paid less on the average than were men in certain broad salary categories in use system-wide. As indicated in Part II, the district court specifically found that individual pay decisions, like individual decisions on filling and promoting to management positions, were essentially subjective ones committed "primarily" to "one's immediate boss." This undisputed fact raises precisely the same normal inferences of facility-based autonomy in pay decisions that we found in respect of promotion-related decisions. And in the same way this makes impossible any fair inference of a statewide pattern or practice that would supply the commonality requirements for class action purposes in respect of the statewide class certified.

Hence, though it may sometimes be appropriate to certify for class action treatment some but not other employment discrimination claims in a single action, see Rule 23(c)(4)(A), this is not such a case. The district court erred in failing to decertify the class action in respect of the unequal pay claims as well as the promotion-related claims.

We find no reversible error in the district court's determination of liability in respect of the named plaintiffs' individual claims of discriminatory denials of promotions to management positions. While the evidence on these claims was in substantial conflict, the district court's careful findings and conclusions appropriately made under the disparate treatment test of McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S. Ct. 1817, 36 L. Ed. 2d 668 (1973) were not clearly erroneous in fact or erroneous in law. The named plaintiffs' detailed evidence that each was qualified for particular promotions that were denied her and given to men was adequate to support the district court's determination that a prima facie case of discrimination under the McDonnell Douglas test had been established in respect of each. Southern Bell's evidence designed to rebut the prima facie cases and to show with respect to each claim legitimate nondiscriminatory reasons for denying the promotions was not sufficiently overpowering to compel acceptance by the district court. The resulting determinations of liability for remedial back and front pay awards in respect of these individual claims are therefore affirmed.23

In support of their individual claims of discrimination in pay, the named plaintiffs relied upon comparisons between their respective salary histories and the salary histories of a group of males within the same salary "levels" or "classes" occupied by the plaintiffs. The record shows that these levels or classes are quite broad, including numbers of different positions requiring widely varying skills and backgrounds, and involving fairly wide ranges of compensation.24 The named plaintiffs' evidence did not attempt to show that within these classes the actual positions held by each during the periods of comparison required skill, effort and responsibility substantially equal to that required to perform the actual positions held by the males whose salaries were compared. Southern Bell contends that this constitutes a total failure of proof of the claims under applicable statutes and decisional law. We agree. 42 U.S.C. §§ 2000e-2(a), (h); Keyes v. Lenoir Rhyne College, 552 F.2d 579 (4th Cir. 1977) (comparisons with males holding common title of "professor" insufficient because not shown to involve substantially equal work); see also Orr v. Frank A. MacNeill & Son, Inc., 511 F.2d 166, 170-71 (5th Cir. 1975); Ammons v. Zia Co., 448 F.2d 117 (10th Cir. 1971).

In consequence, those portions of the district court judgment finding Southern Bell liable to the named plaintiffs on their individual claims of unequal pay are reversed.

Southern Bell has assigned numerous other errors which are either mooted by our disposition or which we find to be without sufficient merit to warrant discussion.

Upon remand the district court will enter a judgment in conformity with this opinion that dismisses the class claims for failure to establish the appropriateness of a class action; alters the judgment in favor of the named plaintiffs on their promotion-related individual claims, to reflect dismissal of the class claim; and dismisses on the merits the named plaintiffs' individual claims of unequal pay.

It may be appropriate in conjunction with dismissal of the class claims to direct the giving of appropriate notice of the dismissal to members of the putative class. Whether this is appropriate or advisable and the means to be used for the purpose are better left to the discretion of the district court which can better assess the likelihood that formal notice is needed to assure protection and determine the most appropriate mode for giving it.25

In view of this disposition, the district court's award of attorney's fees and costs to plaintiffs must be and is vacated. Upon remand, the district court should reassess the award in light of the disposition mandated.

REVERSED IN PART; AFFIRMED IN PART; AND REMANDED WITH DIRECTIONS.