0 & G Spring and Wire Forms Specialty Company (0 & G) is a small company on Chicago’s West Side that manufactures springs and specialty wire forms to order. The shop has about 50 workers, about 35 of whom work at low-skilled jobs in the “secondary department” operating kick and punch presses. 0 & G recruited for these positions by word-of-mouth and by accepting applications from walk-ins off the street, although walk-ins were only considered during certain hiring “windows.” From 1979 until 1985, 0 & G hired 87 people for the secondary department, 58 of whom were walk-in applicants. None of the 87 hires were African-Americans. On November 27, 1985, the Equal Employment Opportunity Commission (EEOC) filed a complaint charging 0 & G with engaging in a pattern and practice of racial and age discrimination in recruiting and hiring in violation of Title VII of the Civil Rights Act of 1964, 42 U.S.C. § 2000e-6(a) (1988), and the Age Discrimination in Employment Act of 1967 (ADEA), 29 U.S.C. § 621 et seq. (1988).1 After a bench trial and subsequent amended opinions, the district court found that 0 & G had engaged in a pattern of intentional discrimination against African-Americans. 705 F.Supp. 400 (N.D.I11.1988).
I. Background
The EEOC alleged that, from 1979 until 1985, O & G intentionally discriminated against African-Americans and persons over age 40 in recruitment and hiring for the *875secondary department, and also that 0 & G’s word-of-mouth recruiting technique had a disparate impact on African-Americans and workers over 40.2 At trial, the EEOC presented evidence that African-Americans represented the largest group in the walk-in pool during the relevant time period, although 0 & G claimed that these individuals did not request applications. Four African-American witnesses also provided anecdotal evidence that they had been rejected for jobs at 0 & G.3
But the heart of the ease was the statistical evidence presented by both parties to calculate how many African-Americans should have been hired based on the relevant labor market serving 0 & G. The EEOC expert, Dr. Pierre de Vise, concluded that African-American availability in the relevant labor market ranged from 22.5% to 31%. Dr. de Vise opined that these figures probably underestimated the African-American availability, since they excluded unemployed people, and African-Americans were overrepresented among the unemployed. 0 & G’s expert, Dr. Glen Meyers, testified that Dr. de Vise’s determinations of African-American availability were inaccurate. Dr. Meyers testified that Dr. de Vise failed to consider 0 & G’s preference for skilled machinists as workers or workers’ own self-selection. Dr. Meyers also testified that 0 & G had legitimate business justifications for relying on wórd-of-mouth and walk-in recruiting and hiring.
The district court found 0 & G liable for violations of Title VII on both the intentional discrimination and disparate impact claims, but found 0 & G not liable on the ADEA claim. The district court agreed with 0 & G that it was reasonable to expect that 0 & G’s work force might be disproportionately composed of recent immigrants, and reasonable for 0 & G to prefer experienced applicants. 705 F.Supp. at 406. But even so, the district court concluded that “no explanation is sufficient to overcome the ‘inexorable zero’ employment of blacks at 0 & G from 1979 though 1985.” Id. African-American availability in the relevant labor market would have tó be extremely low — 6%—to account for no African-American hires and, even if the relevant market had never been precisely defined, the EEOC had proven that availability was in fact significantly higher. Id. The district court amended its decision on May 22, 1989 to clarify its findings.
On January 26, 1990 the district court reconsidered its decision in light of Wards Cove Packing Co. v. Atonio, 490 U.S. 642, 109 S.Ct. 2115, 104 L.Ed.2d 738 (1989), and held that the disparate impact claim failed since the EEOC had not proven that O & G’s word-of-mouth recruiting served no legitimate business purpose. 732 F.Supp. 72 (N.D.Ill.1990). Focusing only on O & G’s walk-in hiring, the district court reaffirmed its decision that O & G was liable for intentional discrimination. The district court again reconsidered its decision on March 20, 1992 after this court decided EEOC v. Chicago Miniature Lamp Works, 947 F.2d 292 (7th Cir.1991), in which we reversed a lower court’s finding of intentional discrimination because the EEOC had not offered sufficient proof of intent. The district court found the two cases dissimilar, and found no reason to disturb its prior finding.4 In later proeeed-*876ings, the district court awarded back pay to be paid pro rata to a class of applicants or would-be applicants to 0 & G. 790 F.Supp. 776 (N.D.I11.1992). 0 & G appeals the district court’s finding of liability on the Title VII claim, and the denial of its motion for attorneys’ fees for the ADEA claim.
II. Sufficiency of evidence
0 & G contends that there was insufficient evidence for the district court to find that it engaged in a pattern and practice of intentional discrimination, arguing primarily that statistical evidence alone cannot prove intentional discrimination. The district court squarely held the opposite, noting that “it is beyond cavil that statistical proof alone is sufficient to establish liability under both the disparate treatment and disparate impact of models.” 705 F.Supp. at 408 (emphasis in original). 0 & G’s arguments to the contrary have been firmly rejected by the Supreme Court and this circuit. Int’l Brotherhood of Teamsters v. United States, 431 U.S. 324, 339-40 n. 20, 97 S.Ct. 1843, 1857 n. 20, 52 L.Ed.2d 396 (1977) (“Since the passage of the Civil Rights Act of 1964, the courts have frequently relied upon statistical evidence to prove a violation ... in many cases the only available avenue of proof is the use of racial statistics to uncover clandestine and covert discrimination by the employer.”); Hazelwood School Dist. v. United States, 433 U.S. 299, 307-08, 97 S.Ct. 2736, 2741, 53 L.Ed.2d 768 (1977); Chicago Miniature, 947 F.2d at 299 (“Statistical evidence of disparities between minority representation in an employer’s work force and minority representation in the community from which employees are hired can prove disparate treatment in a pattern and practice case.”); EEOC v. Sears, Roebuck & Co., 839 F.2d 302, 311 (7th Cir.1988). Reliance on statistical evidence by no means diminishes the plaintiff’s obligation to prove discriminatory intent — but in some cases, statistical disparities alone may prove intent. “Statistics showing racial or ethnic imbalance are probative in a case such as this one only because such imbalance is often a telltale sign of purposeful discrimination; absent explanation, it is ordinarily expected that nondiscriminatory hiring practices will in time result in a work force more or less representative of the racial and ethnic composition of the population in the community from which employees are hired.” Teamsters, 431 U.S. at 339-40 n. 20, 97 S.Ct. at 1857 n. 20.5 Appropriate statistical evidence can also be sufficient to establish a pattern and practice of discrimination — or in other words, that the alleged discriminatory hiring practices were “more than the mere occurrence of isolated or ‘accidental’ or sporadic discriminatory acts.” Teamsters, 431 U.S. at 336, 97 S.Ct. at 1855; Hazelwood, 433 U.S. at 308, 97 S.Ct. at 2742.
But, of course, statistics, like any evidence, are not irrefutable; strong statistics may prove a case on their own, while shaky statistics may be insufficient unless accompanied by additional evidence. Teamsters, 431 U.S. at 340, 97 S.Ct. at 1857; Chicago Miniature, 947 F.2d at 300-01. O & G maintains that the EEOC’s statistical evidence was fatally flawed in several respects. Determining the validity and value of statistical evidence is firmly within the discretion of the district court, and we will reverse its findings only if they are clearly erroneous. Pullman-Standard v. Swint, 456 U.S. 273, 287-90, 102 S.Ct. 1781, 1789-91, 72 L.Ed.2d 66 (1982). In this respect, “a statistical model [in a discrimination case] need not be completely specified and some arguably relevant variables can be omitted in certain cases.” Chicago Miniature, 947 F.2d at 300; Bazemore v. Friday, 478 U.S. 385, 400, 106 S.Ct. 3000, 3009, 92 L.Ed.2d 315 (1986).
Determining the relevant labor market is an essential step in determining whether there are any statistically significant *877deviations between the market and the employer’s hiring patterns. Chicago Miniature, 947 F.2d at 299-800. The EEOC’s expert Dr. de Vise provided three different formulations of 0 & G’s relevant labor market, each with a slightly different geographic or occupational scope.6 The district court did not choose among the three models, but found that they presented a sufficiently accurate range against which the district court could evaluate 0 & G’s hiring record. Under the most conservative definition, African-American availability was at least 22.5%. The district court found that “the statistical probability using standard deviation analysis of no black hires during the period 1979 through 1985 was infinitesimal.” 732 F.Supp. at 74.
0 & G argues that the EEOC statistics do not account for applicant self-selection. Selection bias on the part of the participants in the otherwise-relevant labor market can, of course, account for disparities in the number of people who walk by a shop and the number who walk in. See Chicago Miniature, 947 F.2d at 302, and cases cited therein. But the district court did factor in 0 & G’s claim that its relevant labor market would be disproportionately composed of recent Polish immigrants, since 0 & G offered poor working conditions and low pay, but, as a compensating factor, did not require English. 705 F.Supp. at 406; see also 790 F.Supp. at 779 (district court noted that it considered the language factor). Even this, however, could not account for the absence of any African-American hires, particularly in light of testimony from 0 & G staff and application data indicating that African-Americans represented about 20% of the walk-in applicant pool. 732 F.Supp. at 74. Moreover, 0 & G incorrectly states (and the dissent repeats several times) that the district court found that African-Americans preferred not to work at O & G because its workers spoke Polish or Spanish. Not surprisingly, no such finding of language-based selection bias on the part of African-Americans — nor any shred of evidence that would support such a finding— exists in the record. The dissent’s conclusion that the presence of Polish and Spanish speakers at O & G would, on its own, drive away African-Americans is entirely unfounded, and suggests ethnic attitudes without support in the evidence. It would certainly be improper to impute such biases on appeal. In any event, to make such a variable meaningful, O & G would have to show that African-Americans exhibited this preference in significantly greater proportion to other native-born English speakers. The district court thus considered applicant self-selection, but concluded that, to the extent it was a valid factor, it did not significantly reduce the statistical disparities.
O & G’s expert also testified that, as a small company, O & G could not afford to train new hires, and thus needed to hire experienced workers. The district court disagreed that jobs in the secondary, department required experience, but found that it would be a desirable qualification. O & G contends that the EEOC statistics ignore this preference for skilled workers.7 The *878EEOC statistics were based upon three occupational categories — punch and stamp press operators, machine operators in general and operators — that 0 & G’s own expert conceded were appropriate. This situation contrasts, for example, with the analysis this court found unpersuasive in Chicago Miniature, in which the EEOC presented only a simplistic analysis based on race and the entire Chicago region. 947 F.2d at 301. In any event, even though 0 & G’s work does not require high skill, 0 & G still argues that these occupational categories included too many unskilled workers. But the district court factored in 0 & G’s preference for experienced workers, and still concluded that the disparity between the market and 0 & G’s hires was too great. The district court also considered 0 & G’s argument that it usually hired unemployed workers, and that the EEOC statistics considered only skilled employed workers. This factor could cut in different ways, since the EEOC presented evidence that African-American unemployment was higher than white unemployment. In the end, 0 & G’s failure to hire African-Americans would only be justified in statistical terms if African-American availability were 6% of the labor market — or was discounted by a factor of nearly three from the most conservative EEOC estimate in this case.
The district court thoroughly weighed all of the objections raised here and in the dissent against the EEOC’s statistical evidence. The district court candidly pointed out the areas that were not mathematically provable. But even the use of the most forgiving variables could not reduce the calculation of African-Americans in the relevant labor market to a level that would account statistically for 0 & G’s failure to hire any African-Americans. These are precisely the sort of determinations that we review care-fully but will not disturb unless we are convinced the district court has erred. And 0 & G fails to note that the district court did not rely exclusively on the statistical evidence, but considered it in tandem with the EEOC’s anecdotal evidence. We are led to the same conclusion as the Court in Teamsters that “[i]n any event, fine tuning of the statistics could not have obscured the glaring absence of minoritfies].... The company’s inability to rebut the inference of discrimination comes not from a measure of statistics but from ‘the inexorable zero.’ ” Teamsters, 431 U.S. at 342 n. 23, 97 S.Ct. at 1858 n. 23.8
III. Burden of proof
0 & G also argues that it was subject to an erroneously high burden of proof. 0 & G relies on numerous statements in which the district court pointed out that 0 & G’s evidence was insufficient to outweigh the EEOC’s evidence of intentional discrimination. The district court did err to the extent that it implied that 0 & G was required to refute the EEOC evidence statistic-by-statistic, or offer evidence just as compelling as that presented by the EEOC. Instead, the employer in a disparate treatment case may refute the EEOC’s evidence by showing that it is inaccurate or insignificant. Sears, 839 F.2d at 308. There was never any question about whether 0 & G met any intermediate burdens of production. Addressing 0 & G’s complaints about burden-shifting, the district court noted “the net result was not that 0 & G’s attempt to rebut was legally insufficient. This does not shift the burden of proof to 0 & G; it only means that the prima facie statistical case presented by EEOC was strong enough to overcome 0 & G’s rebuttal by a preponderance of the evidence.” Order of May 22, 1989.
And in any event, after a full trial on the merits, the intermediate burdens are *879of no consequence: we look only to whether “the plaintiffs have demonstrated a pattern or practice of discrimination by a preponderance of the evidence. This is because the only issue to be decided at that point is whether the plaintiffs have actually proved discrimination.” Bazemore, 478 U.S. at 398, 106 S.Ct. at 3007; United States Postal Service Board of Governors v. Aikens, 460 U.S. 711, 714-16, 103 S.Ct. 1478, 1481-82, 75 L.Ed.2d 403 (1983); Hybert v. Hearst, 900 F.2d 1050, 1054 (7th Cir.1990). O & G contends that the district court nonetheless placed an erroneously high ultimate burden of proof on O & G. Although the district court initially believed that O & G carried a burden of persuasion, the district court recognized this error and reevaluated the evidence in its January 26,1990 decision following Wards Cove. The district court recognized that the EEOC carried the burden of persuasion, cf. St. Mary’s Honor Center v. Hicks, — U.S. —, 113 S.Ct. 2742, 125 L.Ed.2d 407 (1993), and weighed the evidence accordingly. O & G points to numerous occasions on which the district court weighed the evidence as examples of an allegedly improper shift in the burden of persuasion. But if this equation held true, any weighing of evidence would reflect an improper shift of the burden of proof, and the district court apparently could never rely on deficiencies in a defendant’s proof. In our view, the district court did not improperly shift any burden to O & G, but properly concluded that the EEOC had proven its case. The district court flagged the bottom line in the case: the “inexorable zero.” In light of the findings that there was significant African-American participation in O & G’s relevant labor market, the district court did not err in concluding that the EEOC had proven its ease by a preponderance of the evidence.
IV. Remedy
Before deciding on the remedy, the district court granted the EEOC leave to publish notice seeking to identify victims of O & G’s discrimination so that the court could fashion an appropriate method of awarding back pay. O & G objected to publication, arguing that the court should first determine whether back pay ought to be awarded, and if so, for what period of time and in what amount. But the district court determined that it needed to know the number of claimants from the outset to give it an idea whether individualized remedies would be feasible, or whether class-wide relief was in order. Stewart v. General Motors Corp., 542 F.2d 445, 452 (7th Cir.1976), cert. denied, 433 U.S. 919, 97 S.Ct. 2995, 53 L.Ed.2d 1105 (1977); Chicago Miniature, 947 F.2d at 300.
The EEOC published a notice addressed to African-Americans who applied or would have applied to O & G for low-skilled jobs from 1979 until 1985. Four hundred and fifty claimants responded. On appeal, O & G argues that the district court should not have allowed the EEOC to publish a class notice before determining the other elements of an appropriate remedy. O & G contends that the notice might reach too broadly and would generate too many false claims to entitlement to relief. But O & G does not claim that such an approach was unlawful or constituted an abuse of the district court’s equitable powers to fashion a remedy, Albemarle Paper Co. v. Moody, 422 U.S. 405, 415-21, 95 S.Ct. 2362, 2370-73, 45 L.Ed.2d 280 (1975). O & G does not request reversal or a remand on this point, but only reimbursement for its expenses and fees “associated with the remedy phase.” But, if the court did not abuse its discretion, we will not reject its choice of remedy. O & G points to no error, and given the district court’s wide discretion to fashion a Title YII remedy, Franks v. Bowman Transportation Co., 424 U.S. 747, 763, 96 S.Ct. 1251, 1263, 47 L.Ed.2d 444 (1976), we do not conclude that the district court abused its discretion by first determining the size of the class and then proceeding to fashion the remedy.
The district court then determined that O & G owed $378,754 in back pay, including benefits and prejudgment interest.9 *880The $378,754 would be paid pro rata to all eligible class members. For the next four years, 0 & G also would be required to fill its new positions with at least equal numbers of class members and non-class members. 0 & G argues that the district court should not have awarded back pay but should have considered alternatives such as establishment of an affirmative action program. But back pay is presumptively appropriate in a Title VII case. Albemarle, 422 U.S. at 421, 95 S.Ct. at 2373; Horn v. Duke Homes, 755 F.2d 599, 606 (7th Cir.1985). It may only be denied for reasons which “if applied generally, would not frustrate the central statutory purposes of eradicating discrimination ... and making persons whole.” Albemarle, 422 U.S. at 421, 95 S.Ct. at 2373. If, for example, the back pay award would bankrupt 0 & G and render it judgment-proof, it might be within the district court’s discretion to alter the award. But here, the district court concluded that 0 & G could afford to pay the entire judgment, a conclusion 0 & G does not dispute on appeal.
0 & G also argues that the district court improperly extended the back pay period. But the court only calculated back pay due from October 22, 1982 (two years before the filing of a charge with the Commission, as required by 42 U.S.C. § 2000e-5(g)); see also Patterson v. Youngstown Sheet & Tube Co., 659 F.2d 736, 740 (7th Cir.), cert. denied, 454 U.S. 1100, 102 S.Ct. 674, 70 L.Ed.2d 641 (1981). The district court included in its calculations individuals who had applied to O & G as early as 1979, since they might still have been employed in 1982. But these individuals could recover only for back pay lost after 1982, not from the date of hire. The district court was clearly entitled to include all victims in the remedy and properly limited recovery in accordance with the two year limitation of Title VII. 42 U.S.C. § 2000e-5(g).
V. Attorneys’ fees
In its December 14, 1988 ruling, the district court found in favor of O & G on the ADEA claim. O & G moved for attorneys’ fees under the Equal Access to Justice Act (EAJA), 28 U.S.C. § 2412 (1988). After a hearing, the district court denied the motion from the bench on November 23, 1992, finding that the EEOC’s position was substantially justified. O & G appeals. The EEOC argues that the EAJA’s substantial justification standard, 28 U.S.C. § 2412(d), does not apply to an ADEA ease, but that even if it does, the EEOC’s position was substantially justified. The question of what statute applies has not been squarely addressed in this circuit, so we first consider whether the *881EAJA applies, and if so, whether the district court erred in denying 0 & G attorneys’ fees.
A.
Under the American rule, parties generally bear their own costs unless Congress specifically provides otherwise. Alyeska Pipeline Co. v. Wilderness Society, 421 U.S. 240, 247, 95 S.Ct. 1612, 1616, 44 L.Ed.2d 141 (1975). But, prior to passage of the EAJA in 1980, even where fee-shifting statutes allowed for costs, the federal government retained its common law immunity from suit unless the fee-shifting statute or some other statute explicitly allowed recovery from the United States (such as Title II of the Civil Rights Act, 42 U.S.C. § 2000a-3(b)). 28 U.S.C. § 2412 (1975); Alyeska, 421 U.S. at 265-67, 95 S.Ct. at 1626. The EAJA significantly overhauled § 2412 to allow for recovery against the federal government. Section 2412(b) waived the government’s immunity from suit and provided that the United States would be hable for fees under the common law to the same extent as any other party. Section 2412(d) carved out an exception to the common law for most litigants, as defined in § 2412(d)(2)(B), holding the government to a higher standard:
Except as otherwise specifically provided by statute, a court shall award to a prevailing party other than the United States fees and other expenses ... incurred by that party in any civil action ... brought by or against the United States ... unless the court finds that the position of the United States was substantially justified or that special circumstances make an award unjust.
28 U.S.C. § 2412(d)(1)(A).
As the text makes clear, the EAJA is a default provision, and does not apply if another statute specifically provides for fees in the same situation as described in § 2412(d). For example, the EAJA does not apply to suits under Title VII. EEOC v. Consolidated Services Sys., 30 F.3d 58 (7th Cir.1994); EEOC v. Kimbrough Investment Co., 703 F.2d 98, 103 (5th Cir.1983). Under Title VII, prevailing defendants may recover fees only if the EEOC’s suit was frivolous. 42 U.S.C. § 2000e-5(k) (“the Commission and the United States shall be liable for costs [including attorney’s fees] the same as a private person”); Christiansburg Garment Co. v. EEOC, 434 U.S. 412, 422, 98 S.Ct. 694, 700, 54 L.Ed.2d 648 (1978) (awarding fees to prevailing Title VII defendants only if suit was frivolous).
The EEOC argues that EAJA § 2412(d) (the substantial justification standard) likewise does not apply to suits brought under the ADEA. Since the ADEA requires the court to award fees to a prevailing plaintiff but does not provide for recovery for prevailing defendants, argues the EEOC, a defendant may only recover under the common-law “bad faith” exception to the American rule, applied to the federal government by EAJA § 2412(b).
From a policy standpoint, the EEOC’s argument that the ADEA should be interpreted like Title VII with respect to fees makes some sense. The ADEA and Title VII share the same goal — elimination of discrimination in the workplace — and their substantive prohibitions are identical. Lorillard v. Pons, 434 U.S. 575, 584, 98 S.Ct. 866, 872, 55 L.Ed.2d 40 (1978). Thus cases construing Title VII are often persuasive authority when interpreting the ADEA. Hayden v. La-Z-Boy Chair Co., 9 F.3d 617, 619 (7th Cir.1993), cert. denied, — U.S.—, 114 S.Ct. 1371, 128 L.Ed.2d 47 (1994); Rogers v. Sugar Tree Products, Inc., 7 F.3d 577, 581 (7th Cir.1993); see also Trans World Airlines, Inc. v. Thurston, 469 U.S. 111, 121, 105 S.Ct. 613, 621, 83 L.Ed.2d 523 (1985); Lehman v. Nakshian, 453 U.S. 156, 163-64, 101 S.Ct. 2698, 2703, 69 L.Ed.2d 548 (1981). And the Supreme Court has consistently made clear that the fee-shifting provisions of the Civil Rights Act are integrally linked to advancing the substantive goal of eradication of discrimination — a goal shared by the ADEA. Newman v. Piggie Park Enterprises, Inc., 390 U.S. 400, 401-02, 88 S.Ct. 964, 966, 19 L.Ed.2d 1263 (1968) (Title II); Albemarle, 422 U.S. at 415, 95 S.Ct. at 2370 (Title VII). These substantive goals underlie Title VII's differential treatment of prevailing plaintiffs and prevailing defendants. Christiansburg, 434 U.S. 412, 98 S.Ct. 694; Fogerty v. Fantasy, Inc., — *882U.S. —, —, 114 S.Ct. 1023, 1028, 127 L.Ed.2d 455 (1994). The Court’s reasoning applies equally to the ADEA: “[T]he plaintiff is the chosen instrument of Congress to vindicate a ‘policy that Congress considered the highest priority ... [and] when a district court awards fees to a prevailing plaintiff, it is awarding them against a violator of federal law ... these policy considerations which support the award of fees to a prevailing plaintiff are not present in the case of a prevailing defendant.” Christiansburg, 434 U.S. at 418-19, 98 S.Ct. at 698-99. The Supreme Court has recently underscored that it is appropriate to consider the policies animating a fee-shifting statute when interpreting similar statutes. Fogerty, — U.S. at—, 114 S.Ct. at 1028; see also Stomper v. Amalgamated Transit Union Local 241, 27 F.3d 316, 320 (7th Cir.1994) (Cudahy, J., dissenting). We might therefore expect the ADEA, so substantively similar to Title VII, to follow a similar fee-shifting regime. See Donnelly v. Yellow Freight Sys., Inc., 874 F.2d 402 (7th Cir.1989), aff'd, 494 U.S. 820, 110 S.Ct. 1566, 108 L.Ed.2d 834 (1990) (ADEA fee-shifting provision interpreted like Title VII); Heiar v. Crawford County, Wis., 746 F.2d 1190 (7th Cir.1984), cert. denied, 472 U.S. 1027, 105 S.Ct. 3500, 87 L.Ed.2d 631 (1985) (ADEA fee-shifting provisions interpreted like 42 U.S.C. § 1988); but see Stom-per (counseling against drawing inferences from Title VII or § 1988).
But although Congress intended the ADEA to mirror Title VII in substance, it very explicitly adopted a different remedial scheme — a modified version of the Fair Labor Standards Act (FLSA), 29 U.S.C. § 201 et seq. Lorillard, 434 U.S. at 578-79, 98 S.Ct. at 869; 29 U.S.C. § 626(b).10 Section 7(b) of the ADEA incorporates by reference the FLSA fee-shifting provision, which provides “the court ... shall, in addition to any judgment awarded to the plaintiff or plaintiffs, allow a reasonable attorney’s fee to be paid by the defendant.” 29 U.S.C. § 216(b); 29 U.S.C. § 626(b). Although § 216(b) makes clear that the allowance of fees to a prevailing plaintiff is mandatory, Alyeska, 421 U.S. at 261 n. 34, 95 S.Ct. at 1624 n. 34, what of prevailing defendants? Either § 216(b) prohibits awarding fees to a prevailing defendant — which would preclude application of the EAJA — or it is silent on the matter— which means the EAJA or the common law steps in to fill the void. The language of the provision, with its emphasis on an award “to the plaintiffs” to be paid “by the defendants, ” strongly suggests by negative implication that it precludes awarding fees to a prevailing defendant. The ADEA, unlike the FLSA, was enacted against the backdrop of Title VII under which defendants may recover fees; the absence of any provision in the ADEA allowing for an award to defendants is worthy of note. Some courts have suggested, although somewhat ambiguously, that the FLSA might preclude awarding fees to a prevailing defendant. See Alyeska, 421 U.S. at 264 n. 37, 95 S.Ct. at 1625 n. 37 (“Congress ... has in some instances provided that either party could be given an award depending upon the outcome of the litigation and the court’s discretion ... while in others it has specified that only one of the litigants can be awarded fees. See ... Fair Labor Standards Act, 29 U.S.C. § 216(b).”); Christiansburg, 434 U.S. at 415-16 n. 5, 98 S.Ct. at 697 n. 5 (contrasting FLSA, which provides for a mandatory award to prevailing plaintiffs, with statutes that allow permissive awards to one party, or awards to either party); see also Richardson v. Alaska Airlines, Inc., 750 *883F.2d 763, 767 (9th Cir.1984) (prevailing defendants may not recover fees under ADEA); Mizrany v. Texas Rehabilitation Comm’n, 522 F.Supp. 611 (S.D.Tex.1981), aff'd, 685 F.2d 1384 (5th Cir.1982) (same). The Chris-tiansburg Court, rejecting the EEOC’s argument that Title VII defendants could only recover attorneys’ fees under the bad faith exception (the same argument the EEOC pursues here), noted that no statutory provision would have been necessary if Congress had intended this default rule to apply. 434 U.S. at 419, 98 S.Ct. at 699. But the Court also remarked “had Congress provided for attorney’s fees awards only to successful plaintiffs, an argument could have been made that the congressional action had pre-empted the common-law rule, and that, therefore, a successful defendant could not recover attorney’s fees even against a plaintiff who had proceeded in bad faith.” Id. at 419 n. 13, 98 S.Ct. at 699 n. 13.
But the negative implication of § 216(b) does not persuasively evince a Congressional intent to preclude application of the common law rule allowing recovery if a plaintiff acted in bad faith. Cf. Sullivan v. Hudson, 490 U.S. 877, 891-92, 109 S.Ct. 2248, 2258, 104 L.Ed.2d 941 (1989) (refusing to read provision of EAJA awarding fees following “adversary adjudication” to preclude recovery of fees following non-adversary adjudication); Krodel v. Young, 576 F.Supp. 390, 393 (D.D.C.1983) (congressional intent must be clear, although statute need not use words “attorneys fees”). Congress must be presumed to know, upon incorporating the FLSA into the ADEA, that in the absence of a specific provision, prevailing defendants would not be able to recover fees absent a showing of bad faith. By explicitly changing this rule with respect to plaintiffs but remaining silent with respect to defendants, the most sensible reading is that the FLSA and the ADEA adopt the common law rule with respect to prevailing defendants. We have already reached this conclusion about the FLSA. EEOC v. Kenosha Unified School Dist. No. 1, 620 F.2d 1220 (7th Cir.1984) (attorneys’ fees only available to prevailing defendant under FLSA under bad faith exception to American rule); see also Matthews v. Allis-Chalmers, 769 F.2d 1215, 1219 (1985), overruled on other grounds, 846 F.2d 457 (7th Cir.1988) (ADEA does not expressly preclude awarding costs to prevailing defendants under Fed.R.Civ.P. 54(d)). Several courts are in accord. See, e.g., Cova v. Coca-Cola, 574 F.2d 958, 962 (8th Cir.1978) (prevailing defendants under ADEA subject to American rule); Gray v. New England Telephone & Telegraph Co., 792 F.2d 251, 260 (1st Cir.1986) (same); Morgan v. Union Metal Mfg., 757 F.2d 792, 796 (6th Cir.1985) (same); Kreager v. Solomon & Flanagan, P.A., 775 F.2d 1541, 1542-43 (11th Cir.1985) (American rule applies to FLSA); EEOC v. Clay Printing, 13 F.3d 813 (4th Cir.1994) (EAJA applies to ADEA prevailing defendant); Donovan v. Dialamerica Marketing Co., 757 F.2d 1376, 1388-89 (3d Cir.1985), cert. denied, 474 U.S. 919, 106 S.Ct. 246, 88 L.Ed.2d 255 (no discussion of § 216(b), but applying EAJA to prevailing FLSA defendant); Steinberg v. St. Regis, 583 F.Supp. 421, 424-25 (S.D.N.Y.1984); Riley v. Resthaven Gardens of Memory, No. 90-1227-C, 1991 WL 261717, 1991 U.S. Dist. Lexis 17804 (D.Kan. Nov. 12, 1991).
Prior to enactment of the EAJA, private employers could recover attorneys’ fees under the FLSA if the plaintiffs allegations were frivolous or in bad faith, but the EEOC was immune. Kenosha Unified School Dish, 620 F.2d at 1228. With the passage of the EAJA this immunity has vanished, and the EEOC is liable under 28 U.S.C. § 2412(b) just as a private plaintiff is liable under common law. But while the EEOC concedes that EAJA § 2412(b) (the bad faith standard) applies to the ADEA, it maintains that the ADEA preempts application of § 2412(d) (the substantial justification standard). This distinction is untenable. The ADEA either preempts application of EAJA § 2412 or it does not. We have held that it does not, and thus nothing precludes application of all of § 2412, including the substantial justification standard of § 2412(d). Accord EEOC v. Clay Printing Co., 13 F.3d 813.
B.
O & G is therefore entitled to attorneys’ fees unless the court finds that the EEOC’s position in the litigation was sub*884stantially justified, or that special circumstances would make an award unjust. 28 U.S.C. § 2412(d)(1)(A). The parties argue at some length about the precise meaning of “substantially justified.” Obviously it requires a stronger showing than mere good faith, but does not require a showing that the government’s position was either correct or even “justified to a high degree.” Instead, the position must be “justified ... in the main — that is, justified to a degree that could satisfy a reasonable person.... [A] position can be justified even though it is not correct, and we believe it can be substantially (i.e., for the most part) justified if a reasonable person could think it correct, that is, if it has a reasonable basis in law and fact.” Pierce v. Underwood, 487 U.S. 552, 565-66 & n. 2, 108 S.Ct. 2541, 2550 & n. 2, 101 L.Ed.2d 490 (1988). The district court reviewed the EEOC’s evidence in the litigation, and ruled that 0 & G was not entitled to fees. We cannot disturb this ruling unless the record “commands the conclusion” that the district court was wrong. Id. at 570-71, 108 S.Ct. at 2552. Although the district court, ruling from the bench, did not facilitate review by clearly setting forth its reasoning, we cannot conclude that it abused its discretion by denying O & G attorneys’ fees.
At trial, the EEOC expert Dr. de Vise testified that O & G had hired only about half as many workers older than 40 than would be expected based on the relevant labor market. The district court found that Dr. de Vise’s analysis was severely flawed, comparing the age of new hires with the age of already employed workers in the relevant market. Under the proper analysis, O & G’s work force included even more older workers than would be expected based on the relevant market. But the EEOC had offered additional evidence to support its claim. Three people testified that they were over 45 and qualified to work at O & G, but had been rejected. The EEOC also introduced two letters written to the EEOC. The first letter was written by O & G’s attorney and stated “many of our machines do require a younger person because of the extensive effort required to operate the machines and move the necessary parts.” The second letter, written by O & G controller Richard Gregg, stated “because of these demanding aspects of secondary work, we prefer to hire younger males between 20-45 years old.” The EEOC argues that these letters alone, but certainly in conjunction with the other evidence, indicate that the EEOC had a reasonable basis for pursuing the ADEA claim.
O & G vigorously contests the probative value of this evidence. With respect to the statistical evidence, O & G points to the district court’s statements at the fees hearing indicating that it found the EEOC’s statistical case on the age claim extremely weak. But the district court, clearly aware of the weaknesses in the EEOC’s statistical case, did not conclude that the weakness rendered the government’s position untenable. O & G also claims that the anecdotal witnesses were not in fact qualified for work at O & G. But weighing the credibility of these witnesses was decidedly within the district court’s prerogative. Although the district court made no specific findings about this testimony, it does not appear to be significantly different from the anecdotal evidence the district court credited on the racial discrimination claim.
Lastly, O & G argues that the letters to the EEOC were repudiated by other testimony, and thus were an insufficient basis for the EEOC to proceed to trial. But the evidence may not have been as thoroughly discredited as O & G contends: although O & G’s attorney testified that he had no personal knowledge of O & G’s hiring preferences and that his letter had merely repeated what Gregg stated, Gregg admitted that he had approved both letters and written his own before he was aware that age discrimination was prohibited by law. Tr. 537, 509. O & G also contends that the EEOC knew before trial that O & G had in fact hired a sufficient number of workers to eliminate any statistical disparity. But again, the weight to be given such a factor, if true, is within the province of the district court. Moreover, the evidence must be evaluated as a whole, and on the basis of all of the evidence, we cannot conclude that the district court erred when it determined that the EEOC was justified in bringing its case. We therefore will not disturb the decision to deny attorneys’ fees.
*885VI.
We think it a rather dangerous suggestion of the dissent that recent immigrants, by clinging to their native tongues, somehow repel other minority applicants and, in effect, deprive them of normal consideration under the discrimination laws. For the reasons given above, the decisions of the district court are Affirmed.
. In the two years after the EEOC filed its complaint, 0 & G hired 12 new people in the secondary department, 5 of them African-Americans. 0 & G attributes the increase to demographic changes in the neighborhood, claiming that in 1985 the neighborhood “switched” from Polish to African-American. O & G also claims that the opening of a nearby market attracted African-Americans and increased foot traffic past O & G.
. The age discrimination claims are only relevant to the cross-appeal on attorneys’ fees, and the merits of the case will be discussed below.
. O & G claims that none of these witnesses applied during a "hiring window" and therefore their testimony is irrelevant. The district court made no finding on this matter, but did explicitly credit the EEOC's anecdotal evidence. 732 F.Supp. 72, 75 (N.D.I11.1990).
. In Chicago Miniature, we found that the EEOC's definition of the relevant labor market was flawed in significant respects. In particular, the geographic scope was defined too broadly and failed to account for the effect that commuting time might have on available labor. Also, the district court completely ignored language factors that might have skewed the applicant pool. Recognizing the heavy burden on a defendant to show that a district court's findings were clearly erroneous, we nonetheless found that in that case the statistics were too unreliable about important variables. 947 F.2d at 301. Although the Chicago Miniature analysis is useful here, the key facts of these two cases are different. Not only were the EEOC's calculations of the relevant labor market far more refined in this case, but the district court took account of language and other variables when evaluating the statistics. .
. This long line of cases was not weakened, as O & G suggests, by our recent decision in EEOC v. Consolidated Service Sys., 989 F.2d 233 (7th Cir.1993). There, we affirmed a district court’s finding that a small company had not engaged in intentional discrimination with its word-of-mouth hiring policy. The court specifically affirmed the potentially "decisive” value of statistical evidence, noting that "Consolidated had some non-Korean applicants for employment, and if it had never hired any this would support, perhaps decisively, an inference of discrimination.” Id. at 236-37.
. If the relevant labor market were defined as punching and stamping press machine operators in the City of Chicago (the narrowest of the relevant occupational and geographic categories in the 1980 census), then African-American availability for hire would be 22.5%. Under the broader occupational category for "machine operators, assemblers and inspectors,” but with a more narrowly focused geographic category based on considering O & G's applicants within a 5-mile radius, the African-American availability would be 23.1%. Under a third analysis, Dr. de Vise looked to employers' EEO-1 reporting forms that show the yearly African-American participation rate from 1980-85 of "operatives," defined as operators of "machine or processing equipment" or “other factory-type duties of intermediate skill level which can be mastered in a few weeks and require only limited training.” In O & G's 60651 zip code, the availability of African-American operatives would be 31.8%. Cf. Hazelwood, 433 U.S. at 309, n. 14, 97 S.Ct. at 2742, n. 14 (standard deviations greater than 2 are statistically significant); Castaneda v. Partida, 430 U.S. 482, 497 n. 17, 97 S.Ct. 1272, 1281 n. 17, 51 L.Ed.2d 498 (1977) (statistics for lawyers).
. The district court found such a preference reasonable, although the parties still dispute just how much of a role experience played in O & G's hiring decisions. O & G contended that almost half of its secondary department employees had previous technical training, Tr. 710, 1530, and most of the others had general factory experience, while the EEOC asserted that most of O & G's hires were unskilled and O & G vice president Joseph Olinyk had testified that jobs in the secondary department were "unskilled" and that *878an inexperienced person could learn how to operate the machines in an hour. Tr. 210.
. We reach the same conclusion with respect to O & G's contention that the EEOC did not adequately prove a "pattern or practice” of discrimination. The EEOC was required to prove a "regular, purposeful, less-favorable treatment of a protected group," King v. General Electric Co., 960 F.2d 617, 623 (7th Cir.1992); in other words, that discriminatory hiring was the standard operating procedure. Teamsters, 431 U.S. at 336, 97 S.Ct. at 1855. Statistics may prove such a pattern, id., and here they clearly did. The EEOC presented witnesses who testified that they did not receive jobs; the district court credited this testimony (importantly, unlike the district court in Consolidated).
. To reach this figure, the court first determined O & G’s hiring shortfall, or the number of African-Americans O & G would have hired absent discrimination. Since one-third of new hires were recruited by word-of-mouth, and at the beginning of the period in 1979 all secondary *880department employees were white, at first most word-of-mouth recruits would also have been white. But the walk-in applicants would be expected to mirror the makeup of the relevant labor market (as the applicant data shows that it did). Since 0 & G hired 58 walk-ins out of a labor market that at the most conservative estimate was 22.5% African-American, 13 hires should have been African-Americans (58 x 22.5%). As the African-American presence in 0 & G's work force increased through non-discrim-inatoiy walk-in hiring, however, it would be expected that more African-Americans would be recruited by word-of-mouth, as occurred in 1986-87. Thus if African-Americans had been hired at a uniform rate in 1979-1985, an average of 6.5 African-Americans would have been employed at 0 & G at any one time, constituting 14% of its work force. Assuming that 14% of the 29 people hired by word-of-mouth from 1979 to 1985 were African-American, 4 of the word-of-mouth hires should have been African-American. 0 & G's total hiring shortfall was therefore 17 (13 + 4).
The district court then determined the back pay award by multiplying the value of wages and fringe benefits actually earned by the 87 hires from 1979 to 1985 by the African-American hiring shortfall percentage (17 out of 87). This number was offset by the amount that those rejected by O & G earned or could have earned in other jobs during the period. § 706(g), 42 U.S.C. § 2000e-5(g). To hold individualized hearings for 451 applicants to determine what they had earned when they were not employed at O & G would have been unreasonable; moreover, given that O & G’s jobs were entry level jobs requiring little experience, it would be impossible to determine which specific class members would have been hired absent discrimination. The court determined that interim wages could be determined by presuming that the employment rate among African-American applicants or would-be applicants to O & G mirrored the African-American employment rate in the relevant labor market. The parties agreed that 22% of African-Americans in the relevant market were unemployed. In sum, the back pay award was determined by multiplying the total wages and benefits of the 87 hires times O & G's African-American shortfall percentage (17 out of 87) times 22%.
. Although other statutes have adopted the language of the FLSA fee-shifting provision, see Stomper, 27 F.3d at 318 n. 1, for some examples, the ADEA is notably explicit in its incorporation of FLSA's entire remedial scheme. The legislative history of the ADEA makes quite clear that "in enacting the ADEA, Congress exhibited both a detailed knowledge of the FLSA provisions and their judicial interpretation and a willingness to depart from those provisions regarded as undesirable or inappropriate for incorporation.” Lorillard, 434 U.S. at 581, 98 S.Ct. at 870. Senator Javits, who proposed applying FLSA to age discrimination and was a floor manager of the bill, made this explicit: "The enforcement techniques provided by [the ADEA] are directly analogous to those available under the Fair Labor Standards Act; in fact, [the ADEA] incorporates by reference, to the greatest extent possible, the provisions of the [FLSA].” 113 Cong.Rec. 31254 (1967), quoted in Lorillard, 434 U.S. at 582, 98 S.Ct. at 871. See also Fogerty, — U.S. ——, 114 S.Ct. 1023 (noting that ADEA expressly incorporated FLSA provisions; refusing to presume similar congressional intent for statute where legislative history was less clear).