Opinion for the Court filed by Circuit Judge WALD.
Opinion concurring in part and dissenting in part filed by Senior Circuit Judge CELE-BREZZE.
WALD, Circuit Judge:
These appeals contest the district court’s dismissal of consolidated individual and class sex discrimination claims against the Director of the United States International Communication Agency (“ICA” or “Agency”), formerly the United States Information Agency. Appellants contend that the district court (1) evaluated under inappropriate legal standards the statistical and testimonial evidence of a pattern and practice of discrimination in hiring, (2) failed to make required fact findings on the class promotion discrimination and retaliation claims, (3) improperly dismissed an individual claim for failure to exhaust administrative remedies, and (4) misapplied the requirements for a prima facie showing of discrimination to another individual claim. We find merit in certain of appellants’ objections and therefore remand the class claims and the individual claim of Rose Kobylinski for further consideration. We affirm, however, the district court’s dismissal of Luba Medina’s individual claim.
I. Background
In March 1977, Luba Medina, a former Agency employee, filed an individual claim for damages and declaratory and injunctive relief under Title VII of the Civil Rights Act of 1964, as amended by the Equal Employment Opportunity Act of 1972, 42 U.S.C. §§ 2000e-2000e-17. Her complaint alleged that, since 1974, the Agency had refused to rehire her in retaliation for her own prior charges of sex discrimination and her husband’s work on behalf of Agency minority employees. She also claimed that she had personally suffered from the Agency’s discriminatory practices against the foreign-born and women. In late 1977, another job applicant, who had been denied employment by the Agency earlier in the year, filed a Title VII class claim on behalf of female applicants and employees against whom the Agency had discriminated in hiring and promotion. In April 1978, the class was conditionally certified “to include all women who have applied for employment with or are currently employed by the United States Information Agency and who have been or continue to be adversely affected by the discriminatory employment practices of the defendant.” Joint Appen[375]*375dix (“J.A.”) at 22. Later that month, an Agency contract employee filed a complaint charging that she had been denied a permanent Agency position on account of sex. In November the three cases were consolidated. In the interim, the district court had permitted Medina and two Agency employees, Josefina Martinez and Rose Kobylinski, to intervene as named plaintiffs and had allowed plaintiffs to supplement the class complaint to include a claim that the Agency maintained “a practice of reprisals against women who have filed sex discrimination charges against the Agency.” J.A. at 28. On April 19, 1979, plaintiffs filed a motion for preliminary injunction to enjoin the defendant “from taking any retaliatory action against individuals who oppose the defendant’s discriminatory practices or otherwise exercise their rights under Title VII.” On May 16, the motion was denied orally without prejudice.
The parties agreed to bifurcate trial of the class claims into “liability” and “remedial” stages,1 and a bench trial on liability was conducted from May 29, 1979 through June 5, 1979. On October 24, 1979, the district court issued an opinion and order which redefined the class to exclude women in clerical positions and dismissed the class claims. Medina v. Reinhardt, Nos. 77-0360, 77-2019 & 78-0762 (D.D.C. Oct. 24, 1979) (Medina I), J.A. at 68.
Plaintiffs filed appeals on December 21, 1979, but on September 19, 1980, this court dismissed the appeals under Fed.R.Civ.P. 54(b)2 because the residual individual claims remained to be heard. Three of the named plaintiffs voluntarily dismissed their individual claims, and trial of Medina’s and Kobylinski’s claims was conducted on December 15 and 16, 1980. On June 15, 1981, the district court rendered its decision dismissing Medina’s claim on the merits and Kobylinski’s claim because she had failed to exhaust her administrative remedies. Medina v. Reinhardt, Nos. 77-0360, 77-2019 & 78-0762 (D.D.C. June 15, 1981) (Medina II), J.A. at 118. This appeal followed.
II. The Class Claims
Although the district court’s “Findings of Fact” discussed rebuttal evidence as well as evidence introduced by plaintiffs to establish their threshold ease, the court ruled in its “Conclusions of Law” that the plaintiff class had failed to establish “a prima facie case of discrimination on the basis of sex,” Medina I at 13, J.A. at 80. The court’s conclusion rested primarily on rejection of both parties’ statistical studies on hiring patterns as “misleading due to a failure to define adequately the relevant labor market from which the Agency draws for qualified [376]*376personnel,” id. at 3, J.A. at 70. The court’s objection was that the Census occupational categories used for comparison “with the jobs in issue at the Agency simply do not match.” Id. at 6, J.A. at 73. We find, however, that the district court’s opinion reflects a basic misperception of the relevancy and role of statistical evidence in the plaintiffs’ prima facie showing; hence, we remand for a redetermination of whether plaintiffs can make out a prima facie case of sex discrimination. Further, we must remand because the court made no findings or comment on plaintiffs’ evidence of Agency reprisals against women asserting their rights under Title VII.
Had the court credited either appellants’ or appellee’s definition of the relevant labor market, it would have found “disparities between the women employed at the Agency and the external labor pool of (1) Electronic Technicians, (2) Radio Broadcast Technicians, (3) Writers/Editors, and (4) Foreign Information Specialists.” Id. at 8, J.A. at 75. In 1977, when the class action was initiated, these four categories accounted for a major part of the Agency’s nonclerical positions. See, e.g., United States Information Agency FY-1978 Affirmative Action Report (Plaintiff's Exhibit No. 22(b)). Consequently, on remand, the district court should reconsider whether these disparities alone or in combination with testimonial evidence3 are sufficient to raise an inference of discrimination in hiring and, if so, whether that inference was adequately rebutted. Upon remand, the court should also address the class retaliation claim.
A. Relevant Labor Market
The 1972 amendments to the Civil Rights Act of 1964 came in response to the “persistence of discrimination” and the consequent need for more effective enforcement. H.R.Rep.No.238, 92d Cong., 1st Sess. 3 (1971), U.S.Code Cong. & Admin.News 1972, p. 2137. The legislative history particularly focused on the seriousness of sex discrimination, id. at 4-5, and explicitly recognized the need “[t]o correct ... entrenched discrimination in the Federal service.” Id. at 24, U.S.Code Cong. & Admin.News 1972, p. 2159. It is noteworthy that Congress itself relied on “statistical evidence” to prove the existence of sex discrimination in higher level government jobs.
Statistical evidence shows that minorities and women continue to be excluded from large numbers of government jobs, particularly at the higher grade levels.
This disproportionate distribution of minorities and women throughout the Federal bureaucracy and their exclusion from higher level policy-making and supervisory positions indicates the government’s failure to pursue its policy of equal opportunity.
[377]*377Id. at 23, U.S.Code Cong. & Admin.News 1972, p. 2158. See S.Rep.No.415, 92d Cong., 1st Sess., 421-23 (1971). Congress thus extended to federal employees the right to bring individual and class actions under Title VII.
In a Title VII suit, the' cláimant “carries the initial burden of showing actions taken by the employer from which one can infer, if such actions remain unexplained, that it is more likely than not that . .. the employer is treating ‘some people less favorably than others because of their race, color, religion, sex or national origin.’ ” Furnco Const. Corp. v. Waters, 438 U.S. 567, 576-77, 98 S.Ct. 2943, 2949, 57 L.Ed.2d 957 (1978) (quoting International Bhd. of Teamsters v. United States, 431 U.S. 324, 335 n.15, 97 S.Ct. 1843, 1854 n.15, 52 L.Ed.2d 396 (1977)). When a plaintiff submits sufficient evidence to permit such an inference, Title VII gives it the status of a “legally mandatory, rebuttable presumption.” Texas Dept. of Community Affairs v. Burdine, 450 U.S. 248, 254 n.7, 101 S.Ct. 1089, 1094 n.7, 67 L.Ed.2d 207 (1981). Because unlawful discriminatory intent is typically elusive of direct proof, Congress has deemed it appropriate to then require an explanation of the defendant.
In a sex discrimination class action charging disparate treatment, appropriate statistical comparisons may be used to indicate whether similarly situated men and women have been treated similarly, see, e.g., Valentino v. United States Postal Serv. (USPS), 674 F.2d 56, 69 (D.C.Cir.1982) (quoting Valentino v. United States Postal Serv., 511 F.Supp. 917, 940 (D.D.C.1980)), and, if not, whether the difference in treatment shown supports an inference of discriminatory intent. See, e.g., Teamsters, 431 U.S. at 325 n.15, 97 S.Ct. at 1854 n.15. Where specialized skills are legitimately required for employment, “[t]he proper comparison is between the composition of the [employer’s] work force and the qualified population.” Davis v. Califano, 613 F.2d 957, 963 (D.C.Cir.1979) (As Amended Feb. 14, 1980). See Valentino v. USPS, 674 F.2d at 68. (“When the job qualifications involved are ones that relatively few possess or can acquire, statistical presentations that fail to focus on those qualifications will not have large probative value.”) We have recently restated, however, that not every conceivable qualification for every separate job must be taken into account in making out a prima facie class claim of discrimination: “[T]he qualifications a Title VII plaintiff must grapple with .. . are threshold or ‘minimum objective’ qualification.” Id. at 71 n.24 (quoting Davis v. Califano, 613 F.2d at 964)). Thus, plaintiffs must identify the population likely to possess the minimum objective qualifications required of Agency employees (the relevant labor pool) and compare the proportion of women in that population with the proportion of women employed in the Agency. The comparisons in turn must show disparities of sufficient magnitude that they are statistically unlikely to have occurred by chance. We are then entitled to assume that “absent discriminatory employment practices, the proportion of the protected group in each of the job classifications and grade levels would approximate the proportion of the protected group with the minimum necessary qualifications . . . . ” Id. at 964. See, Teamsters, 431 U.S. at 339 n.20, 97 S.Ct. at 1856 n.20. Thus, statistically significant disparities between the composition of an employer’s work force and the labor pool from which the employer draws indicate that similarly situated people have been treated differently and “alone may in a proper case constitute prima facie proof of a pattern or practice of discrimination.” Hazelwood School Dist. v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977).
Here, because the district court did not reach the issue, we have no occasion to consider whether the magnitude of the statistical disparities shown was adequate to infer discriminatory motive. We are concerned in this appeal only with whether there is “a basis for a reasonable assumption” that the comparison population was qualified for Agency positions. Metrocare v. Washington Metropolitan Area Transit Auth. (WMATA), 679 F.2d 922, 930 (D.C.Cir.1982). In this case, the experts testify[378]*378ing on both sides proceeded through trial on the assumption that the population sufficiently well-qualified to be employed in Agency occupational categories was the population employed in those same occupations outside the Agency. We think this is a reasonable threshold assumption which follows from the Supreme Court’s reasoning in Hazelwood School Dist. v. United States. In Hazelwood, a school district was charged with racial discrimination in teacher hiring, and United States Census data recording employment in the relevant occupational categories were used to calculate the disparities that formed the basis for plaintiffs’ prima facie case. The Supreme Court specifically approved the technique, noting that “[t]he comparative statistics . . . were properly limited to public school teachers, and therefore this is not a case ... in which the racial-composition comparisons failed to take into account special qualifications for the position in question.” 433 U.S. at 308 n.13, 97 S.Ct. at 2742 n.13. Thus, Hazel-wood established that the proportion of a protected group actually employed elsewhere in the relevant occupation(s) is a meaningful measure of the proportion of the protected group qualified for employment by the defendant. The district court’s opinion here, however, raises the question whether there is too much diversity within the occupations involved in this case to permit reliance on the Hazelwood assumption as a basis for the plaintiffs’ prima facie showing. The district court concluded that the Census data used by the experts on both sides here was not sufficiently reflective of the qualifications required for Agency positions. The court insisted on “statistical data which matches those job categories at the Agency and the specific requirements thereof,” Medina I at 7, J.A. at 74 (emphasis supplied), and concluded in its “Findings of Fact”:
10. ... The job categories used by the parties’ experts do not correspond with the jobs in the defendant Agency. Neither do plaintiffs’ nor defendant’s experts adequately explain that the tasks actually performed by the employees at the Agency, in the job categories analyzed, correspond in any more than a very general and speculative way to those utilized by the parties’ experts.
11. Cross-mapping of actual employee activities for purposes of comparison with statistics concerning available labor pools is appropriate and useful where the inquiry is of general non-specialized skills. While statistics are helpful and useful in many cases, it must be understood that it cannot be argued or found in this case that precise labor pool availability figures can be derived to determine the number of females available for employment in such specialized fields as, for example, Cambodian language news analyst/writer/broadcaster.
Id. at 8-9, J.A. at 75-76 (footnote omitted) (emphasis supplied).
While definition of the relevant labor market is normally reviewable under the “clearly erroneous” standard as an “essentially factual matter within the special competence of the district court,” Castaneda v. Pickard, 648 F.2d 989, 1003 (5th Cir. 1981); see Hazelwood, 433 U.S. at 312-13, 97 S.Ct. at 2744, “if the trial court bases its findings upon a mistaken impression of applicable legal principles, the reviewing court is not bound by the clearly erroneous standard.” Inwood Laboratories, Inc. v. Ives Laboratories, Inc., — U.S. —, — n.15, 102 S.Ct. 2182, 2189 n.15, 72 L.Ed.2d 606 (1982). A close scrutiny of the legal underpinnings of the district court’s fact finding is appropriate here because the court’s decision was expressly based on its interpretation of the standard of proof enunciated in Hazelwood and Teamsters. The district court observed that Hazelwood “indicated that statistics comparing the employer’s work force and the relevant labor market must be based on the labor pool truly relevant to the employer’s potential work force,” and concluded that “the statistics suffer from the following deficiency, as noted by the Supreme Court in Teamsters : Imprecise definitions of the relevant labor-market when particular qualifications are required for the job(s) in question.” Medina I at 12, J.A. at 79. We conclude, how[379]*379ever, based on our examination of these cases, that the standard of precision the district court demanded, far from being mandated by these cases, is unprecedented and unjustifiable, insofar as it results in a total rejection of the Census data as a basis for statistical comparisons to establish a prima facie case.
The methods employed in this case by the experts on both sides to identify Census categories comparable to Agency positions, in fact, closely track that adopted in Hazel-wood and by other courts, see, e.g., Rivera v. City of Wichita Falls, 665 F.2d 531 (5th Cir. 1982); Croker v. Boeing Co. (Vertrol Div.), 437 F.Supp. 1138 (E.D.Pa.1977), aff’d, 662 F.2d 975 (3d Cir. 1981). Both experts subdivided the Agency work force into occupational categories and sought to translate each Agency category into Census terminology (“cross-map”) by reference to the U. S. Department of Commerce, Bureau of the Census, Alphabetical Index of Industries and Occupations (1971) (Defendant’s Exhibit No. 2) which lists “approximately . .. 23,000 occupation titles in alphabetical order.” Id. at iii. The Alphabetical Index explains the design of the Census classification system which groups those titles under some 440 occupational categories. Each category includes all the titles considered to be part of the same occupation.
To organize and make understandable the information relating to the many thousands of industries and occupations, a system of homogeneous grouping or classification must be used. Homogeneous titles are grouped together to form the various categories which comprise the system .... In this Index each title is identified by the code for that category to which it is assigned.
For example, plaintiffs’ expert explained the composition of the Census category “Editors and Reporters.”
Census Code 184, covering editors and reporters, is a list of about 100 titles which all fit into a journalistic type of occupational group, including just, for example, editor, feature writer, foreign correspondent, newspaper writer, and newspaper editor.
Trial Transcript (“Tr.”) at 82 (May 29,1979) (testimony of M. Rosenblum).
The defendant’s expert testified that in the “overwhelming majority of occupations” cross-mapping is accomplished by looking up the Agency position title in the Index and identifying the Census category to which it belongs. Tr. at 19 (June 1,1979) (testimony of S. Wolfbein). Where relevant Agency job titles were not included in the Alphabetical Index, defendant’s expert testified that he translated Agency categories into Census terminology based on job descriptions provided by the Agency. Tr. at 23 (June 1, 1979) (testimony of S. Wolfbein). Plaintiffs’ expert testified that he consulted job descriptions in order to cross-map all the relevant positions.
I consulted the 118 Manual to read the job description, as published by Civil Service, covering those Civil Service titles and codes that are used by all federal agencies.
In a number of these instances I also consulted material published by the Agency, itself, to augment and fill in additional descriptions.
So that I got a better sense in my own mind of specifically which Census occupational category would be appropriate for this cross-mapping exercise.
Tr. at 83 (May 29, 1979) (testimony of M. Rosenblum). Plaintiffs’ expert testified that he also consulted an Office of Personnel Management (“OPM”) study that translated white collar civil service jobs into Census terms, although he disagreed with OPM’s cross-mapping in one instance.
Because the Census has fewer occupational categories (approximately 440) than the more detailed Civil Service system (over 1,000), the cross-mapping necessarily involved fitting some Agency occupational categories at issue in the class action into broader Census categories. For example, based on Census coding, the defendant’s expert placed both the Agency positions entitled “Radio Broadcast Technician” and “Electronic Technician” in the Census category entitled “Electrical and electronic en[380]*380gineering technicians.” Tr; at 20, 93-94 (June 1, 1979) (testimony of S. Wolfbein). He also placed both “Writers/Editors” and “Foreign Information Specialists” in the Census category “Editors and Reporters.” Id. 103-04. We are satisfied that comparing Agency occupational categories to the broader Census categories is appropriate because all types of jobs the Census includes within any given Census occupational category are sub-specialties of that occupation; thus, such aggregations retain “generally similar job skills” in common. Valentino v. USPS, 674 F.2d at 68. (“The burden of comparing appropriate groups in terms of minimum objective qualifications, onerous here because of the disparate occupational categories involved, is far more tractable when all members of the class are professional, administrative or technical employees with generally similar job skills and seek [employment in or] advancement to positions involving those same skills.”). Our conclusion is supported in this case by the more refined cross-mapping attempted where Agency occupations involved skills arguably reflected in more than one Census occupational category. As we have pointed out, the experts agreed on the basic methodology involved in. identifying the relevant labor pool although they disagreed as to which Census occupational category more properly encompassed certain Agency positions. As these disagreements came only in fine-tuning the comparisons, however, they do not deprive the statistics of probative value, but, in fact, enhance it since the disagreements caused the experts to focus on specific Agency job requirements and tasks and thus accomplish the cross-mapping with considerable attention to detail.4 [381]*381To the extent that the experts disagreed on the appropriate Census category to which Agency categories should be compared, we of course defer under the “clearly-erroneous” standard, see Fed.R.Civ.P. 52(a), to the district court’s judgment as to which comparison has the greater probative value. And, we would not second-guess the district court as to other areas of disagreement between the experts which the court did not decide, e.g., whether 1970 Census data or 1978 Labor Department data provided the appropriate set of figures. We decide only that the cross-mapping by both experts here provided an adequate basis from which to derive meaningful disparity figures in order to decide if a prima facie case of discrimination in hiring was made out.
A review of the statistical comparisons sanctioned in Hazelwood bolsters our conclusion that the district court imposed an inappropriately high standard of precision between Agency and Census job categories. The Hazelwood Court was satisfied with data that limited the relevant labor pool to those in the general Census occupational category of secondary school teachers although this data aggregated diverse teaching positions not subdivided on the basis of subject matter taught. Thus, the district court mistakenly relied on Hazel-wood for authority that plaintiffs must provide data comparing the labor market for every combination of skills required in every one of the more than 2,000 Agency jobs at issue. We do not believe a plaintiff is required to prove that each individual in the comparison pool is qualified in every way for a particular Agency position. The objective is to define “a population that closely approximates the characteristics of those who would be likely to apply” and “meet legitimate threshold qualification requirements.” D. Baldus & J. Cole, Statistical Proof of Discrimination 120 (1980) (emphasis supplied). The focus thus should be on whether the Census statistics give us a meaningful estimate of the proportion of women in the labor market reasonably likely to possess the minimum qualifications needed for the Agency jobs in question.
We agree with the district court that the ICA positions at issue are properly treated differently from the bulk of federal government jobs which are generally professional, administrative and managerial positions for which no differentiated training or educational standards are imposed as minimal qualifications. And we agree as well that the test was not met in a case like Valentino, where the statistics “did not group employees by job category,” 674 F.2d at 70, nor “hone in on the wide variety of minimum objective qualifications required of applicants for the diverse .. . positions” at issue. Id. at 61. In Valentino, where discrimination in promotion was charged, it would indeed have been “irrational to assume ‘equal qualifications’ to fill engineering or secretarial vacancies,” as the plaintiffs urged, simply because employees were “educated the same number of years and employed by the government for the same length of time.” Id. at 71. See also Metrocare v. WMATA, 679 F.2d at 930 (no showing that “persons now holding secretarial or clerical jobs are qualified for [promotion to] managerial positions”). The data in this case, however, did hone in on the basic technical skills — “the minimum objective qualifications,” Valentino v. USPS, 674 F.2d at 68 (quoting Davis v. Calif ano, 613 F.2d at 964)—prerequisite to employment in particular Agency occupational categories. The expert testimony reveals the comparisons of Agency and Census occupational categories [382]*382were based on common job requirements and were accomplished in some instances with much greater precision than in Hazel-wood. It should be noted again that in Hazelwood the comparison pool included public school teachers whether they taught, for example, natural science or a foreign language.
Therefore we do not deem it fatal to plaintiffs’ prima facie case that the Census occupational data failed to take account of foreign language skills prerequisite to employment in certain Agency positions. “[N]ot every conceivable factor relevant to [an employment] decision must be included in the statistical presentation ... . ” Davis v. Califano, 613 F.2d at 964. See, e.g., Trout v. Hidalgo, 517 F.Supp. 873 (D.D.C.1981):
Certainly, plaintiffs’ expert did not, in his analysis, account for each of the factors that the government suggests should have been considered. It is also true that a model which incorporated additional potentially relevant factors (such as type or quality of education and experience) would form a more perfect foundation for determinations regarding allegations of discrimination. However, defendants have furnished no evidence that inclusion of the missing variables or refinement of others would have altered rejection of the hypothesis of no discrimination. Indeed, they failed to offer any evidence indicating that type of education and experience or quantity of experience per age was distributed unequally among . .. women and men in the . . . population.
517 F.Supp. at 881 (emphasis supplied).
Here, many, if not most, of the jobs involved do not require foreign language skills at all.5 Thus, to'the extent that the district court rejected the statistics for failure to account for such skills, the court imposed an additional and unnecessary requirement for a large number of Agency positions. Further, with respect to positions which include specific foreign language skills among the minimum objective qualifications (e.g., “Cambodian language news analyst/writer/broadcaster”), the court articulated no basis for the assumption that such skills are in fact unevenly distributed between men and women gener-’ ally or in the particular occupations involved. The more logical assumption, barring proof to the contrary, is that equal numbers of men and women possess skill in any given language; thus, the proportion of women qualified for Agency positions would not necessarily change if this variable were included in the occupational data. And, practically, statistical data, so far as we can tell from the record, are simply not available correlating 440 Census occupational categories with several dozen foreign language skills;6 in their absence, we think it appropriate here to afford plaintiffs the benefit of a rebuttable presumption of an equal distribution of the relevant language skills.7 We underline that we are dealing [383]*383here with the showing necessary for a prima facie case only. “In a Title VII case, the allocation of burdens and the creation of a presumption by the establishment of a prima facie case is intended progressively to sharpen the inquiry into the elusive factual question of intentional discrimination.” Texas Dept. of Community Affairs v. Burdine, 450 U.S. at 225 n.8, 101 S.Ct. at 1094 n.8 (emphasis supplied). Exactness is not required at the prima facie stage. As a consequence, in rebuttal, a defendant need only raise “a genuine issue of fact as to whether it discriminated” and need not even “persuade the court that it was actually motivated” by nondiscriminatory reasons. Id. at 254,101 S.Ct. at 1094. The defendant here is certainly entitled to rebut plaintiffs’ showing with evidence, more readily available to it than to plaintiffs, that, as to certain jobs with foreign language requirements, there are disproportionately fewer qualified women candidates available or even that bona fide recruitment efforts have resulted in a proportionately lower number of qualified female applicants than men. Cf. EEOC v. Radiator Specialty Co., 610 F.2d 178, 185 n.8 (4th Cir. 1979) (“Requiring the defendant to show the inappropriateness of general population statistics in such situations follows the principle of allocation of proof to the party with the most ready access to the relevant information.”). We find it significant here, however, that the defendants themselves did not argue to the trial court that failure to control for language invalidated the occupational comparisons.8
[384]*384We therefore cannot accept the district court’s total rejection, as too imprecise, of both experts’ comparisons of Agency occupational categories with Census occupational categories. Were trial courts to apply Hazelwood and Teamsters as the court did here, statistical evidence would rarely be acceptable in Title VII class actions because statistical evidence is virtually always lacking in the degree of precision demanded by the district court. “[I]n most cases, conditions are far from ideal, with incomplete qualification data and non-random samples being the rule rather than the exception,” D. Baldus & W. Cole, supra, at 26-27. And yet the Supreme Court’s “cases make it unmistakably clear that ‘[statistical analyses have served and will continue to serve an important role’ in cases in which the existence of discrimination is a disputed issue.” Teamsters, 431 U.S. at 339, 97 S.Ct. at 1856 (quoting Mayor of Philadelphia v. Education Equality League, 415 U.S. 605, 620, 94 S.Ct. 1323, 1333, 39 L.Ed.2d 630 (1974)).
Thus relevant labor pool statistics are commonly used although it is rarely possible to be exact in the definition of the relevant labor pool. Sometimes imprecision works to the detriment of plaintiffs as well as defendants. For example, a comparison labor pool based on Census employment statistics does not include all those qualified. “[C]ensus statistics analyzing the population by job skill include[] in each skill category only people actually employed in those skill categories. People qualified for, but not employed in, such positions [are] omitted from the statistics quantifying the proportion of the population eligible for the type of employment in question.” Rivera v. City of Wichita Falls, 665 F.2d at 544 n.19. Certain defects in statistical evidence may, of course, be fatal to a plaintiff’s case, as in Valentino where comparisons were grossly imprecise or in a case, hypothesized in Va[385]*385¡entino, where the sample size is inordinately small. Valentino v. USPS, 674 F.2d at 66 n.12 (citing Wilkins v. University of Houston, 654 F.2d 388, 409 n.37 (5th Cir. 1981)) (“[T]he breakdown of highly specialized workplaces into occupational categories for the purpose of examining the treatment of similarly qualified employees may yield numbers too small to conduct certain types of statistical analyses relied upon to show discrimination in workplaces less specialized.”) But because “statistical measures are necessarily imperfect in differing ways and varying degrees,” the courts generally “accept what figures are available; allow for imperfections, skewing factors, and margins of error; and then take the figures for what they are worth. Sometimes this is much, sometimes little.” Phillips v. Joint Legislative Committee on Performance and Expenditure Review, 637 F.2d 1014, 1025 (5th Cir. 1981), cert. denied, — U.S. —, 102 S.Ct. 2233, 72 L.Ed.2d 845 (1982).
In the usual case, statistics are not intended to “conclusively prove intentional discrimination.... In recognition of [the] limits on the potential of statistics as a basis for an inference, the courts have given statistical proofs a question-raising, burden-shifting function.” D. Baldus & W. Cole, supra, at 26-27. We find the base data here to be sufficiently precise and consistent with statistical and legal norms to permit an inference of discrimination if statistically significant disparities exist. We therefore remand for reconsideration of whether plaintiffs made a prima facie showing of Agency discrimination in hiring.
B. Required Findings
Plaintiffs also protest on appeal that, with respect to the class claims of promotion discrimination and retaliation,9 the district court’s opinion was deficient under Fed.R.Civ.P. 52(a) which requires that a court sitting without a jury “find the facts specially and state separately its conclusions of law thereon.”10 It is established that the requirement of fact findings cannot be met by a “statement of ultimate fact without the subordinate factual foundations for it which must be the subject of specific findings.” O’Neill v. United States, 411 F.2d 139, 146 (3d Cir. 1969). Further, the fact findings must touch all material issues. “For this court to exercise adequately its power of review, the district court must make specific findings about the nature and truth of [plaintiffs’] allegations.” Borrell v. ICA, 682 F.2d 981 at 992 (D.C.Cir.1982). Because the district court’s opinion is bereft of reference to the retaliation claim, we must remand for findings on this issue. We are satisfied, however, with the court’s findings on the promotion discrimination claim.
To support an inference of discriminatory promotion practices, the plaintiffs introduced undisputed government statistics showing the small percentage of women in higher level Agency positions. E.g., the Agency’s FY-1978 Affirmative Action Plan, Sec. C, Table 3 (Plaintiffs’ Exhibit No. 22(b)); U. S. Civil Service Commission, Report on Review of Personnel Management in the United States Information Agency (Plaintiffs’ Exhibit No. 23). The plaintiffs’ proposed findings of fact with respect to the promotion claim were based on this statistical evidence and they object on appeal that the district court’s opinion failed to include any reference to the data or the inferences to be drawn therefrom. We, too, [386]*386find it troubling that, while the district court devoted five pages of fact findings to the statistical evidence related to the hiring discrimination claims, the court ignored the statistical evidence presented on the promotion claims. The court, however, acknowledged the allegation of promotion discrimination, stating the issue before it as “[w]hether the defendant’s hiring, promotion and salary practices constitute patterns or practices of discrimination.... ” Medina I at 3, J.A. at 70. And, although the court did not specifically discuss the statistical evidence on promotion practices, it made findings based on defendant’s testimonial evidence, concluding:
In addition to [defendant’s witnesses] very credible testimony, the fact that these women have attained the positions they now occupy, and have done so by rapid and consistent advancement, is dis-positive of the absence of any pattern or practice of discrimination based on sex at the Agency at all relevant periods in this litigation.
Medina I at 11, J.A. at 78. The “ultimate fact,” that there existed no pattern or practice of discrimination, was thus supported by the specific finding of instances of accelerated promotion of women and the credible testimony of women defense witnesses regarding the absence of discrimination in “the advancement of women in any manner.” Id. at 11. While this court would have been aided by an explicit statement of the district court’s reasons for rejecting the inference plaintiffs urged be drawn from the statistics, the district court provided findings “sufficient for a clear understanding of the basis of the decision.” 9 C. Wright & A. Miller, Federal Practice and Procedure § 2577 at 697 (1971).
In contrast, the district court’s opinion failed to acknowledge the class retaliation claim. The court’s recitation of the procedural history does not refer to the supplemental complaint alleging retaliatory practices nor do the “Findings of Fact” or “Conclusions of Law” address the claim. Because a retaliation claim does not depend on whether the challenged employment practices are determined to be unlawful, see Parker v. Baltimore & O. R. Co., 652 F.2d 1012, 1018-19 (D.C.Cir.1981), the court is required to address the retaliation issue independently.
III. The Individual Claims
The individual discrimination claims of Luba Medina and Rose Kobylinski were the subject of a separate trial and a second opinion in which the court dismissed Medina’s claim on the merits and denied jurisdiction over Kobylinski’s claim because she had not filed a charge with the Equal Employment Opportunity Commission (EEOC), and thus had failed to exhaust her administrative remedies as required by 42 U.S.C. § 2000e-16. Since the trial, however, the Supreme Court has decided that the requirement of timely filing is akin to a statute of limitations and is not a jurisdictional prerequisite, Zipes v. Trans World Airlines, 455 U.S. 385, 102 S.Ct. 1127, 71 L.Ed.2d 234 (1982), and this court has held that “the critical factor in determining whether an individual Title VII plaintiff must file an EEOC charge, or whether he may escape this requirement by joining with another plaintiff who has filed such a charge, is the similarity of the two plaintiffs’ complaints.” Foster v. Gueory, 655 F.2d 1319, 1322 (D.C.Cir.1981). We reverse the district court’s dismissal of Kobylinski’s claim because we find that her claim was so similar to that made by Martinez, who had filed an EEOC charge and with whom Kobylinski intervened as a named plaintiff, “that it can fairly be said that no conciliatory purpose would be served by filing separate EEOC charges.” Id. We affirm, however, the court’s dismissal of Medina’s claim as based on fact findings that are not “clearly erroneous.”
A. Exhaustion of Administrative Remedies
In Foster v. Gueory, this court reversed the district court’s denial of a motion to intervene in a pending employment discrimination suit. The motion had been denied on the ground that the parties who sought [387]*387intervention had failed to exhaust their administrative remedies, but this court concluded that the purposes of the exhaustion requirement had been served by the initial plaintiffs’ filing. The “principal functions of the EEOC filing requirement” are to enable “the EEOC to provide the alleged wrongdoer with notice and to permit possible conciliation.” Id. at 1323. The Foster court concluded that separate filing is required of co-plaintiffs if there exists “a real possibility that one of the claims might be administratively settled while the other can be resolved only by the courts.” Id. at 1322. But the court held that where two plaintiffs allege that they were similarly situated and received the same discriminatory treatment, the purposes of the exhaustion requirement are adequately served if one plaintiff has filed an EEOC complaint.11 Here, the claims of Martinez and Kobylinski are virtually identical. Each claimed intentional sex discrimination (as opposed to disparate impact). Each was a GS-11 Agency employee who based her claim on defendant’s failure to promote her (as opposed to a failure to hire). Each also charged that they performed like services for less pay than men of similar qualifications and experience. Thus, we find sufficient similarity between these claims to doubt the likelihood that conciliation would prove successful as to one where it had failed as to the other.
As the district court found in its order granting the motion to intervene, Hartman v. Reinhardt, No. 77-2019 (D.D.C. Sept. 18, 1978), J.A. at 23, plaintiff Martinez had exhausted administrative remedies with regard to her claims. See Notice of Final Decision of Agency (Aug. 30, 1978) (from EEO officer to Martinez). We therefore hold that Kobylinski was not required to file an EEOC charge, and remand her case to the district court for a decision on the merits.
B. Medina’s Claim
Appellant Medina protests that the dismissal of her individual claims was based on the district court’s misinterpretation of the standard of prima facie proof enunciated in McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973). However, in dismissing Medina’s claim, the court ultimately relied on the adequacy of defendant’s rebuttal evidence. As the court’s findings are not “clearly erroneous,” we affirm the court’s disposition of Medina’s claim.
Medina claims that the Agency discriminated against her both on the basis of sex and in retaliation for her previous EEOC complaints and her husband’s representation of minorities in EEOC actions. At trial she presented evidence regarding three separate instances of alleged discrimination: (1) the denial of the opportunity to retest for a newsroom position following an unsatisfactory test performance several months earlier; (2) the Agency’s rejection of her application for a radio production [388]*388position; and (3) the Agency’s rejection of her application for a position as a foreign language broadcaster.
The district court found that Medina was entitled to retake the test and ordered the Agency to allow a retest, but the court further found that the Agency’s denial of Medina’s right to retest was a mistake rather than discrimination. As to the Agency’s rejection of both job applications at issue, the court found that Medina’s proof failed to satisfy the last of the four prerequisites to a prima facie showing enumerated in McDonnell Douglas. The court understood McDonnell Douglas to require that a discrimination plaintiff show:
(1) he or she is a minority group member;
(2) he or she applied for a job with the defendant and was qualified for the position;
(3) he or she was rejected despite his or her qualification; and
(4) after the rejection, the job remained open, and the defendant sought other applications.
Medina II at 8, J.A. at 125. To the extent that the court treated these factors as delineating the only circumstances that might give rise to an inference of unlawful discrimination, the court erred. In fact, McDonnell Douglas defines but one “model” of such circumstances. Texas Dept. of Community Affairs v. Burdine, 450 U.S. 248, 253 & n.6, 101 S.Ct. 1089, 1093 & n.6, 67 L.Ed.2d 207 (1981). But the district court’s error is not dispositive because the court further stated that even if a prima facie case had been made out, the Agency had articulated legitimate nondiscriminatory reasons for its actions and that defendant’s proof showed these reasons were not pretextual. This conclusion rested on the court’s finding that “other qualified individuals were selected in lieu of Ms. Medina.” Medina II at 10-11, J.A. at 127-28.
In reviewing the court’s findings that the challenged personnel actions resulted from mistake in one instance and reflected legitimate employment decisions in the others, we are bound to give “due regard ... to the opportunity of the trial court to judge the credibility of the witnesses,” and we may not set aside these findings unless they are “clearly erroneous.” Fed.R. Civ.P. 52(a). While, again, we would have been aided on review by discussion of the specific evidence upon which the court relied, we find, in the record, evidence that supports the court’s findings. In assessing the qualifications of the man selected in lieu of Medina as Radio Production Specialist, the court relied on the application he submitted to the Agency which described a long career as a “News Director/Announcer.” Defendant’s Exhibit No. 38, J.A. at 409. The court based its findings as to the qualifications of persons selected as foreign language broadcasters upon the testimony of the Chief of the Ukranian Service of the Voice of America, who specifically identified her reasons for selecting the other candidates over Medina. Tr. at 200-01 (Dec. 16, 1980) (testimony of O. Dragan). We find no reason to abandon the presumption that the trial court correctly assessed the evidence and the witnesses’ credibility. Finally, the court did not explain its conclusion that Medina was denied a retest by mistake rather than due to discrimination. We note, however, that the testimony about the incident recounts only that Medina was told there would be “no point” to a retest, Tr. at 102 (Dec. 15, 1980) (testimony of L. Medina), and that her initial performance had received “fairly severe ratings,” EEOC Report of Investigation, Attachment F — 1 (Affidavit of Bernard Kamenske (Nov. 29, 1976)), Defendant’s Exhibit No. 39 (Pt. II). The district court could infer from this evidence that, while Medina ought to have been allowed to retest, which the court ordered, she was discouraged from doing so because of the extreme unlikelihood of her improving her performance sufficiently to meet the Agency’s standards rather than because of discriminatory intent. Accordingly, we affirm the court’s dismissal of Medina’s claims of discrimination.
In view of the foregoing, the case is remanded to the district court for proceedings not inconsistent with this opinion.
So Ordered.