This is an appeal by plaintiffs from a judgment denying on the merits their individual and class claims brought under Title VII of the Civil Rights Act of 1964, 42 U.S.C. Sec. 2000e, et seq. and Title I of the Civil Rights Act of 1866, 42 U.S.C. Sec. 1981, alleging sex and race discrimination in hiring, job assignment, and promotion practices of the defendant, Bloomsburg Mills, Inc. (Bloomsburg) in its Abbeville, South Carolina, textile mill facility. Appellants challenge the district court's interlocutory orders prohibiting communications by them or their counsel with unnamed members of the certified class or potential members of an expanded class; the court's several refusals to expand the class to include black males; and the court's determination on the merits that Bloomsburg had not discriminated against the individual plaintiffs or the certified class in any of the respects alleged.

We conclude that the district court did not abuse its discretion in declining to expand the class to include black males, but we hold that the court's orders prohibiting communications were erroneous in law; that its findings of no discrimination on the class and individual hiring claims were in critical respects clearly erroneous, requiring reversal and remand for entry of a remedial decree; and that the court's failure to make adequate findings in respect of the promotion and job assignment claims requires vacation of that portion of its judgment and remand for further proceedings conformable to Fed.R.Civ.P. 52.

* This action was commenced on April 3, 1973, by ten black women who on that date filed a complaint against Bloomsburg in the United States District Court for the District of South Carolina. Though all but one of the named plaintiffs was alleged only to have been denied employment by Bloomsburg, the complaint also alleged across-the-board race discrimination against blacks as a class in Bloomsburg's hiring, job assignment, and promotion practices at its Abbeville, South Carolina, textile plant.

Notwithstanding that all of the named plaintiffs were black women, the original complaint in terms alleged only race, not sex, discrimination.1 On April 19, 1973, before any class action certification order was entered, the original complaint was amended by the addition of a paragraph alleging in pertinent part that "defendant has further discriminated ... against plaintiffs and female members of the class they represent on the basis of their sex and race, in denying to them equal hiring opportunities to those accorded white female applicants." Almost a year later, on March 5, 1974, the district court, with consent of plaintiffs' then lead counsel, conditionally defined and certified the class as including only black female applicants and employees either denied employment or employment opportunities because of their race or sex.

Immediately following entry of this class certification order, plaintiffs retained new lead counsel who from that point on waged an unsuccessful effort in the district court to have the class redefined to include black males.2 This effort was complicated and heavily influenced by new lead counsel's early, court-aborted efforts to contact and obtain information from extant class members and potential class members, both female and male. Specifically, when new plaintiff's counsel, by an open letter dated December 4, 1974, announced their representation and sought to arrange a meeting with present and former black Bloomsburg employees to discuss the case, Bloomsburg immediately obtained an ex parte temporary restraining order prohibiting the announced meeting and barring plaintiffs and their counsel from initiating further contacts with black class members or non-class member employees or former employees of Bloomsburg. With some minor modifications, this order was converted, after hearing, into an interlocutory injunction that remained in effect throughout proceedings in the district court.3 Its practical effect was, as intended, to prevent all contact by plaintiffs or their counsel with any present or former black Bloomsburg employees except the named plaintiffs and, later, those members of the certified class who during a limited pre-trial claim-filing period filed notices of individual claims.

After protracted discovery and motion practice involving massive document filings and several abortive trial settings, the action finally went to bench trial on August 19, 1980. Evidence was received during ten days of trial and the case was submitted for decision on September 26, 1980. On December 30, 1982, the district court entered judgment dismissing the action on the basis of findings of fact and conclusions of law by which the court determined that appellants had failed to prove discrimination as alleged in respect of any of their individual or class claims.

This appeal followed.

We deal first with appellants' related contentions that the district court abused its discretion in declining to redefine the class to include black male discriminatees and in entering and then refusing to lift its injunctive ban on communication with members or potential members of the plaintiff class as defined or as it might be expanded to include black males.

First off, Gulf Oil Co. v. Bernard, 452 U.S. 89, 101 S. Ct. 2193, 68 L. Ed. 2d 693 (1981), decided while the case was under submission following trial in the district court, established what plaintiffs had unsuccessfully contended throughout: that the ban on communication was legally erroneous.4 That leaves the question, however, whether that error made similarly erroneous--as an abuse of discretion--the court's continued refusal to redefine the class to include black males.5

Appellants point to the difficulty the ban created for them in locating potential black male witnesses or in obtaining information from black males that might demonstrate that they had experienced all or aspects of the across-the-board discrimination alleged in behalf of the black female-only class certified. And appellants further point to the fact that the district court's refusal to redefine the class was based in substantial part upon the legally erroneous view that the attempts by plaintiffs' counsel to solicit black male employee participation in the action were legally improper. They contend that these in combination establish the taint of the legally erroneous communication ban upon the discretionary refusal to expand the class.

There is obvious merit in appellants' contention that the district court's legally erroneous view as reflected in its communication ban affected the class definition determination. But in the end we do not find in it a sufficient basis for reversing the district court's refusal to redefine and expand the class.

The determination of proper class composition is simply an element, though obviously an important one, of the more fundamental decision whether an action may proceed on any terms as a class action. As such, it, like the more fundamental decision, is committed to the informed, partially rule-guided discretion of the district courts. Those courts must, after all, manage this typically difficult type of litigation and our review of their class definition determinations must therefore be highly deferential. We may reverse those determinations for abuse only where manifest injustice would otherwise plainly result to affected persons and where the management prerogatives of the trial court would not be improperly subverted by our action.

This means, of course, that we may not reverse a class definition determination simply because we might disagree with it and believe from the record we review that we would have decided things differently. Our deference must take into account the trial court's rightful concern that parties seeking to represent classes in litigation should early and as precisely as circumstances permit make and be prepared to live with tactical decisions respecting the scope of the class claims they intend to advance and hence the scope of the classes allegedly affected. While trial court efficiency, management prerogatives, and convenience may not be allowed to subvert fundamental fairness in individual cases, the maintenance of procedural regularity and protection of the integrity of trial court processes over the long haul are also matters with which we must be concerned in exercising our appellate review function. See Jones v. Caddo Parish School Board, 735 F.2d 923, 938 (5th Cir.1984) (Higginbotham, J., concurring).

Guided by these principles, we cannot hold that the district court's refusal to expand the class to include black males was an abuse of discretion, though reasons clearly existed for taking the other course, and taking it surely would not either have abused discretion. In the pre-trial discovery stage of the proceedings, it became manifest that plaintiffs' evidence would, as it later did, demonstrate employment practices that arguably disadvantaged black males in the same way as black females, though not in comparable degree. And, as indicated, a principal reason given by the district court for refusing to redefine the class is now revealed to have been legally erroneous.

On the other hand, there were valid reasons for the court to refuse expansion of the class. Primarily there was the fact that almost four years after the filing of the underlying EEOC charge by the plaintiff Lewis, and almost a year after commencement of this action, plaintiffs' counsel, with no suggestion of coercion, nor of mistake as to controlling law, nor of lack of opportunity to consider tactical alternatives, formally agreed to a female-only class definition. There was the further fact that, as of that time, only black females had sought to advance either individual or class claims in the action; and that, so far as the record shows, no black male had then filed an EEOC charge against Bloomsburg or sought to be added as a named plaintiff in the action.6 Though there is evidence of some confusion on the court's part as to the significance of the plaintiffs' race-plus-sex pleading amendment,7 it is indisputable that the amendment, though most ambiguously, at least suggested problems of black male-black female conflicts of interest if the class were redefined. As plaintiffs sought later to demonstrate, those potential conflict problems were technically manageable by subclass definition or, as was unsuccessfully attempted, by formally abandoning any sex-discrimination claim. But the district court was nevertheless entitled to weigh in the balance the management difficulties that such belated changes in the earlier consented course of proceedings would entail.

Against these valid, though ultimately manageable, administrative and procedural reasons for not expanding the class we may properly consider whether they are outweighed by such manifest injustice that they cannot be allowed to prevail. As matters developed, we see no such resulting prejudice to the two potentially affected categories of persons.

First, we see no such legally cognizable injustice to the named plaintiffs or to the female-only class. Only if the court had used the absence of black males as class members to restrict proof of race discrimination practices solely to their effects upon black women might the named plaintiffs or members of the certified class have had any claim of possible prejudice. This might arguably have deprived them unfairly of proof of general race discrimination of which they were special victims. But the proof was not so limited and remains in the record for consideration in any further proceedings.

The possibility of significant injustice to putative black male class members is not more compelling.8 In the first place, no black male was legally prevented solely by his non-inclusion in this class from pursuing his own individual claim, or from becoming a member of another class.9 Concededly, failure of inclusion would indirectly bar any male who, because of his failure timely to file an individual EEOC charge, could only have survived as a viable Title VII claimant by inclusion in this or another class. See Inda v. United Air Lines, Inc., 565 F.2d 554, 559 (9th Cir.1977). But that of course is the common lot of all strangers to litigation who might be but are not included in particular class actions and whose individual claims are then, for whatever reasons, time-barred. Such a loss of potential rights of action could only be charged to a court as abuse of discretion (as opposed to being charged to the person's own inaction, neglect, or choice) if the outsider had reasonably relied to his detriment on an assumption created or encouraged by the court that he would be included in the class and need not therefore protect his interests by other means.

Here, the record belies any possibility of such reasonable reliance--even assuming that there may have been subjective expectations and demonstrable detriment of this kind to some black males. From the date of the class action's certification, the district court steadfastly refused every formal motion to redefine the class and never by any of its orders held out any prospect of its expansion. While, as a practical matter, we may be dubious about the extent to which outsiders to class litigation actually rely upon or even have knowledge of court proceedings affecting their putative status as parties, we must perforce ascribe to them a form of constructive notice in our administration of the class action device. See Washington v. Finlay, 664 F.2d 913, 929 (4th Cir.1981).

Because we conclude that there were valid administrative and procedural reasons for declining to expand the class as originally defined by consent of the parties and no countervailing fundamental injustice to any persons affected by the decision, we cannot find abuse of discretion in the district court's action.

Plaintiffs' principal individual and class claims concerned Bloomsburg's hiring practices at its Abbeville plant. The individual claims were of specific failures to hire because of plaintiffs' race and sex; the class claims were of a pattern or practice of disparate treatment or disparate impact adversely affecting members of the class denied employment.

In support of these claims, plaintiffs offered general background evidence concerning Bloomsburg's organization and operations in the Abbeville plant; specific testimonial evidence about Bloomsburg's hiring practices during the period in issue; extensive statistical evidence concerning Bloomsburg's hiring record and workforce composition; and the individual testimony of plaintiffs and members of the class concerning specific failures to hire.10

The general background evidence was largely stipulated, as was the accuracy of the base statistical data respecting hiring and workforce composition. Much of the testimonial evidence respecting general hiring practices and a large part of the expert opinion testimony as to the statistical significance of the base statistical data was essentially undisputed. We first summarize the most relevant background evidence which, for one reason or another, was not in issue.

At the critical times, Bloomsburg was a textile manufacturer headquartered in New York with plants in Bloomsburg, Pennsylvania and the one here in issue in Abbeville, South Carolina. Concentrating essentially on the production of small amounts of profitable, specialized weaves suitable for women's high fashion apparel, it made relatively short runs of styles and frequently ran several styles simultaneously. Its operations were therefore characterized by fairly sudden adjustments to a volatile end-market, with consequent adjustments in the size and composition of its workforce. It also experienced the usual ups and downs in production demands and capabilities stemming from a variety of causes that traditionally have affected the textile industry in general. These also affected its demands for labor over the period in issue. Its workforce was generally typical of the southern textile industry in terms of skill requirements--consisting essentially of unskilled or semi-skilled workers, some with prior textile experience when hired, some without.

The Abbeville plant was opened in 1963 and from that time through the time in issue employed on an erratically fluctuating basis between 250 and 425 workers. It was one of several textile mills in the county, being the second largest employer among these. Its labor supply was essentially drawn from the local workforce population, and from the blue-collar segment of that population rather than from the white-collar professional, clerical segment. When the plant was opened in 1963, a deliberate effort was made in its initial hiring to employ workers with prior textile mill experience. This effort was essentially successful, but over the ensuing years, for a number of reasons which are to some extent disputed, the proportion of those initially hired who had previous textile experience declined, as of course did the proportion of those in the overall workforce. The workforce was characterized over the period in issue by a significant turnover rate and by considerable rehiring of former employees.

During the period in issue the Abbeville plant was divided into several departments under departmental heads who were responsible for personnel decisions, including hirings, in their respective departments. A plantwide personnel manager--of whom there were four in succession during the period in issue--was in general supervision and control of personnel matters. From September 1963 when the plant opened, until September 1971 the personnel manager was James Hemminger. For reasons that will appear, it was upon the hiring practices during the period of Hemminger's tenure that plaintiffs ultimately based their claim of class-wide discrimination in hiring.

During that period Bloomsburg had no formally adopted written hiring standards to guide its personnel manager, department heads, or others in making hiring decisions. When the plant opened in 1963, a general directive was given by management that so far as possible the most experienced and skilled applicants should be hired, in order to avoid the need for on-the-job training during start-up time. Presumably in consequence, initial hires were preponderantly experienced in textile work. Among the first year's hires only 11, or 6.3 percent, were black, and none of these was a black woman.

Once the plant was in operation, hiring of replacement or additional workers was done on a loosely structured basis, still with no written standards, throughout Hemminger's tenure. In practice, hiring was largely controlled by an initial screening process left for the most part to the receptionist in the plant's on-site office or to the "girls happening to be at the window" when applicants presented themselves. At that point the receptionist, or whoever was on office duty, had the applicant complete an application which was then placed on file unless an appropriate vacancy happened then to exist. In the latter situation, the applicant might be directly referred to a department head in whose department the vacancy existed. Otherwise the application was placed on file and remained "alive" in that file for a fixed period. Department heads advised the office when they had vacancies, and it was on this basis that the receptionist processed applications. When a vacancy developed, the receptionist went to the applicant files and from among the live applications picked one to refer to the department head, or occasionally directly referred an applicant who happened to be in the office at the time. Though department heads were of course under no obligation to take the applicant referred, in practice they routinely did. It was not the practice to check back with the receptionist to see if there were other, possibly more qualified, live applications then on file. Nor did the receptionist operate under any specific standards related either to relative times of filing or to relative qualifications in choosing between applicants in live files.

Building upon this stipulated or essentially undisputed evidence of Bloomsburg's general organization and mode of operation in making hiring decisions, plaintiffs offered more specific evidence to establish that throughout the period of Hemminger's tenure, from 1963 to 1971, Bloomsburg's hiring practices either intentionally or by disparate impact discriminated against black females who applied for employment.11

In presenting their statistical evidence, plaintiffs relied essentially upon the familiar procedure of comparing for the relevant period the "observed" annual rate of hires of black females with the "expected" rates based upon the proportional availability of black females in the relevant labor pool, then having the results of the comparison assessed for statistical significance through expert opinion testimony. Because Bloomsburg had improperly disposed of employment applications for most of the relevant period, cf. EEOC v. American National Bank, 652 F.2d 1176, 1195 (4th Cir.1981), applicant flow data was not available to supplement or test the validity of the plaintiff's raw hiring rate statistical evidence. The available pool of black female applicants therefore had to be established for statistical comparison purposes by other means. Census data showing the percentage of black females in the relevant labor pool at critical intervals and state Employment Security Commission data were used for the purpose.

In view of the district court's treatment of the statistical evidence and of our resolution of this appeal, it is not necessary for our purposes to summarize or analyze that evidence in all its voluminous detail. For reasons that will appear, it suffices here that the evidence convincingly demonstrated that throughout the period of Hemminger's tenure the hiring rate of black women was far below that to be expected from their proportion in the relevant labor pool.12 Among the salient items of this statistical evidence was data and testimony demonstrating that through this period the hiring rate of black women was in the range of 5-8 standard deviations below the "expected" level.13 Unchallenged expert opinion testimony established that such a disparity was statistically significant in the sense that the probability that it might be explained by chance alone was no better than one in a thousand. This clearly satisfied the standards we have imposed for finding from statistical evidence of this sort, whether or not supported by other non-statistical evidence, a pattern or practice of race or other illegal employment discrimination. See, e.g., EEOC v. Federal Reserve Bank of Richmond, 698 F.2d 633, 647-48 (4th Cir.1983), rev'd on other grounds, sub nom. Cooper v. Federal Reserve Bank of Richmond, --- U.S. ----, 104 S. Ct. 2794, 81 L. Ed. 2d 718 (1984); American National Bank, 652 F.2d at 1192. See also Moultrie v. Martin, 690 F.2d 1078, 1082-83 (4th Cir.1982) (acceptable statistical proof of discrimination in jury selection).

Here, confronted with the overpowering weight of this statistical evidence, the district court did conclude that it established a prima facie case of disparate impact against the class.14 That finding is not seriously challenged on appeal, as indeed it could not be on the evidence and under our precedents.15

Instead, issue is actually joined here only on the district court's follow-up determination that Bloomsburg's evidence established that the reason for the demonstrated disparate impact was Bloomsburg's preference for experienced rather than inexperienced labor, and that this experience requirement was a legitimate business necessity which, under Griggs v. Duke Power Co., 401 U.S. 424, 432, 91 S. Ct. 849, 854, 28 L. Ed. 2d 158 (1971), relieved Bloomsburg from liability under Title VII.

Appellants challenge that dispositive determination and we agree with them that it is legally erroneous and cannot stand as the basis for the judgment in Bloomsburg's favor. We hold instead that the district court's business necessity determination--an ultimate finding of fact--is clearly erroneous on the record we review.

Our reasons are several. They include obvious misapprehensions on the district court's part as to controlling legal principles respecting the "business-necessity" defense to a class disparate impact claim. And they include failure or refusal by the district court to consider essentially unchallenged documentary evidence of Bloomsburg's actual hiring practices which completely undercut any rational basis for a finding that an experience requirement had anything significantly to do with the disparate hiring rate of black women.

We look first to the court's inexplicable failure or refusal to consider critical evidence negating the defense. Bloomsburg's evidence offered to establish the existence and impact of an experience requirement in its hiring practices was exclusively testimonial and essentially anecdotal. No broad-based documentary or statistical evidence was offered to substantiate the general assertions that such a requirement explained and justified the demonstrated disparity in hiring rates during the period of Hemminger's tenure. The testimony that was offered in support was of the vaguest sort, consisting essentially of general assertions by management witnesses that such a requirement had been orally communicated to persons engaged in and responsible for hiring. But the hard, undisputed evidence as to how hiring was actually done revealed the unsystematic process earlier noted in which any experience requirement that existed was primarily committed for implementation, with no formal written (or oral) standards, to the low-level clerical employees in the plant office who received, filed, and then chose between applications for employment. The evidence was unchallenged that at least throughout the period of Hemminger's tenure, these people essentially called the shots on which applicants were hired; that departmental heads almost invariably relied upon their selections in filling vacancies in their departments, conducting no independent studies of the applicant files to monitor the receptionist's or other clerical employee's determinations of relative experience or other qualifications; and that neither did anyone in higher level management exercise any effective monitoring of the process to insure that any experience requirement or other qualification was being effectively administered. The results of this manifestly subjective hiring process were, as the district court found, a significantly disproportionate number of black women hires during the critical period, hence a prima facie case of disparate impact.

In more specific rebuttal of Bloomsburg's effort to establish a business-necessity experience requirement defense, appellants offered evidence, essentially documentary and statistical, which, if accepted, wholly undercut that theory of defense. We merely summarize its salient, essentially uncontested, elements as they properly have been emphasized by appellants on this appeal.

The overall picture presented by this evidence was of an initial hiring pattern in which prior experience was arguably a significant, dominant factor. In the plant's first year of operation, 1963, about half of the white workers hired had previous experience in the job for which they were hired or had comparably useful experience. Within a year's time however, only about 30% of white men and 15% of white women hired (or rehired) had such experience, with 70% and 80% respectively having none. This was the typical experience thereafter during the period of Hemminger's tenure. From 1967 to 1971, about 85% of white men and 80% of white women hired had no such experience.

Presumably unable to challenge the basic accuracy of this rebuttal evidence indicating the meager or non-existent role of experience in actual hiring practices, Bloomsburg sought to avoid its effect by explaining it away. The proffered testimonial explanation was that as the general employment situation improved over this period, skilled workers in general were attracted to better paying jobs than those available in textiles and that this forced Bloomsburg out of its preference for experienced workers.

Notwithstanding that this counter-theory effectively conceded that the experience preference did not in fact cause the proven disparate impact (indeed that it was a non-factor in the challenged hiring practices), the district court found the experience preference a sufficient explanation of and defense to the disparate impact that it had found established.16

Because of the great deference we owe to the trial judge as trier-of-fact, see Anderson v. City of Bessemer City, --- U.S. ----, 105 S. Ct. 1504, 84 L. Ed. 2d 518 (1985), we have carefully canvassed the record to see if there is a plausible basis in the evidence for this seemingly illogical finding. Our conclusion, regrettably reached, is that such a finding could only have been made because of a failure to appreciate or a refusal to consider appellants' basically uncontradicted documentary evidence or, perhaps more importantly, to consider Bloomsburg's ultimate concession by its countering explanation that the experience preference was simply not a significant factor in determining hiring rates during the critical period. This is confirmed--though negatively--by the court's failure even to address appellants' evidence on the point in its findings of fact.

Our conviction that a fundamental mistake has been made in this critical aspect of the court's fact-finding process is increased by the court's obvious confusion about the relationship between the disparate treatment and disparate impact theories of claim and defense in Title VII class actions. From the district court's discussion of its application of the business necessity defense, it is plain that the court confused that affirmative defense and its persuasion burden with the much less burdensome defensive riposte in the McDonnell Douglas17 presumption-based proof scheme applicable to individual disparate treatment cases, and that it erroneously applied the latter in finding the prima facie case of disparate impact defeated.18

Of course, we recognize that disparate treatment and disparate impact discrimination theories and proof schemes are not to be applied in all Title VII cases, individual and class alike, with wooden inflexibility and in unvarying accordance with the details of their original formulations, nor in mutually exclusive fashion. See, e.g., Wright v. Olin Corp., 697 F.2d 1172, 1184-85 (4th Cir.1982). But, as we also have pointed out, their basic elements, though adaptable in detail, are not in the final analysis simply interchangeable; at bottom they reflect critical substantive differences between the two theories. See id. at 1185-86. Without engaging here in detailed conceptual analysis to demonstrate this, see, id., it suffices to note that finding a prima facie case of class disparate impact defeated by the mere production or "articulation" of evidence that might be thought sufficient to rebut the McDonnell Douglas presumption of individual disparate treatment completely skews the theories and involves a fundamental misapprehension and misapplication of controlling legal principle.19 This legal misapprehension, considered in conjunction with the court's closely related failure or refusal to take into account the uncontested evidence and Bloomsburg's implicit concession that an experience requirement was not even a significant factor--not to say a business necessity--in Bloomsburg's hiring during the critical charge period convinces us that a fundamental mistake has been made in the district court's fact-finding on this critical issue. See United States v. United States Gypsum Co., 333 U.S. 364, 395, 68 S. Ct. 525, 542, 92 L. Ed. 746 (1948); Miller v. Mercy Hospital, Inc., 720 F.2d 356 (4th Cir.1983). The resulting finding that an experience requirement effectively explained and justified as business necessity the disparate impact earlier found by the court has no plausible support in the record, cf. Anderson, --- U.S. at ----, 105 S. Ct. at 1512, and in consequence is clearly erroneous. Fed.R.Civ.P. 52(a).

With this clearly erroneous finding set aside, the district court's unchallenged finding that Bloomsburg's hiring practices had a discriminatorily disparate impact upon members of the plaintiff class during the period from March 18, 1969 to September 15, 1971,20 stands dispositive in favor of the class claim for that period of time. The judgment dismissing the class claim of discrimination in hiring must therefore, to that extent, be reversed and the case remanded for further proceedings.

Upon remand, the district court is instructed to enter an appropriate remedial decree in favor of the class, and to direct such further proceedings as may be required to effectuate it.21 See American National Bank, 652 F.2d at 1201 (comparable remand upon reversal); see also Sledge v. J.P. Stevens & Co., 585 F.2d 625, 643, 652-53 (4th Cir.1978).

Because any named plaintiff denied employment during the indicated period was entitled to benefit of the presumption arising from proof of disparate impact upon the class, see American National Bank, 652 F.2d at 1201, the district court's denial of their individual claims without benefit of the presumption must also be reversed, and we do so. Dismissal of individual claims of discrimination outside the indicated period are affirmed as based on findings not clearly erroneous.

In any stage II monetary relief proceedings ordered on remand any class members making individual claims of hiring discrimination during the indicated period will be entitled to benefit of the presumption and to be free of any preclusive effects from prior rulings respecting their testimony. See id.

Appellants' other substantive claim was of discrimination in job assignments which allegedly resulted in class members' systematic relegation to lower paying jobs in Bloomsburg's workforce.22 To prove this claim they relied again principally upon statistical evidence. Because we find it necessary to remand for proper factfinding in respect of this claim, we summarize only so much of that evidence as is required to demonstrate the inadequacy of the court's treatment of this claim and to point up the need for adequate first instance factfindings. In the process, we intend no expression of opinion upon the ultimate merits of the claim nor of the findings that properly should be made from the evidence.

Plaintiffs' statistical case was built essentially upon two sets of data reflecting year-end pay rates--one constructed by plaintiffs, from which average annual pay rates were calculated; the other, supplied by Bloomsburg, showing actual pay rates for the last payroll period of each year. Each set was stipulated to be comprehensive and statistically representative. The plaintiffs' data covered the years 1963 through 1975; Bloomsburg's, the years 1965 through 1979. Both applied only to non-salary wage rates.

The plaintiffs' basic plan of proof was first to demonstrate general disparities in pay rates, traceable to job assignments, between blacks and whites and, more particularly, between black women and white males over the period of time covered by the data. This was followed by statistical data designed to show that the pattern of job assignments, hence pay disparities revealed, were not explicable by those factors, considered singly or in combination, that most commonly explain job assignments, hence wage differentials on grounds unrelated to race or gender: seniority, education, and experience. To supplement and explain the raw data, plaintiffs relied upon the testimony of Dr. Charles R. Mann, a witness duly qualified as an expert statistician in the relevant field.

It suffices for our purposes here to say that the plaintiffs' statistical evidence, if accepted as essentially accurate and given the probative force urged for it, might have supported an ultimate finding of a pattern or practice of disparate treatment of, or of disparate impact upon, the class. In rough summary, and in its most salient respects, the evidence indicated that for the years 1966 through 1979 the hourly rates of pay for white male employees was substantially greater than that for black women workers, and that the revealed differentials were statistically significant. It further revealed that in raw mathematical terms these differentials existed from the times of initial hires throughout the coincident periods of employment of white male and black female employees hired at approximately the same times. It further demonstrated a facially disproportionate concentration of blacks in lower-paying jobs and of whites in higher-paying jobs, from the times of initial hires on through coincident periods of employment.

In order to anticipate the obvious defensive response to this data and its raw statistical significance, plaintiffs offered further statistical evidence in the form of exhibits and expert testimony designed to eliminate as explanations of the revealed differentials the most obvious non-discriminatory factors of experience, seniority, and education. Again, in rough appraisal of that evidence it can be said that if accepted for accuracy and given the probative force urged for it, it might well have supported the finding it was designed to produce.

Using workforce data that actually traced individual employment histories of Bloomsburg's entire workforce through the period covered by the payroll data basis, Dr. Mann performed and explained to the court a series of multiple regression analyses that, per his testimony, effectively ruled out the indicated non-race and non-gender factors as statistically acceptable explanations of the revealed pattern of job assignments and resulting pay differentials. For example, Dr. Mann testified that his multiple regression analysis for the year 1969 revealed that the estimated effect of race on pay rate for the black women employees, taking into account the factors of experience, education, and seniority, was to reduce their wage by 41.3 cents an hour. Comparable effects were said to be revealed for the other three years studied--1972, 1975 and 1978. Supplementing this estimate of the effect of race upon wage rates in general as deduced from the multiple regression analyses, Dr. Mann further testified that the statistical data revealed greatly disproportionate concentrations of comparably experienced white males and black females in highest-paying and lowest-paying jobs respectively. For example, in 1973, 26% of white males but only 5% of black women with no prior work experience or with no prior textile experience were revealed to have been assigned to Bloomsburg's highest-paying jobs. Comparable examples for other years and involving other combinations of comparable qualifications were said by Mann to be revealed by the data.

To counter this statistical case, Bloomsburg relied essentially upon challenges to the legal as well as statistical significance of the revealed pay differentials and job-assignment concentrations. The essential accuracy of the raw payroll data and of the individual employment histories of Bloomsburg's employees from which the wage differentials and job assignment concentrations were drawn was not challenged, as presumably it could not have been. The probative force of Dr. Mann's multiple regression analyses and of his opinions based upon them were only sought to be challenged by cross-examination and by the expert testimony of Bloomsburg's qualified expert witness, a Dr. Trent. The thrust of the challenge by both means was upon the capability of Mann's statistical analysis to demonstrate that the evident wage differentials and disproportionate job assignments were race and gender-related rather than being attributable to non-discriminatory factors.

We have carefully studied both experts' testimony on this critical point and are persuaded that the following is a fair summary of the gist of each. Mann frankly conceded, as of course common sense required, that no statistical analysis of this type can ever rule out every conceivable non-discriminatory factor as explanation for a demonstrated disparate effect upon a protected minority group, or establish beyond question that only discriminatory factors are its cause. Trent ultimately testified to no more than that Mann's analysis was subject to this conceded limitation.

In argument to us, and presumably to the district court, Bloomsburg contended, however, that Mann's multiple regression analyses were simply unworthy of any acceptance for the purpose offered because of this limitation. But Bloomsburg offered no countering evidence tending to suggest that a particular factor not worked into Mann's regression analyses would indeed provide a more substantial non-discriminatory explanation or at least by that specific means draw Mann's analysis in direct question. Indeed, the basic accuracy of Mann's methodology was not seriously challenged by Bloomsburg's expert, whose testimony was essentially aimed only at pointing up the methodology's ultimate limitation and in questioning some of its incidental aspects.23

As indicated, this is but a general summary of the more salient aspects of a truly massive evidentiary record dealing with the claim of job assignment, hence wage, discrimination. The district court disposed of the claim in what we can only characterize, regrettably, as a cursory, unsystematic and, in several respects, logically implausible, way. Because we conclude that the court's disposition of this claim failed even substantially to comply with the requirements of Fed.R.Civ.P. 52, we summarize the court's treatment of the claim in order to point up this critical deficiency.

In the first place, despite the obvious fact that plaintiffs claimed and their evidence tended to demonstrate that the wage differentials resulted both from original job assignments and failures to promote, the court dealt specifically in its formal findings of fact and conclusions of law only with Bloomsburg's promotion practices and their impact upon members of the class. Taking this unjustifiably narrow view of the nature of the claim and of the evidence offered to support it, the court essentially rejected it on the basis of a finding that members of the class did not expect to be promoted, hence made no effort to be. The court's only mention of the wage-differential claim in its concluding summary of Findings and Conclusions is this: "3) defendant did not discriminate against black females, individually, or against the class in granting promotions on the basis of race [or] 4) ... of sex."

These "findings" and "conclusions" simply fail to address in terms the broader and more fundamental claim--in which "promotion" practices as such were merely an element--of general discrimination against black women in respect of the wages paid them as a class by virtue of the jobs they were given to do. The broader claim of wage-job assignment discrimination was only addressed obliquely in the course of the court's outright rejection (presumably for all purposes and in respect of all claims) of Dr. Mann's credibility as an expert witness. This was laid by the court to the two basic factors. First, to the conceded inability of any statistical analysis in employment discrimination cases to demonstrate beyond question that race or gender-discrimination has caused any demonstrated disparity in treatment of protected groups. But cf. Dothard v. Rawlinson, 433 U.S. 321, 331, 97 S. Ct. 2720, 2727, 53 L. Ed. 2d 786 (1977) (employer not entitled to rest on speculative deficiencies in employees' facial showing of discriminatory impact); Sledge v. J.P. Stevens & Co., 585 F.2d 625, 634-35 (4th Cir.1978) (same); Grant v. Bethlehem Steel Corp., 635 F.2d 1007, 1015 (2d Cir.1980) (statistical proof not challengeable by speculative need for "fine tuning"). Second, upon the fact that while Mann conceded that he had never been to Bloomsburg's Abbeville plant or to any textile mill and knew nothing about them, by contrast, Bloomsburg's expert witness, Trent, had visited Bloomsburg's plant and while there had "checked and verified some of the statistics."24 On these grounds, the court simply dismissed the plaintiffs' entire statistical case: "This Court finds as a fact that Trent's testimony is credible and Mann's testimony lacks credibility, and the court cannot rely on what Mann said as a basis for deciding any issue in this litigation."

In requiring that a bench trial court "shall find the facts specially and state separately its conclusions of law thereon," Fed.R.Civ.P. 52(a) does not attempt to specify the breadth or depth in which findings shall be made, nor their exact form or style, nor their detail--whether they shall include "subsidiary," or "basic," or "historical," or "ultimate," or "mixed" fact findings--nor any other specific quality. Appellate courts generally, and wisely, have taken a flexible view about all these matters, eschewing hypertechnicality in assessing the sufficiency of particular findings made under this wisely non-specific rule directive. See generally 9 C. Wright & A. Miller, Federal Practice & Procedure: Civil Sec. 2579. Nevertheless, "specially" means something in the rule's context. Basically it means sufficiently wide in scope of coverage, sufficiently deep in factual detail, and sufficiently clear in style and mode of expression to reveal to the legally-trained observer (most importantly, reviewing courts) not only the ultimate factual premises for the trial court's legal conclusions, but the basic factfinding processes by which, starting from the evidence laid before it, the court arrived at the ultimate factual premises. See, e.g., Schneiderman v. United States, 320 U.S. 118, 129-31, 63 S. Ct. 1333, 1338-40, 87 L. Ed. 1796 (1943); Knapp v. Imperial Oil & Gas Products Co., 130 F.2d 1, 3 (4th Cir.1942).

The system owes this much revelation of its processes to litigants and the public. Only so can the system discharge its assumed responsibility to those entities to make the processes of justice as well as its end product as visible as possible. In turn the system must demand as much of its trial bench in order to insure, so far as appellate review makes this possible, that the system's factfinding processes are fair and rational.

In the nature of things, whether particular trial court findings are sufficient for these purposes must be left to the judgment of the appellate courts required to review them under the "clearly erroneous" standard also imposed by Fed.R.Civ.P. 52(a). Confronted with findings whose sufficiency for these purposes is questionable, appellate courts have had essentially two recourses. They might swallow hard and reverse or affirm on the basis that though mediate factfinding or factfinding processes are missing or obscure, they can be safely implied and that, as implied, they did or did not result in clearly erroneous ultimate findings. Or, they might, whether in utter despair or simple prudence, forego any effort to review and remand to have it done right in the first place.

In following the first course, appellate courts have undoubtedly fudged at times in implying findings where none are recorded or, indeed, where none may actually have been consciously made. To take that course is an understandable appellate impulse. Not only may it be thought justified to avoid a foolish insistence upon form when substance seems apparent ("in order to find this, the trial court must necessarily first have found that"), but it may sometimes be thought justified to avoid wasteful, time-consuming remand when the result (formally perfected findings leading to the same end) seems preordained.

Whatever its source and however worthy its purpose, however, indulging the impulse runs the risk of making first instance findings when none in fact may have been made by the district court, or occasionally of "implying" a finding that is in fact at odds with that actually "made" but not recorded by the trial court.

We run through all this in order to say that any impulse we may have had here either to affirm or reverse in review of factfindings so meager in scope as those just summarized seems to us now practically foreclosed as an acceptable procedure by the Supreme Court's intervening decision in Anderson, --- U.S. ----, 105 S. Ct. 1504. That important decision helpfully clarifying the appellate review function under Rule 52(a) did not deal directly with the sufficiency of trial court findings to permit review in the first place. But we think that its critical emphasis upon role-allocation as the essential underpinning of the Rule, see Anderson, --- U.S. at ----, 105 S. Ct. at 1512, strongly suggests that the proper recourse for courts of appeals confronted with district court findings of questionable sufficiency is ordinarily to remand for proper first instance factfinding.

To attempt review in such circumstances creates the dual risks, clearly of concern to the Anderson Court, of engaging in first-instance appellate factfinding under the guise of filling gaps in trial court findings and, in that very process, of engaging in an exhaustive, resource-wasteful record review that would not be required if the district court adequately performs its specially allocated factfinding function in the first place. The generally heightened degree of deference that Anderson commands courts of appeals to give to all aspects and modes of district court factfinding implies a commensurately heightened degree of responsibility in district courts to make and record findings of fact that facially demonstrate a full consideration of the factual issues necessary to decision. And this, in turn, implies a heightened responsibility on the part of courts of appeals to insist, even at the risk of delay, on having the factfinding process carried out properly at the level intended rather than to assume, even indirectly, a factfinding role.

Here, were we to attempt review, we could not carry it through as contemplated by the Rule's basic allocation of functions. In order to affirm, we would have to declare the district court's dispositive findings a plausible account of the evidence, see Anderson, --- U.S. at ----, 105 S. Ct. at 1512, and we cannot do that.

As indicated, the court's sparse findings essentially finesse the critical thrust of the plaintiffs' statistical evidence of substantial wage disparities traceable to job assignments after seniority, education levels and experience have been taken into account. Furthermore we simply cannot accept as plausible the court's stated basis for wholesale rejection of the critical testimony of plaintiffs' expert statistical witness.

In pointing this out, we do not suggest that there may not be a plausible basis for rejecting outright the whole of this statistical evidence or, at least, for finding it not sufficiently probative of the claim. There may be flaws in the statistical methodology other than the purely peripheral concerns noted obliquely in the court's memorandum of decision. Plaintiffs' expert may somehow have revealed wherein his testimony, largely devoted to explaining the statistical significance of unchallenged raw data, was wholly unworthy of belief or, at least, did not prove the claim of discrimination by the sufficient preponderance of evidence. But that is not revealed in the findings we are given, nor is it plausibly accounted for in the one specific reason cited by the court: that the expert was unfamiliar with textile operations and had not visited Bloomsburg's plant. Only by making our own first instance assessments of the statistical and legal significance of the massive quantity of statistical evidence in the trial record might we construct on our own a plausible basis for rejecting that evidence as not probative of the ultimate fact in issue.

Neither could we with confidence reverse on the basis that the ultimate finding of "no discrimination in promotion practices" is clearly erroneous. We can only say with confidence that it is an incomplete basis for disposing of the job-assignment/wage differential claim. Nor could we confidently say without engaging in first instance assessment of the statistical evidence that its total and cursory rejection by the court was clearly erroneous. We can only say that the court's stated basis for its rejection--essentially couched in terms of witness credibility rather than intrinsic or methodological untrustworthiness of the statistical evidence presented through that witness--are too meager to permit rational appellate review as opposed to second-guessing de novo review.

Therefore, as indicated, we must remand for reconsideration and proper disposition of the entire claim of discrimination in job assignments, hence in wages, against the class. We leave to the district court upon remand whether it may do this upon the record as presently made, or whether the record should be reopened for receipt of additional evidence; whether it will invite or permit additional written and oral submissions of counsel; and, as indicated, whether it will attempt in any way to control communications incident to claim filing or other proceedings between counsel, parties, and class members in accordance with Bernard. We only instruct that the court's findings of fact and conclusions of law on the present or a reopened record should meet the requirements outlined in this opinion in order to serve the purposes indicated.25V

In summary, we find no abuse of discretion in the district court's refusal to expand the class originally certified; we hold that the court erred in imposing its communication ban, but that whether one should be imposed upon remand should be left to the district court, guided by Bernard; we reverse the court's dismissal of the class and individual hiring discrimination claims and remand for entry of a remedial decree and further proceedings in accordance with this opinion; and we vacate that portion of the judgment dismissing the job assignment/wage discrimination claim and remand for its reconsideration and proper disposition on the present or a reopened record in accordance with this opinion.

SO ORDERED.