FERGUSON, Circuit Judge:

The Equal Employment Opportunity Commission (EEOC) appeals the district court's decision in the EEOC's gender discrimination action against General Telephone Company of the Northwest (GenTel). The EEOC argues that the district court erred in holding that GenTel did not discriminatorily deny its female employees access to higher-paying positions in violation of Title VII.

Initially, this panel affirmed the decision of the district court in a memorandum disposition. Upon reconsideration, that decision is withdrawn. We now reverse the decision of the district court.

I.

This action arose as the result of numerous gender discrimination charges against GenTel filed with the EEOC. The EEOC brought an action in which it contended that GenTel had engaged in a variety of improper practices which resulted in women being denied access to higher paying jobs within the company.

At a bench trial, the district court received oral and written testimony, along with anecdotal evidence of specific instances of sex discrimination. statistical evidence in the form of regression analyses, and evidence regarding GenTel's equal employment opportunity policies and programs. The court rejected the EEOC's regression analyses showing substantial disparities in the treatment of men and women in various job classifications and its anecdotal evidence of sex discrimination, and concluded that the EEOC had failed to prove that GenTel engaged in a company-wide pattern or practice of intentional discrimination in violation of Title VII, 42 U.S.C. § 2000e. The court based its judgment on the collective effect of four findings: (1) that the EEOC's statistical studies failed "to analyze in a meaningful way the extent to which career interests differed between males and females"; (2) that men and women had substantially different job and career interests, which GenTel demonstrated were related to job placements; (3) that GenTel had an active commitment to equal opportunity in employment; and (4) that the gender balance of GenTel's work force compared favorably with statistics for other public utilities and craft vocations in Washington state and nationwide. After concluding that GenTel was the prevailing party, the district court awarded GenTel attorney's fees and costs.

The EEOC appeals the district court's decision on two grounds. First, it argues that the district court erred in admitting evidence of GenTel's equal employment opportunity efforts after having exempted from discovery relevant self-critical materials. The EEOC also contends that the district court erred by not finding discrimination where fewer women than men were initially assigned to higher-paying jobs as allegedly demonstrated by its statistical and other evidence.

GenTel cross-appeals, claiming that the court erred in calculating the costs and fees to which it was entitled as the prevailing party. GenTel argues that it was entitled to an additional $425,000 in costs for expert witness fees. GenTel also claims, in the event of a reversal of the district court judgment, that the district court lacked proper jurisdiction over those of the EEOC's claims which alleged gender discrimination in the promotion of hourly-based workers in various job classifications.

II. A.

At trial, the district court admitted voluminous evidence offered by GenTel regarding its equal opportunity efforts. This evidence consisted in part of testimony from GenTel management and other employees charged with oversight and implementation of the company's affirmative action programs and complaint process. The remainder of this evidence was comprised of over two hundred pages of exhibits, including policy statements, in-house newspaper articles, and letters referring to GenTel's commitment to equal opportunity. The EEOC repeatedly objected to the admission of this evidence. The EEOC argues on appeal that admission of this evidence was an abuse of discretion because the court had previously precluded the EEOC from discovering relevant self-critical material. We agree.

It is clear that affirmative action or equal opportunity evidence is relevant to and probative of an employer's intent not to discriminate. See Coser v. Moore, 739 F.2d 746, 751 (2d Cir. 1984); Craik v. Minnesota State Univ. Bd., 731 F.2d 465, 472 (8th Cir. 1984); Ottaviani v. State Univ. of New York, 679 F. Supp. 288, 309 (S.D.N.Y. 1988).

Some courts have, however, extended a qualified privilege to self-critical portions of an employer's equal opportunity efforts, rendering them exempt from discovery under appropriate circumstances. See, e.g., Coates v. Johnson Johnson, 756 F.2d 524, 551 (7th Cir. 1985); Jamison v. Storer Broadcasting Co., 511 F. Supp. 1286, 1296-97 (E.D.Mich. 1981), aff'd in part and rev'd in part, 830 F.2d 194 (6th Cir. 1987). Underlying this privilege is the goal of removing discrimination from the work place. But when an employer voluntarily uses evidence of its equal opportunity efforts to prove nondiscrimination, it "opens the door" and waives whatever qualified privilege may have existed. Coates, 756 F.2d at 552. As the Seventh Circuit has explained,

[A]n employer should not be able to offer its affirmative action policy before the trier of fact as a manifestation of nondiscrimination and at the same time be able to hide self-critical evaluations that may undercut the employer's portrayal of its efforts. Fairness requires that the qualified privilege not be allowed to mask discrimination when the overall policy behind the privilege is directed toward eliminating it.

Coates, 756 F.2d at 552. Thus, facilitating one-sided presentation of a defense prevents the factfinder from getting "the full picture" of a defendant's conduct by precluding the plaintiff from enjoying a fair opportunity to challenge the evidence and the defendant's theory in offering it. Cf. Yatvin v. Madison Metro. School Dist., 840 F.2d 412, 415 (7th Cir. 1988) ("just as the establishment of a bonafide affirmative action plan might help rebut a claim of sex discrimination . . . so the violation of such a plan might help support such a claim" (citation omitted)): Here, the district court exempted from discovery relevant self-critical materials thus leaving the EEOC ill-equipped to effectively cross-examine those of GenTel's witnesses who testified concerning the implementation and efficacy of GenTel's equal opportunity efforts. Thus, the district court erred in admitting GenTel's equal opportunity evidence.

Furthermore, it is clear from the decision issued by the district court that the EEOC was prejudiced by this error. Fifteen of the court's fifty-three findings related to GenTel's equal opportunity programs and policies. Moreover, of all of the exhibits and testimony presented by GenTel at trial, the court discussed this evidence in a vastly more detailed manner than any other genre of GenTel's evidence. Since the court's judgment was based in primary part on this evidence, the court's admission of the evidence was prejudicial error. We therefore hold that the district court abused its discretion by admitting GenTel's equal opportunity evidence, once it had exempted GenTel's self-critical material from discovery.

B.

The EEOC also contends that the district court erred in assessing its statistical data. The court held that (1) in general, none of the EEOC's statistical studies sufficiently proved discrimination, and (2) more specifically, the EEOC's statistical analyses contained various "critical flaws" including the failure to adequately account for differences in job interest among men and women. These determinations regarding the EEOC's statistical analyses mirror the criticisms GenTel made of the EEOC's data at trial. The court's findings and conclusions as to the probative value of this evidence, however, are not supported with any explanation of how adequately adjusting for the alleged flaws would have eliminated the disparities shown in the EEOC's statistical analyses. This appears to be because GenTel never offered any such evidence. Thus, the EEOC challenges the court's determinations regarding its statistics in claiming that it proved, through statistics and evidence of specific instances, that GenTel had impermissibly discriminated against women in hiring.

While this circuit has not yet addressed precisely what kind of evidence will overcome the probative force of a plaintiff's statistical evidence, decisions of the Supreme Court and other circuits have given us guidance.

1.

Bazemore v. Friday, 478 U.S. 385, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986), is the Supreme Court's most relevant discussion of the validity of statistical studies in the form of regression analyses. In Bazemore, the Court stated that "[a] plaintiff in a Title VII suit need not prove discrimination with scientific certainty. . . ." 478 U.S. at 400, 106 S.Ct. at 3009 (Brennan, J., concurring for a unanimous court). The Court also rejected the Fourth Circuit's holding that an analysis which accounts for the major factors in an employment decision "must be considered unacceptable as evidence of discrimination," merely because some variables have been omitted, id. at 399-400, 106 S.Ct. at 3008 (quoting Bazemore v. Friday, 751 F.2d 662, 672 (4th Cir. 1984)), since "a regression analysis that includes less than `all measurable variables' may serve to prove a plaintiff's case." Id. 478 U.S. at 400, 106 S.Ct. at 3009.

Since Bazemore, several circuits have applied the principles articulated in that decision when addressing the use of regression analyses to prove discrimination. The majority of those courts have concluded that Bazemore requires that the defendant do more than simply point out possible flaws in the proponent's statistical analyses in order to rebut the inference of discrimination raised by the statistical evidence.

In Sobel v. Yeshiva Univ., 839 F.2d 18 (2d Cir. 1988), the Second Circuit addressed the use of regression analyses to prove a claim of gender-based wage discrimination at a university medical school. The court rejected the University's bare contention that the plaintiff had left out several important variables that would explain some of the disparity in salary which the plaintiffs' experts had attributed to gender discrimination. Id. at 34. The court found that the University did not show that the apparent gender disparity would in fact be reduced if these variables were taken into account. It then found that the University's experts had "simply criticized plaintiff's failure to include [the variables], offering no reason, in evidence or analysis, for concluding that they correlated with sex and therefore were likely to affect the sex coefficient." Id.; cf. Bazemore, 478 U.S. at 403 n. 14, 106 S.Ct. at 3010 n. 14 ("Respondents' strategy at trial was to declare simply that many factors go into making up an individual employee's salary; they made no attempt that we are aware of — statistical or otherwise — to demonstrate that when these factors were properly organized and accounted for there was no significant disparity between the salaries of blacks and whites.") (emphasis added).

The Second Circuit then held that Bazemore "require[s] a defendant challenging the validity of a multiple regression analysis to make a showing that the factors it contends ought to have been included would weaken the showing of a salary disparity made by the analysis." 839 F.2d at 34; see also Rossini v. Ogilvy Mather, Inc., 798 F.2d 590, 604 (2d Cir. 1986) (holding that trial court abused its discretion in rejecting plaintiffs' statistical table "on the speculative basis that the table's results might `possibly' have been different" if an unaccounted-for factor had been included.).

Sobel is consistent with earlier holdings of the District of Columbia and Eighth Circuits. Those courts also concluded that under Bazemore a defendant cannot overcome a strong statistical showing of discrimination merely by making an unsubstantiated assertion of error.

In Palmer v. Shultz, 815 F.2d 84 (D.C. Cir. 1987), the District of Columbia Circuit found that

[i]mplicit in the Bazemore holding is the principle that a mere conjecture or assertion on the defendant's part that some missing factor would explain the existing disparities between men and women generally cannot defeat the inference of discrimination created by plaintiff's statistics . . . . [I]n most cases a defendant cannot rebut statistical evidence . . . without introducing evidence to support the contention that the missing factor can explain the disparities as a product of a legitimate, nondiscriminatory selection criterion.

Id. at 101 (footnote omitted); see also Segar v. Smith, 738 F.2d 1249, 1287 (D.C.Cir. 1984), cert. denied sub nom., Meese v. Segar, 471 U.S. 1115, 105 S.Ct. 2357, 86 L.Ed.2d 258 (1985) ("Absent such evidence [that purported flaws explain disparities], we are left . . . with mere speculation and conjecture."); D. Baldus J. Cole, Statistical Proof of Discrimination, vii (1987 Supp.) (when statistical evidence is challenged on methodological grounds, the burden should be on the challenger to present evidence that the statistics are defective and how that flaw biases the results).

The Eighth Circuit reached a similar result in Catlett v. Missouri Highway and Transp. Comm'n, 828 F.2d 1260 (8th Cir. 1987), cert. denied, ___ U.S. ___, 108 S.Ct. 1574, 99 L.Ed.2d 889 (1988). That case involved successful class and individual claims of sex discrimination in the state's hiring of highway maintenance workers. The class had presented statistics showing a substantial disparity between the number of women in the relevant work force and the number hired by the state during relevant time periods. The court held that "[w]hile Missouri argues that the class in defining the relevant work force failed to consider the actual interest of otherwise qualified men and women in maintenance work, Missouri bore the burden of introducing evidence to show this failure was significant . . . ." Id. 828 F.2d at 1266; see also Sobel, 839 F.2d at 35 ("In short, the simple fact of imperfection, without more, does not establish that plaintiffs' model suffers from underadjustment, even though men score higher on the proxies.").

The only other circuit which has considered the issue is the Seventh Circuit in Equal Employment Opportunity Commission v. Sears, Roebuck Co., 839 F.2d 302 (7th Cir. 1988). Sears involved claims of sex discrimination in hiring, promotions, wages, and the availability of paid pregnancy leave. At trial, the EEOC relied primarily on regression analyses based on information obtained from Sears. Sears did not present its own regression analyses, but instead offered evidence intended to undermine the two assumptions it claimed lay at the foundation of the EEOC's statistical analyses: that men and women had equal interests and equal qualifications for commission sales positions. Id. at 312-13. The district court concluded that the EEOC's statistical studies were "virtually meaningless" because they failed, inter alia, to adequately capture differences in male and female interests and qualifications. Equal Opportunity Employment Commission v. Sears, Roebuck Co., 628 F. Supp. 1264, 1305 (N.D.Ill. 1986).

The Seventh Circuit affirmed the district court's conclusion, holding that the court's findings regarding the EEOC's statistical evidence were not "manifestly erroneous". 839 F.2d at 310 (relying on Soria v. Ozinga Bros., 704 F.2d 990, 995 n. 6 (7th Cir. 1983)). Dissenting from the majority's statistical discussion, Judge Cudahy criticized the majority for uncritically accepting the defendant's unsupported contention that differences in job interests and other subjective qualities would account for the huge statistical disparities observed between rates of hiring of men and women. He argued that by accepting the defendant's contentions, the majority had placed an unreasonably heavy burden on the plaintiffs by requiring them to "disprov[e] the enormous significance that Sears attributes to unmeasurable variables." Id. 839 F.2d at 363 (emphasis in original).

We agree with the dissent and reject the approach taken by the Sears majority which places a very heavy — and possibly insurmountable — burden on the plaintiff with respect to establishing the probativeness of proffered statistical data. See id. at 363, 365. Moreover, the Sears majority opinion is at odds with the holdings of the District of Columbia, Second, and Eighth Circuits, which we find to be more persuasive. We also find those holdings to be more in keeping with the Supreme Court's reasoning in Bazemore, and with the general rules of procedure held by the Supreme Court to be most consistent with the purpose of Title VII. We therefore adopt the reasoning in those cases, i.e., that the defendant cannot rebut an inference of discrimination by merely pointing to flaws in the plaintiff's statistics.

We note that our approach to regression analyses in Title VII litigation does not conflict with this circuit's earlier treatment of such statistical evidence in Penk v. Oregon State Bd. of Higher Educ., 816 F.2d 458 (9th Cir. 1987). In Penk, women faculty members brought a Title VII action challenging the salary, promotion, and tenure practices of the Oregon state system of higher education. In support of their claims, plaintiffs offered multiple regression analyses to show a pattern and practice of discrimination "at every institution in every year studied in pay, rank, and tenure decisions." Id. at 464. Defendant Oregon State Board of Higher Education rebutted plaintiffs' proffered statistical evidence by offering its own statistical evidence and by challenging the "accuracy and importance" of plaintiffs' evidence. Id. While the district court admitted plaintiffs' multiple regression evidence, it later chose to discount the weight of that evidence during its deliberations following the bench trial. In its memorandum decision, the court held that plaintiffs' statistical analyses omitted or inadequately represented several critical decision-making variables, such as teaching quality, community and institutional service, and quality of research and scholarship. Id. at 465. In concluding that the plaintiffs had failed to establish intentional discrimination, the court emphasized that these missing variables "must have had a significant influence on salary and advancement decisions." Id.

Our decision in Penk is easily distinguished from this case because the statistical flaws in Penk were of a substantially different character than the alleged flaws in the EEOC's regression analyses. In Penk, the plaintiffs' regression analyses failed to account for arguably the most critical factors in academic promotion and tenure decisions — the subjective variables of teaching quality, community and institutional service commitment, and research and scholarship quality. Id. at 464 (academic and professional employment advancement decisions properly include a "high regard for subjective personnel qualities and characteristics"); see also Zahorik v. Cornell University, 729 F.2d 85, 92-94 (2nd Cir. 1984). Thus, it is clear that the statistical omissions in Penk were so central to academic employment decisions that the defendants, by merely pointing out such omissions, could defeat any inference of discrimination. See Sheehan, 839 F.2d at 103; Palmer, 815 F.2d at 101. The alleged flaws in the EEOC's regression analyses in this case, on the other hand, concern the failure to account for gender-based differences in career interests. While failure to include such employment interests may render the regression analyses less precise, merely pointing to such an imperfection does not, without more, defeat a showing of intentional discrimination established by the regression analyses. See Catlett, 828 F.2d at 1266 (mere assertion by defendant that plaintiff's statistical analysis failed to account for gender-based interest differences in highway maintenance positions did not defeat inference of discrimination; employer bore burden of introducing evidence to show omission was significant).

2.

The record in this case shows that the district court did not properly assess the probative value of the EEOC's evidence in light of GenTel's attacks upon it. Once the EEOC had introduced its statistical studies and other evidence, GenTel countered in two ways: (1) by attacking the EEOC's statistics; and (2) by offering its own evidence of nondiscrimination.

a.

In its Findings of Fact, the district court adopted GenTel's assertions and made a generalized finding that the EEOC's statistical analyses failed to establish impermissible discrimination because they contained various "critical flaws" — the most significant being the failure to adequately analyze differences in job interest between men and women and how this affects job placement. However, as noted above, GenTel was required to do more than simply point out possible flaws in the EEOC's data to overcome the statistical showing of discrimination.

The EEOC offered regression analyses which showed a disparity in hiring that correlates with sex, while controlling for the major legitimate factors, raising a strong inference of discrimination. GenTel cannot defeat that showing of discrimination simply by pointing out possible flaws in EEOC's data. Rather, GenTel had to produce credible evidence that curing the alleged flaws would also cure the statistical disparity — proof which GenTel did not offer. Thus, we hold that the district court erred in uncritically accepting GenTel's assertion that the plaintiff's analyses were flawed where GenTel had failed to show that if the EEOC had "adequately" accounted for the alleged flaws, the disparities in its analyses would have been eliminated. See Sobel, 839 F.2d at 34; Catlett, 828 F.2d at 1266; Palmer, 815 F.2d at 101; Segar, 738 F.2d at 1287. The court's error is harmless, however, unless it also erred in its conclusion that GenTel's own evidence overcame any inference of discrimination raised by the EEOC's evidence.

b.

GenTel's affirmative defense consisted of three types of evidence: its own regression analyses; testimonial and documentary evidence of its equal opportunity efforts; and statistics comparing the percentages of women in craft positions in the Northwest and in the entire United States. The district court found that GenTel's statistics established a correlation between job interests and placement sufficient to dispel any inference of discrimination created by the EEOC's analyses. The court rendered this determination because although it acknowledged that GenTel's statistical data had limited probative value by itself, its conclusions "were reinforced by other testimony and exhibits received at trial."

That "other testimony and exhibits," however, consisted in primary part of the self-laudatory evidence of GenTel's equal employment opportunity policies and programs — the admission of which we have held was error. With only a one-sided view of GenTel's equal employment opportunity efforts, it is not surprising that the district court found the conclusions of GenTel's questionable statistical analyses to be corroborated.

Other evidence which the court found substantiated GenTel's statistical conclusions showed that GenTel's record of employing women compared favorably to figures for the Northwest and the nation as a whole. The court did not rely heavily on this evidence, however, because it recognized that while offering some insight, the comparative evidence "do[es] little to assist us in understanding whether women were precluded from higher paying jobs at General Telephone. . . ."

As the district court itself recognized, GenTel's non-equal opportunity evidence has limited probative value. Thus we are unable to agree with the court's conclusion that GenTel overcame any inference of discrimination created by the EEOC's evidence, when considered in light of only the properly admitted evidence.

C.

In its cross-appeal, GenTel alleges that the charge of one of the named plaintiffs, Terry Freeman, was not timely filed with the EEOC and therefore cannot serve as a foundation for jurisdiction. GenTel argues that the EEOC thus failed to comply with certain jurisdictional requirements necessary for the court to try those EEOC claims which involved allegations of gender discrimination in the promotion of hourly-based workers in several job classifications. The EEOC contends that the Freeman charge was in fact timely filed, and even if it was not, there were other charges which provide adequate bases for the court's jurisdiction over the claims. Although this claim was made below in a motion to dismiss and renewed in the pretrial conference order, the district court never made a determination as to the timeliness of the Freeman charge.

A determination, however, must be made on this question before a decision can be made as to the merits of these particular claims. Because the record is inadequate for this court to assess the timeliness of the Freeman charge, we remand to the district court to make the proper inquiries and render the necessary determination as to whether the jurisdictional requirements of the relevant causes of action have been met.

CONCLUSION

The district court failed to assess properly the probative value of the EEOC's statistical evidence, in light of GenTel's opposing evidence. We cannot determine how the district court would have ruled with regard to the EEOC's ultimate burden of persuasion if it had accorded proper treatment to the statistical evidence. In addition, the district court erred in admitting the affirmative action evidence after denying discovery of critical self-assessments.

Accordingly, our previous disposition is withdrawn and vacated, the decision of the district court is REVERSED and the case REMANDED to the district court for a new trial.

REVERSED AND REMANDED.