Dukes v. Wal-Mart Stores, Inc.

Lead Opinion

Opinion by Judge HAWKINS; Concurrence by Judge GRABER; Dissent by Judge IKUTA; Dissent by Chief Judge KOZINSKI.

MICHAEL DALY HAWKINS, Circuit Judge:

Plaintiffs allege that Wal-Mart, Inc., discriminates against women in violation of Title VII of the Civil Rights Act of 1964. After detailed briefing and hearing, the district court certified a class encompassing all women employed by Wal-Mart at any time after December 26, 1998, and encompassing all Plaintiffs’ claims for injunctive relief, declaratory relief, and back pay, while creating a separate opt-out class encompassing the same employees for punitive damages. We affirm¹ the district court’s certification of a Federal Rule of Civil Procedure 23(b)(2) class of current employees with respect to their claims for injunctive relief, declaratory relief, and back pay. With respect to the claims for punitive damages, we remand so that the district court may consider whether to certify the cláss under Rule 23(b)(2) or (b)(3). We also remand with respect to the claims of putative class members, who no longer worked for Wal-Mart when the complaint was filed so that the district court may consider whether to certify an additional class or classes under Rule 23(b)(3).

BACKGROUND

Plaintiffs’ Third Amended Complaint,² brought on behalf of certain named plaintiffs and those similarly situated, asserts claims against Wal-Mart for sex discrimination under Title VII of the 1964 Civil Rights Act. Plaintiffs allege that women employed in Wal-Mart stores: (1) are paid less than men in comparable positions, despite having higher performance ratings and greater seniority; and (2) receive fewer — and wait longer for- — promotions to in-store management positions than men. Plaintiffs contend that Wal-Mart’s strong, centralized structure fosters or facilitates gender stereotyping and discrimination, that the policies and practices underlying this discriminatory treatment are consistent throughout Wal-Mart stores, and that this discrimination is common to all women who work or have worked in Wal-Mart stores.

Plaintiffs sought to certify a nationwide class of women who have been subjected to these allegedly discriminatory pay and promotion policies. The proposed class consists of women employed in a range of Wal-Mart positions, from part-time entry-level hourly employees to salaried managers. The class seeks injunctive and declaratory relief, back pay, and punitive damages, but not traditional “compensatory” damages. Plaintiffs proposed that the district court certify the following class pursuant to Rule 23:

All women employed at any Wal-Mart domestic retail store at any time since December 26, 1998 who have been or may be subjected to Wal-Mart’s challenged pay and management track promotions policies and practices.

Dukes v. Wal-Mart Stores, Inc., 222 F.R.D. 137, 141-42(N.D.Cal.2004).

After the parties had conducted extensive discovery and filed copious briefs, the district court heard oral argument. At the hearing, Wal-Mart emphasized the “historic” nature of Plaintiffs’ motion, including the size of the putative class, involving women employees at Wal-Mart’s 3,400 stores in 41 regions. The court acknowledged Wal-Mart’s concerns but noted that, while the class size was large, the issues were not unusual.³

District Court Proceedings

The district court issued an eighty-four-page order granting in part and denying in part Plaintiffs’ motion for class certification. See id. at 187-88. With respect to Plaintiffs’ claims for equal pay, the district court granted Plaintiffs’ certification motion as to issues of alleged discrimination and all forms of requested relief. With respect to Plaintiffs’ promotion claim, the court’s holding was mixed. The court certified the proposed class with respect to issues of alleged discrimination (including liability for punitive damages, as well as injunctive and declaratory relief); however, the court rejected the pro posed class with respect to the request for back pay, determining that data relating to the challenged promotions were not available for all class members. The court also exercised its discretion to provide for notice and an opportunity for employees to opt-out of the punitive damages portion of the class.

The Appeal

Pursuant to Rule 23(f), Wal-Mart appealed, contending that the district court erred by: (1) concluding that the class met Rule 23(a)’s commonality and typicality requirements; (2) eliminating Wal-Mart’s ability to respond to individual Plain tiffs claims; and (3) failing to recognize that Plaintiffs’ claims for monetary relief predominated over their claims for injunctive or declaratory relief. Plaintiffs cross-appealed from the district court’s limitation of back pay relief for many of Plaintiffs’ promotion claims.

DISCUSSION

I. Standard and Scope of Review

A district court’s decision regarding class certification is not only reviewed for abuse of discretion, Staton v. Boeing Co., 327 F.3d 938, 953 (9th Cir.2003), but also subject to “very limited” review, to be reversed “only upon a strong showing that the district court’s decision was a clear abuse of discretion,” Armstrong v. Davis, 275 F.3d 849, 867(9th Cir.2001) (internal quotation marks omitted). See also Millowitz v. Citigroup Global Mkts., Inc. (In re Salomon Analyst Metromedia Litig.), 544 F.3d 474, 480 (2d Cir.2008) (“When reviewing a grant of class certification, we accord the district court noticeably more deference than when we review a denial of class certification.”); Gonzales v. Free Speech Coal, 408 F.3d 613, 618 (9th Cir. 2005) (“Abuse of discretion is a highly deferential standard, under which the appellate court cannot substitute its view of what constitutes substantial justification for that of the district court; rather, the review is limited to assuring that the district court’s determination has a basis in reason.” (internal quotation marks omitted)); Blyden v. Mancusi, 186 F.3d 252, 269 (2d Cir.1999) (“A district court’s decision to certify a class is reviewed for abuse of discretion, and ‘[a] reviewing court must exercise even greater deference when the district court has certified a class than when it has declined to do so.’ ” (quoting Marisol A. v. Giuliani, 126 F.3d 372, 375 (2d Cir.1997) (per curiam))); Doninger v. Pac. Nw. Bell, Inc., 564 F.2d 1304, 1309(9th Cir.1977) (stating that “the judgment of the trial court should be given the greatest respect and the broadest discretion” (internal quotation marks omitted)).

A court abuses its discretion if its decision is premised on legal error. Hawkins v. Comparet-Cassani, 251 F.3d 1230, 1237 (9th Cir.2001). Moreover, the district court’s factual findings as to the applicability of Rule 23 criteria are entitled to the traditional deference given such determinations. See Local Joint Executive Bd. of Culinary/Bartender Trust Fund v. Las Vegas Sands, Inc., 244 F.3d 1152, 1161(9th Cir.2001).

Rule 23 “provides district courts with broad discretion to determine whether a class should be certified, and to revisit that certification throughout the legal proceedings before the court.” Armstrong, 275 F.3d at 871 n. 28. If evidence not available at the time of certification disproves Plaintiffs’ contentions that common issues predominate, the district court has the authority to modify or even decertify the class, Gen. Tel. Co. of Sw. v. Falcon, 457 U.S. 147, 160, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982) (“Even after a certification order is entered, the judge remains free to modify it in the light of subsequent developments in the litigation.”), or use a variety of management devices to address the individualized issues that have arisen, see 2 Alba Conte & Herbert B. Newberg, Newberg on Class Actions § 4.26(4th ed.2002).⁴

Our review is limited to whether the district court correctly selected and applied Rule 23 criteria. Bogus v. Am. Speech & Hearing Ass’n, 582 F.2d 277, 289 (3d Cir.1978); see also Waste Mgmt. Holdings, Inc. v. Mowbray, 208 F.3d 288, 295 (1st Cir.2000) (“An abuse occurs when a court, in making a discretionary ruling, relies upon an improper factor, omits consideration of a factor entitled to substantial weight, or mulls the correct mix of factors but makes a clear error of judgment in assaying them.”). Thus, if Plaintiffs demonstrate that they meet Rule 23’s requirements, they should be allowed to pursue their action as a class.

II. Standards For Class Certification Under Rule 23

A district court may certify a class only if:

(1) the class is so numerous that joinder of all members is impracticable;

(2) there are questions of law or fact common to the class; (3) the claims or defenses of the representative parties are typical of the claims or defenses of the class; and (4) the representative parties will fairly and adequately protect the interests of the class.

Fed.R.Civ.P. 23(a).

The district court must also find that at least one of the following three conditions is satisfied: (1) the prosecution of separate actions would create a risk of: (a) inconsistent or varying adjudications, or (b) individual adjudications dispositive of the interests of other members not a party to those adjudications; (2) the party opposing the class has acted or refused to act on grounds generally applicable to the class; or (3) questions of law or fact common to the members of the class predominate over any questions affecting only individual members, and a class action is superior to other available methods for the fair and efficient adjudication of the controversy. See id. 23(b).

The party seeking certification bears the burden of showing that each of the four requirements of Rule 23(a) and at least one requirement of Rule 23(b) have been met., Zinser v. Accufix Research Inst., Inc., 253 F.3d 1180, 1186 (9th Cir.), amended by 273 F.3d 1266 (9th Cir.2001).

The parties vigorously contest the standards governing the district court in finding the Rule 23 requirements satisfied. For a number of reasons, we must clarify this standard. First, the parties’ briefs demonstrate the degree of debate over the standard in this circuit. Second, a number of recent cases in other circuits have endeavored to clarify the standard, and we find it prudent to follow suit given evidence of confusion. Third, it is only very recently that any case in this circuit has interpreted language under the standard that drifts away from our clear case law, and we write to clarify our precedent. Fourth, we are not aware of a circuit that has detailed the issue in the Title VII context, and the typical situation presented in such a case implicates significant differences in the doctrine that require explanation to reach a resolution here and in future Title VII class action certification decisions.

A. Supreme Court Authority

The Supreme Court has provided clear, if sometimes misunderstood, guidance on the issue of what standards a district court applies when deciding whether to certify a class.

The leading case describing these standards is Falcon, where the plaintiff successfully certified a class of Mexican-Americans claiming racial discrimination. After a full trial on the merits, however, the district court’s findings distinguishing the named representative’s claims from those of the class called the propriety of continued class certification into question.⁵ On appeal, the Supreme Court held that because of the district court’s findings at trial, Falcon’s “complaint provided an insufficient basis for concluding that the adjudication of his claim of discrimination in promotion would require the decision of any common question concerning the failure of [General Telephone] to hire more Mexican-Americans.” Id. at 158., 102 S.Ct. 2364

The Court described the implications of its holding, noting that “[s]ometimes the issues are plain enough from the pleadings to determine whether the interests of the absent parties are fairly encompassed within the named plaintiffs claim, and sometimes it may be necessary for the court to probe behind the pleadings before coming to rest on the certification question.” Id. at 160, 102 S.Ct. 2364. In either case, the Court held, “a Title VII class action, like any other class action, may only be certified if the trial court is satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.” Id. at 161, 102 S.Ct. 2364. In conducting this rigorous analysis, the Court explained that “the class determination generally involves considerations that are enmeshed in the factual and legal issues comprising the plaintiffs cause of action.” Id. at 160, 102 S.Ct. 2364(internal quotation marks omitted).

Illustrating further, the Court approvingly cited Judge Godbold’s concurring opinion in the Fifth Circuit case that had announced the “across-the-board” rule the Court was reviewing in Falcon. Judge Godbold’s concurrence addressed the role of the district court in understanding the likely course of the litigation, and the Supreme Court praised his focus on “the need for more precise pleadings.” Id. at 160, 102 S.Ct. 2364(internal quotation marks omitted). The Court approved of his statement that “ ‘without reasonable specificity the court cannot define the class, cannot determine whether the representation is adequate, and the employer does not know how to defend.’ ” Id. at 161, 102 S.Ct. 2364(quoting Johnson v. Ga. Highway Express, Inc., 417 F.2d 1122, 1126 (5th Cir.1969) (Godbold, J., specially concurring)).

Falcon thus provides relatively straightforward guidance. When considering class certification under Rule 23, district courts are not only at liberty to, but-must, perform a rigorous analysis to ensure that the prerequisites of Rule 23(a) have been satisfied. See id. at 160-61, 102 S.Ct. 2364. It does not mean that a district court must conduct a full-blown trial on the merits prior to certification. A district court’s analysis will often, though not always, require looking behind the pleadings, even to issues overlapping with the merits of the underlying claims. See id. As we describe in more detail below, every circuit to have considered this issue, including our own previous decisions, has reached essentially the same conclusion: Falcon’s central command requires district courts to ensure that Rule 23 requirements are actually met, not simply presumed from the pleadings.

We also agree with the Second Circuit’s recent decision in Miles v. Merrill Lynch & Co. (In re Initial Pub. Offerings Securities Litigation) (“IPO ”), which explained that, to the extent lower courts have evidenced confusion regarding the Rule 23 standard after Falcon, this confusion has existed because those courts have misread a Supreme Court statement made eight years before the Court handed down Falcon. See 471 F.3d 24, 33-34 (2d Cir.2006). Specifically, courts have misunderstood Eisen v. Carlisle & Jacquelin, in which the Supreme Court stated, ‘We find nothing in either the language or history of Rule 23 that gives a court any authority to conduct a preliminary inquiry into the merits of a suit in order to determine whether it may be maintained as a class action.” 417 U.S. 156, 177, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974). “This statement has led some courts to think that in determining whether any Rule 23 requirement is met, a judge may not consider any aspect of the merits....” IPO, 471 F.3d at 33. It has “led other courts to think that a judge may not do so at least with respect to a prerequisite of Rule 23 that overlaps with an aspect of the merits of the case.” Id.

As the IPO court recognized, the distinguishing features of Falcon and Eisen are the purposes for which the certifying court is using the underlying facts — whether to address a merits issue unnecessarily or to determine whether, for example, the plaintiffs have demonstrated questions of law or fact common to their proposed class. See IPO, 471 F.3d at 32-35. An easy way to understand this distinction is to analogize to a familiar dispute over the admissibility of alleged hearsay evidence. Offered for the truth of the matter asserted, an out-of-court statement by someone not testifying is, of course, inadmissible hearsay. However, that same evidence used to impeach a witness may be properly admitted. So, too, with inquiries into facts overlapping with merits issues in the Rule 23 context. It is whether courts are using the facts to probe the plaintiffs’ claims of compliance with Rule 23, or to hear either parties’ claims directed to stand-alone merits issues, that renders a court’s use of the facts proper or improper. Courts have thus strayed from Falcon when they have myopically invoked Eisen to avoid considering facts properly relevant to the Rule 23 determination because the facts happen to be relevant to the later merits inquiry as well.

B. Case Law in Other Circuits

Like our decision today, and in part because of courts’ misunderstanding of Eisen, many of our sister circuits have recently been called upon to clarify the standard that a district court applies when deciding whether to certify a class under Rule 23.

Though IPO is the most notable of these decisions, other courts often trace the explanatory effort back to the Seventh Circuit’s decision in Szabo v. Bridgeport Machines, Inc., 249 F.3d 672 (7th Cir.2001). In Szabo, the court rejected a nationwide products liability class because the class was generally unsuitable for class-wide resolution because of choice of law problems on the state breach-of-warranty and fraud claims. Id. at 674. The Seventh Circuit also found that the district court below had “certified the class without resolving factual and legal disputes that strongly influence the wisdom of class treatment” because the district court had stated that the plaintiffs’ “allegations in the complaint are accepted as true for purposes of the class motion.” Id. at 675. Analogizing to the preliminary jurisdictional inquiries that district courts routinely make under Federal Rules of Civil Procedure 12(b)(1) and 12(b)(2), the court explained that, “[bjefore deciding whether to allow a case to proceed as a class action, therefore, a judge should make whatever factual and legal inquiries are necessary under Rule 23.” Id at 676.

The Second Circuit’s decision in IPO agreed with Szabo, providing what is now the leading case on the extent to which a district court must resolve Rule 23 issues that overlap with the merits of the case. IPO held that factual disputes concerning each of the Rule 23 factors must be analyzed and resolved. 471 F.3d at 41. This is a similar holding to our previous explanation — discussed in more detail below— that a district court must make “determinations” that the prerequisites of Rule 23(a) have been satisfied before it certifies a class, “which may require review of the same facts and the same law presented by review of the merits.” Falcon, 457 U.S. at 161, 102 S.Ct. 2364; Blackie v. Barrack, 524 F.2d 891, 897(9th Cir.1975). IPO explained that “the obligation to make such determinations is not lessened by overlap between a Rule 23 requirement and a merits issue.” 471 F.3d at 41. IPO expressly rejected the Second Circuit’s approach in Caridad v. Metro-North Commuter Railroad, 191 F.3d 283, 291-93 (2d Cir.1999), and Wal-Mart Stores, Inc. v. Visa U.S.A. Inc. (In re Visa Check/MasterMoney Antitrust Litigation) (“Visa Check”), 280 F.3d 124, 135 (2d Cir.2001), which had permitted class certification based on “some showing” that the Rule 23 factors were met, obviating the need to assess conflicting expert testimony pertinent to the Rule 23 inquiries. IPO, 471 F.3d at 40.

Since IPO, a number of other circuits have detailed the issue. The First Circuit has reviewed these appellate cases, noting that “[o]ur sister circuits agree that when class criteria and merits overlap, the district court must conduct a searching inquiry regarding the Rule 23 criteria, but how they articulate the necessary degree of inquiry ranges along a spectrum which suggests substantial differences.” Brown v. Am. Honda (In re New Motor Vehicles Canadian Export Antitrust Litig.), 522 F.3d 6, 24 (1st Cir.2008) (“New Motor Vehicles ”).

Though, of course, different circuits have used different words in articulating the review necessary, we think New Motor Vehicles overstates the degree of difference among the circuits. The core holding across circuits that have considered the issue is essentially unanimous: district courts must satisfy themselves that the Rule 23 requirements have been met before certifying a class, which will sometimes, though not always, require an inquiry into and preliminary resolution of disputed factual issues, even if those same factual issues are also, independently, relevant to the ultimate merits of the case. See, e.g., 1 Joseph M. McLaughlin, McLaughlin on Class Actions § 3:12 (6th ed. 2009) (“Consensus is rapidly emerging among the United States Courts of Appeal. The First, Second, Third, Fourth, Fifth, Seventh, Eighth, Tenth and Eleventh Circuits have expressly adopted certification standards that require rigorous factual review and preliminary factual and legal determinations with respect to the requirements of Rule 23 even if those determinations overlap with the merits.”).

A closer discussion of the cases the First Circuit has cited demonstrates the truly narrow range in which this “spectrum” actually exists. Requiring the district court to “make specific findings that each Rule 23 criterion is met” represents, the First Circuit has claimed, “around the more rigorous end of [the] spectrum,” and has been embraced by the Second, Fourth, Fifth, Seventh — and we would add Ninth — Circuits. New Motor Vehicles, 522 F.3d at 24 (citing IPO, 471 F.3d at 33, 41; Unger v. Amedisys Inc., 401 F.3d 316, 321-22(5th Cir.2005) (requiring courts to find facts favoring class certification through the use of “rigorous, though preliminary, standards of proof’); Gariety v. Grant Thornton, LLP, 368 F.3d 356, 366 (4th Cir.2004) (requiring that “the factors spelled out in Rule 23 ... be addressed through findings”); Szabo, 249 F.3d at 675-76); see Blackie, 524 F.2d at 897.

On the other end of this “spectrum,” according to the New Motor Vehicles court, are cases in “the Third and Eighth Circuits [which] sometimes require an inquiry into and preliminary resolution of disputes, but they do not require findings and do not hold that such inquiry will always be necessary.” New Motor Vehicles, 522 F.3d at 24 (citing Blades v. Monsanto Co., 400 F.3d 562, 566-67 (8th Cir. 2005); Newton v. Merrill Lynch, Pierce, Fenner & Smith, Inc., 259 F.3d 154, 166 (3d Cir.2001)).

In our review of these cases, we find this “spectrum” of certification standards narrower and more internally consistent than does the First Circuit. To begin with, we respectfully disagree with New Motor Vehicles'^ characterization of the Third Circuit’s approach. The case New Motor Vehicles cited, Newton, 259 F.3d at 174-77, only discussed a “presumption” of conformity with Rule 23 in the context of the fraud-on-the-market presumption in a securities class action.

The Third Circuit has since provided clarification. It has noted that Newton does not hold that a district court can presume Rule 23 requirements met from contested pleadings outside the fraud-on-the-market context. Vallies v. Sky Bank, 591 F.3d 152, 162-63 (3d Cir.2009). It has also demonstrated its conformity with the other circuits’ standard more broadly. In In re Constar International Inc. Securities Litigation, the court explained that “we require that each Rule 23 component be satisfied.” 585 F.3d 774, 780 (3d Cir. 2009) (emphasis added) (citing In re Hydrogen Peroxide Antitrust Litig., 552 F.3d 305, 310 (3d Cir.2009) (“Hydrogen Peroxide ”)). Recognizing the importance of the class certification decision, the court emphasized that “district courts, where appropriate, [are] to ‘delve beyond the pleadings to determine whether the requirements for class certification are satisfied.’ ” Id. (quoting Hydrogen Peroxide, 552 F.3d at 316; Newton, 259 F.3d at 167). It further elucidated, “[a]n overlap between a class certification requirement and the merits of a claim is no reason to decline to resolve relevant disputes when necessary to determine whether a class certification requirement is met.” Id. (internal quotation marks omitted). The Third Circuit thus requires a district court “determination” that the “requirements” of Rule 23 have been “satisfied.” Id. While we readily recognize the benefit of the additional cases available to us in conducting our review, we respectfully disagree with the First Circuit’s conclusion that the Third Circuit allows a more lax standard of review than do other circuits.

New Motor Vehicles is on firmer ground describing the Eighth Circuit’s decision in Blades, though we find the language in that case more nuanced than New Motor Vehicles describes, and we disagree that Blades “suggests substantial differences” from other circuits. New Motor Vehicles, 522 F.3d at 24. The Blades court cited Falcon in describing the standard as follows:

To determine whether common questions predominate, a court must conduct a limited preliminary inquiry, looking behind the pleadings. In conducting this preliminary inquiry, however, the court must look only so far as to determine whether, given the factual setting of the case, if the plaintiffs [sic] general allegations are true, common evidence could suffice to make out a prima facie showing for the class....

The preliminary inquiry at the class certification stage may require the court to resolve disputes going to the factual setting of the case, and such disputes may overlap with the merits of the case. See Szabo v. Bridgeport Machs., Inc., 249 F.3d 672, 676-77 (7th Cir.2001). Nonetheless, such disputes may be resolved only insofar as resolution is necessary to determine the nature of the evidence that would be sufficient, if the plaintiffs general allegations were true, to make out a prima facie case for the class.

Blades, 400 F.3d at 566-67 (some citations omitted) (emphasis added).

While this language is not as definitive as that used by the Second Circuit in IPO, for example, it sets up essentially the same standard. Under Blades, district courts in the Eighth Circuit must “resolve” factual disputes going to the “setting of the case,” which we understand to mean the factual circumstances dictating whether the plaintiffs have met the Rule 23 requirements. Id. at 567. Supporting this understanding is Blades’s later statement explaining that “[t]o certify a class action under Rule 23(b)(3), the Court mast find that: 1) common questions predominate ... and 2) class resolution is superior to other available methods for the fair and efficient adjudication of the controversy.” Id. at 569 (emphasis added). Finally, like IPO, Blades understands Eisen as prohibiting the district court from making preliminary findings on merits issues not related to the Rule 23 resolution. See id. at 566-67; see also Richard A. Nagareda, Class Actions in the Administrative State: Kalven and Rosenfield Revisited, 75 U.Chi. L.Rev. 603, 616 (2008) (“Nagareda, Class Actions ”). Not surprisingly, then, IPO cited Blades as a key case supporting its conclusion to require district court determinations that each of the Rule 23 requirements is met. IPO, 471 F.3d at 38.

We thus view whether an appellate court requires district courts to “resolve,” id.; “find,” Unger, 401 F.3d at 319-20; or “determine,” IPO, 471 F.3d at 40-41, that Rule 23 requirements have been met, as essentially, even if not precisely, the same standard.⁶ To characterize such usage as embodying “substantial differences” seems to create a distinction where none exists. See New Motor Vehicles, 522 F.3d at 24; see also Richard A. Nagareda, Class Certification in the Age of Aggregate Proof, 84 N.Y.U. L.Rev. 97, 113-14 (2009) (Nagareda, Aggregate Proof) (describing federal courts as moving toward an “essentially uniform” standard that is “now-settled law”). Under any of these articulations, the charge to the district court follows the analysis the IPO court explained, and we agree, that Falcon and Eisen require. IPO, 471 F.3d at 41. The district court must analyze underlying facts and legal issues going to the certification questions regardless of any overlap with the merits. However, this does not mean a district court should put the actual resolution of the merits cart before the motion to dismiss, summary judgment, and trial horses by reaching out to decide issues unnecessary to the Rule 23 requirements.

Having discussed our sister circuits’ treatment of this issue, we now turn to the standard in our circuit as established by our previous cases.

C. Ninth Circuit Precedent

Since we first addressed the issue in the wake of the Eisen and Falcon decisions, our cases have made clear that a district court inquiry overlapping with the merits is permissible, and often required, under Rule 23. See Moore v. Hughes Helicopters, Inc., 708 F.2d 475, 480 (9th Cir.1983) (“Although some inquiry into the substance of a case may be necessary to ascertain satisfaction of the commonality and typicality requirements of Rule 23(a), it is improper to advance a decision on the merits to the certification stage.” (citation omitted)). Later, we said that a district court is “at liberty to consider evidence which goes to the requirements of Rule 23 even though the evidence may also relate to the underlying merits of the case.” Hanon v. Dataproducts Corp., 976 F.2d 497, 509 (9th Cir.1992) (internal quotation marks omitted); see also Staton v. Boeing Co., 327 F.3d 938, 954 (9th Cir.2003) (“The court may not go so far, of course, as to judge the validity of [the merits] claims.... But the breadth and consistency of class counsel’s initial evidence places the district court’s finding of commonality well within that court’s discretion.” (citation omitted)). We have also specifically cited Eisen in rejecting a party’s argument that Eisen prohibits any inquiry overlapping with the merits and explaining that such an overlapping inquiry is sometimes necessary under Rule 23. See Blackie, 524 F.2d at 897(“Nor is the class issue separable from the merits in all cases (including this one). The common questions, typicality, conflicts and adequacy of representation, and predominance tests, are determinations ... which may require review of the same facts and the same law presented by review of the merits.” (citations omitted)).⁷

We have adhered to the Supreme Court’s guidance in holding that “[a] class may only be certified if we are ‘satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.’ ” Hanon, 976 F.2d at 509 (quoting Falcon, 457 U.S. at 161, 102 S.Ct. 2364). We have also, however, recognized that Falcon contemplated cases in which “the issues are plain enough from the pleadings” and do not require analysis beyond those papers. 457 U.S. at 160, 102 S.Ct. 2364; Chamberlan v. Ford Motor Co., 402 F.3d 952, 961 (9th Cir.2005) (per curiam) (“The Supreme Court thus recognized over twenty years ago that a rigorous analysis does not always result in a lengthy explanation or in-depth review of the record.”); Hanlon v. Chrysler Corp., 150 F.3d 1011, 1023 (9th Cir.1998). Thus, in some cases, the pleadings will be sufficient to render the certification decision, and in others, the district court must make determinations on facts overlapping with the merits.

In addition, our cases have understood that Falcon, like Rule 23 itself, requires “questions of law or fact that were common to the claims.” 457 U.S. at 158, 102 S.Ct. 2364 (emphasis added). Therefore, we have found class certification proper where a “plaintiff is attacking defendants’ discriminatory practices against females, and this is not just as it applied to plaintiff only,” because such a claim “identifies a common legal issue, discrimination against women, and a common factual problem, discrimination as applied” by the defendant. Bouman v. Block, 940 F.2d 1211, 1232 (9th Cir.1991) (internal quotation marks omitted); see also Rodriguez v. Hayes, 591 F.3d 1105, 1122-23 (9th Cir. 2010) (“[T]he commonality requirements ask[s] us to look only for some shared legal issue or a common core of facts.”). We have also, of course, rejected the “across-the-board” claims that Falcon prohibited, and we have denied class certification when a rigorous analysis did not show the class would implicate common questions. See, e.g., Lozano v. AT & T Wireless Servs., Inc., 504 F.3d 718, 730 (9th Cir.2007) (citing Falcon in rejecting certification of a class action based on the unconscionability of a class action waiver because the district court did not perform the requisite analysis). As one example, we have explained that “commonality is satisfied where the lawsuit challenges a system-wide practice or policy that affects all of the putative class members,” because such a system implicates common factual questions. Armstrong, 275 F.3d at 868.

Our most recent statement considering Falcon and Eisen comports with this understanding as well. Writing for the court in United Steel Workers v. ConocoPhillips Co., Judge Bybee explained that it is the plaintiffs theory that matters at the class certification stage, not whether the theory will ultimately succeed on the merits. See 593 F.3d 802, 808-09 (9th Cir.2010). We thus found reversible error because the district court did not consider whether the plaintiffs legal theory “was one in which common issues of law and or fact did not predominate,” and instead found the class did not meet Rule 23’s requirements because “ ‘there can be no assurances that [plaintiffs] w[ould] prevail on this theory.’ ” Id. (citing the district court) (first emphasis added). In short, we have consistently held that when considering how the facts and legal issues apply to a class under Rule 23(a), the district court must focus on common questions and common issues, not common proof or likely success on the questions commonly raised. See id.

For the most part, the dozens of district court cases in this circuit parsing Eisen, Falcon, and our precedent, have considered Eisen’s language in context, concluding that they must make determinations that the requirements of Rule 23 have been met, and acknowledging that these determinations will sometimes require examining issues that overlap with the merits. See, e.g., Endres v. Wells Fargo Bank, No. C 06-7019, 2008 WL 344204, at *2 (N.D.Cal. Feb. 6, 2008) (“[T]he court is not permitted to make a preliminary inquiry into the merits of the plaintiffs’ claims at that stage of the litigation. Nevertheless, the court must consider evidence relating to the merits if such evidence also goes to the requirements of Rule 23.” (internal quotation marks and citation omitted)); Westways World Travel, Inc. v. AMR Corp., 218 F.R.D. 223, 230 (C.D.Cal.2003) (“Although the Supreme Court once stated that courts have no authority to conduct a preliminary inquiry into the merits of a suit in order to determine whether it may be maintained as a class action, [citing Eisen ] later cases make clear that a preliminary inquiry into the merits is sometimes necessary to make a meaningful determination of class certification issues. At the same time, the court is required to take the allegations of the plaintiffs complaint as true. Plaintiff bears the burden of showing that the elements of Rule 23 are met.” (citations omitted)).⁸ These cases have correctly threaded the needle between Eisen, Falcon, and our precedent.

In the less typical instances in which district courts in this circuit have been led astray, a common reason seems to have been a misreading of our statement in footnote 17 of Blackie v. Barrack, in which we explained that district courts are “bound to take the substantive allegations of the complaint as true, thus necessarily making the class order speculative in the sense that the plaintiff may be altogether unable to prove his allegations.” 524 F.2d at 901 n. 17; see also Edwards v. City of Long Beach, 467 F.Supp.2d 986, 991 (C.D.Cal.2006) (“The court is bound to take the substantive allegations in the complaint as true.”). But, that statement is taken out of context when relied on in this manner.⁹ The subsequent sentences in Blackie, which the erring district courts almost universally neglect to include, explained that, although “the court may not put the plaintiff to preliminary proof of his claim, it does require sufficient information to form a reasonable judgment. Lacking that, the court may request the parties to supplement the pleadings with sufficient material to allow an informed judgment on each of the Rule’s requirements.” 524 F.2d at 901 n. 17 (emphasis added). Discussing when a certification inquiry overlaps with the merits, we further explained in Blackie:

Nor is the class issue separable from the merits in all cases (including this one). The common questions, typicality, conflicts and adequacy of representation, Fed.R.Civ.P. 23(a), and predominance tests, Fed.R.Civ.P. 23(b)(3), are determinations ... which may require review of the same facts and the same law presented by review of the merits.

524 F.2d at 897.

A better reading of Blackie, then, is to understand it as having the meaning that our cases, and the majority of district court opinions, have ascribed to it: Blackie is entirely consistent with the Supreme Court’s guidance, explicitly requiring a district court to probe behind the pleadings if doing so is necessary to make findings on the Rule 23 certification decision. Thus, Blackie does not, and could not, require the district court to unquestioningly accept a plaintiffs arguments as to the necessary Rule 23 determinations. The sense in which Blackie referred to class certification as speculative is the same way in which we most recently, and more artfully, described the inquiry:

[A] court can never be assured that a plaintiff will prevail on a given legal theory prior to a dispositive ruling on the merits, and a full inquiry into the merits of a putative class’s legal claims is precisely what both the Supreme Court and we have cautioned is not appropriate for a Rule 23 certification inquiry.

United Steel Workers, 593 F.3d at 808-09.

In short, the explanation we provide today is not a new standard at all. Though a small number of district courts in this circuit have misunderstood Blackie and relied on Eisen in the way that the Second Circuit has cautioned against, and that we now reject, the precedent from this court is consistent. We are unable to find a single case in our court that incorrectly relied on the “no merits inquiry” language from Eisen in certifying a class without examining necessary issues because they overlapped with the merits. Cf. Caridad, 191 F.3d at 291-92(applying the “no merits inquiry” language from Eisen out of context to prohibit the proper Rule 23 analysis). Our explanation confirms what our decisions have held for more than twenty-five years: A district court must sometimes resolve factual issues related to the merits to properly satisfy itself that Rule 23’s requirements are met, but the purpose of the district court’s inquiry at this stage remains focused on, for example, common questions of law or fact under Rule 23(a)(2), or predominance under Rule 23(b)(3), not the proof of answers to those questions or the likelihood of success on the merits. Falcon, 457 U.S. at 158, 102 S.Ct. 2364(requiring a “specific presentation identifying the questions of law or fact that were common to the claims of respondent and of the members of the class he sought to represent” (emphasis added)); United Steel Workers, 593 F.3d at 808-09 (district court erred in basing its decision on whether the plaintiffs could prove merits rather than whether their claims implicated common questions of law or fact). Of course, for class certification to be proper, the common questions cannot simply arise from artful pleading. Plaintiffs must raise questions that the district court concludes, after a rigorous analysis, are susceptible to common resolution at a later stage.

D. Clarifying the Standard

1. The Proper Standard of Rule 23 Adjudication

This review of Supreme Court dictates, as well as our own and other circuits’ treatment of the issue, leads us to recognize a number of constant holdings across circuits that we must incorporate into our clarification of the proper standard governing a district court in considering a Rule 23 motion for class certification.

To begin with, at the class certification stage, while Eisen prohibits a court from making determinations on the merits that do not overlap with the Rule 23 inquiry, district courts must make determinations that each requirement of Rule 23 is actually met. This bedrock rule is consistent with the Supreme Court’s statements, other circuits’ decisions, and our longstanding precedent. While plaintiffs need not make more than allegations as to their substantive claims, whether the suit is appropriate for class resolution must be actually demonstrated, not just alleged, to the district court’s satisfaction.

In addition, in the cases such as IPO in which courts have recognized how Eisen is sometimes misunderstood, those circuits have uniformly reserved discretion with the district court to avoid a trial-level inquiry at the certification stage despite the need to find the Rule 23 requirements met.¹⁰ See, e.g., Blades, 400 F.3d at 567 (“The closer any dispute at the class certification stage comes to the heart of the claim, the more cautious the court should be in ensuring that it must be resolved in order to determine the nature of the evidence the plaintiff would require.”). District courts must maintain the ability to cut off discovery to avoid either party bootstrapping a trial or summary judgment motion into the certification stage. Nearly every circuit to consider the issue, including our own, has recognized the practical importance of the certification decision as leverage for settlement, yet Rule 23 gives neither party the right to turn the certification decision into a trial.

When reading recent class certification cases, one also notices the prevalence of securities fraud cases, and particularly the fraud-on-the-market presumption, in the evolution of the Rule 23 standard. See, e.g., Oscar Private Equity Invs. v. Allegiance Telecom, Inc., 487 F.3d 261, 264 (5th Cir.2007); IPO, 471 F.3d at 30-31; Bowe v. Polymedica Corp. (In re PolyMedica Corp. Secs. Litig.), 432 F.3d 1, 6-7(1st Cir.2005); Unger, 401 F.3d at 323-24; Gariety, 368 F.3d at 364-66; West v. Prudential Sec., Inc., 282 F.3d 935, 937-38(7th Cir.2002).

It is an interesting feature of the case, law’s development that it has occurred largely in these fraud-on-the-market cases, in which a plaintiff typically must show the six “basic elements” of a securities fraud action, Dura Pharms., Inc. v. Broudo, 544 U.S. 336, 341-42, 125 S.Ct. 1627, 161 L.Ed.2d 577 (2005), and the efficient market component of the causation element overlaps with the merits. We must, then, be aware that applying the same procedural rules, such as Rule 23, can lead to different outcomes when the underlying legal and factual framework is different. See, e.g., Hohider v. United Parcel Serv., Inc., 574 F.3d 169, 182-85 (3d Cir.2009) (describing how the pattern and practice evidentiary framework from Title VII cases is not perfectly applicable in the context of the Americans with Disabilities Act).

Thus, in contrast to a securities class action based on a fraud-on-the-market theory, in a pattern and practice discrimination case, a plaintiff will typically not come to court in the first place without anecdotal evidence. For practical purposes, assuming a plaintiff possesses anecdotal evidence, the plaintiffs statistical evidence does not overlap with the merits, it largely is the merits. See Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 986-87, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988); Int’l Bhd. of Teamsters v. United States, 431 U.S. 324, 337-39, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977). This means that disputes over whose statistics are more persuasive are often not disputes about whether the plaintiffs raise common issues or questions, but are really arguments going to proof of the merits. See Nagareda, Class Actions, supra, at 619-20 (“Nowhere is this [disputed statistics] problem more acute than in employment discrimination class actions centered on statistical analysis of an aggregate nature.”). Of course, if one party argued that the other party’s statistics were unreliable or based on an unaccepted method, this may be an issue the district court would have to resolve to determine whether the potentially problematic statistics were even capable of raising a common question.

In fact, after IPO, but before his recent elevation to the Second Circuit, Judge Gerard Lynch recognized this problem in a Title VII gender discrimination case in which the plaintiffs were seeking to certify on the same grounds as those here. Predictably faced with a dispute regarding whether the plaintiffs’ statistics demonstrated commonality, Judge Lynch addressed IPO’s application in a Title VII, gender discrimination context.¹¹

The problem, the district court explained, was that “[i]n deciding that the class certification order complied with In re IPO, it is important to note that disparate impact cases present unique difficulties in analyzing the commonality requirement of Rule 23(a). Plaintiffs in disparate impact cases often rely on statistical evidence to prove the merits of their claim.” Hnot v. Willis Group Holdings Ltd., 241 F.R.D. 204, 210-11 (S.D.N.Y.2007). The court continued, “[i]n such a ease, however, the same evidence must be considered to determine whether a plaintiff has satisfied the commonality requirement.” Id. at 211.

In resolving this problem and certifying the class, the district court noted, “[contrary to defendants’ assertions, In re IPO does not stand for the proposition that the Court should, or is even authorized to, determine which of the parties’ expert reports is more persuasive. Defendants ignore the fact that In re IPO specifically rejected this interpretation of Rule 23.” Id. at 210. Instead, Judge Lynch explained, “In re IPO reiterated that ‘experts’ disagreement on the merits — whether a discriminatory impact [can] be shown — [is] not a valid basis for denying class certification.’ ” Id. (alterations in original) (quoting IPO, 471 F.3d at 35). Thus, the court could only “examine the expert reports as far as they bear on the Rule 23 determination.” Id.

Judge Lynch similarly clarified, with regard to the disparate treatment issue in the case before him, that “plaintiffs and defendants disagree on whose statistical findings and observations are more credible, but this disagreement is relevant only to the merits of plaintiffs’ claim — whether plaintiffs actually suffered disparate treatment — and not to whether plaintiffs have asserted common questions of fact or law; plaintiffs’ ultimate success at trial on the merits requires an answer to that question, specifically that defendants actually did discriminate against plaintiffs.” Id. at 210-11. “By asking the Court to decide which expert report is more credible, defendants are requesting that the Court look beyond the Rule 23 requirements and decide the issue on the merits, a practice In re IPO specifically cautions against.” Id. at 210.

Thus, in addition to demonstrating Rule 23’s proper implementation in various legal contexts (securities class actions versus Title VII claims), Hnot also illustrates a second feature of the certification cases that we must address. Because the fraud-on-the-market cases are typically decided under Rule 23(b)(3)’s predominance requirement, a related feature of the circuit court cases considering the proper standard a district court applies, when deciding whether to certify a class, is the difference between cases describing review of evidence under Rule 23(a) and those under Rule 23(b).¹² We are unaware of any argument claiming that a different standard should apply in the two inquiries, and the cases generally treat the precedent as applying equally to both contexts. We do, however, note that the inquiry cannot be divorced from the text of Rule 23, which, of course, requires plaintiffs to make different showings under different parts of the rule. This distinction helps order the doctrine, keeping in mind the different requirements under Rule 23(a) and (b) and how those requirements may play out in a district court’s certification decision.

This insight derives from the Supreme Court’s explanation of the predominance test under Rule 23(b)(3). Predominance, the Court explained, “tests whether proposed classes are sufficiently cohesive to warrant adjudication by representation,” Amchem Prods., Inc. v. Windsor, 521 U.S. 591, 623, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997), a standard “far more demanding” than the commonality requirement of Rule 23(a), id. at 623-24, 117 S.Ct. 2231. Thus, we would expect that cases in which the parties are contesting facts underlying the Rule 23(b)(3) determination may often require more determinations by the district court than those in which Rule 23(a)(2) is the primarily contested issue. Rule 23(a)(2) is about invoking common questions, see Falcon, 457 U.S. at 158, 102 S.Ct. 2364, whereas Rule 23(b)(3) requires a district court to formulate “some prediction as to how specific issues will play out in order to determine whether common or individual issues predominate in a given case,” New Motor Vehicles, 522 F.3d at 20 (internal quotation marks omitted). We thus should not be surprised that a district court will have to make more precise factual determinations under Rule 23(b)(3) than under Rule 23(a)(2). This insight does not apply a procedural rule differently in two like contexts; it applies the text of Rule 23 itself to the district court’s task.

While we find the case law across circuits more uniform than some courts have implied, see id. at 24, to the extent it is not, this result may be because of courts’ failure to recognize this key difference between a district court’s job under Rule 23(a)(2) and its job under Rule 23(b)(3).¹³ This conclusion is further supported when one notices that Falcon's primary inquiry was under Rule 23(a), 457 U.S. at 156-58, 102 S.Ct. 2364, whereas many of the circuit courts discussed above were concerned with Rule 23(b)’s predominance test and, thus, required the district court to make more detailed determinations as to how the facts at issue would play out — whether they would predominate — in the merits of the litigation. See, e.g., Hydrogen Peroxide, 552 F.3d at 310 (Rule 23(b)(3) predominance); New Motor Vehicles, 522 F.3d at 26 (same); IPO, 471 F.3d at 35 (same); In re PolyMedica Corp. Secs. Litig., 432 F.3d at 3 & n. 4 (same); Unger, 401 F.3d at 320, 324-25 (same); Blades, 400 F.3d at 566 (same); Gariety, 368 F.3d at 364-65 (same); Szabo, 249 F.3d at 676-77 (same).

Though some courts have noted the difference between Rule 23(a) and 23(b) in passing, see Hydrogen Peroxide, 552 F.3d at 310 — 11,¹⁴ we are not aware of any decisions detailing the difference, and some courts have failed to note the distinction at all. See, e.g., IPO, 471 F.3d at 33 n. 3 (“We see no reason to doubt that what the Supreme Court said about Rule 23(a) requirements applies with equal force to all Rule 23 requirements, including those set forth in Rule 23(b)(3).”). We think a great deal of any disparity in different courts’ explication of the Rule 23 standard can be explained by this difference. The lesson for future district courts is that, in a given case, the text of Rule 23(a), as compared to Rule 23(b), may require them to determine more or different facts (typically more under Rule 23(b)(3)) to determine whether the plaintiffs have met their Rule 23 burden.

In short, these observations, which include the Supreme Court’s direction, long-standing precedent in this court, and treatment from other circuits, lead us to the following explanation of the proper standards governing a district court’s adjudication of a Rule 23 motion for class certification. First, when considering class certification under Rule 23, district courts are not only at liberty to, but must, perform a rigorous analysis to ensure that the prerequisites of Rule 23 have been satisfied, and this analysis will often, though not always, require looking behind the pleadings to issues overlapping with the merits of the underlying claims. It is important to note that the district court is not bound by these determinations as the litigation progresses. Second, district courts may not analyze any portion of the merits of a claim that do not overlap with the Rule 23 requirements. Relatedly, a district court performs this analysis for the purpose of determining that each of the Rule 23 requirements has been satisfied. Third, courts must keep in mind that different parts of Rule 23 require different inquiries. - For example, what must be satisfied for the commonality inquiry under Rule 23(a)(2) is that plaintiffs establish common questions of law and fact, and answering those quéstions is the purpose of the merits inquiry, which can be addressed at trial and at summary judgment. Fourth, district courts retain wide discretion in class certification decisions, including the ability to cut off discovery to avoid a mini-trial on the merits at the certification stage. Fifth, different types of cases will result in diverging frequencies with which the district court will properly invoke its discretion to abrogate discovery. As just one example, we would expect a district court to circumscribe discovery more often in a Title VII case than in a securities class action resting on a fraud-on-the-market theory, because the statistical disputes typical to Title VII cases often encompass the basic merits inquiry and need not be proved to raise common questions and demonstrate the appropriateness of class resolution. Plaintiffs pleading fraud-on-the-market, on the other hand, may have to establish an efficient market to even raise common questions or show predominance.

2. The Dissent’s “Significant Proof’ Standard

In addition to setting out our own review, we feel compelled to discuss the dissent’s consideration of this issue. The Supreme Court’s decisions in Falcon and Eisen are the primary guides to our ruling in this case. The dissent, in suggesting that we are unfaithful to Falcon, seeks to create a new class action requirement based on a hypothetical in one sentence of Supreme Court dicta; conflates, or at least fails to distinguish, the posture of Falcon and the present case; ignores the weight of the many cases in other circuits arriving at the same standard we have described above; and renders itself unpersuasive by critiquing the district court’s eighty-four-page analysis as insufficiently rigorous. In doing so, the dissent is unfaithful to the actual distinctions the Supreme Court relied upon in Falcon, and variously depicts Falcon as instituting a “significant proof’ “burden” or “ ‘significant proof requirement” that Falcon did not create. Dissent at 633, 637-38.

We read Falcon, as has nearly every Court of Appeals to consider the question, as creating the standard we describe above. But in discussing what it views as the analysis required by Falcon, the dissent quotes a portion of a sentence of Falcon dicta in footnote 15 as standing for the requirement that plaintiffs cannot prevail at the certification stage without showing “[significant proof that an employer operated under a general policy of discrimination.” Id. at 632. However, the entire footnote sentence reads as follows: “Significant proof that an employer operated under a general policy of discrimination conceivably could justify a class of both applicants and employees if the discrimination manifested itself in hiring and promotion practices in the same general fashion, such as through entirely subjective decision-making processes.” Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. 2364 (emphasis added).¹⁵

Falcon’s discussion of two distinct processes — hiring and promotion — for which “significant proof’ could prove sufficient to certify a single class, is an unusually high standard that Plaintiffs here need not meet because they did not present the distinct legal theories of recovery that the Falcon plaintiffs, both employees and applicants, had pursued together in one class. “The question before the district court was not whether [Plaintiffs] have definitively proven disparate treatment and a disparate impact; rather, the question was whether the basis of [Plaintiffs’] discrimination claims was sufficient to support class certification.” Brown v. Nucor Corp., 576 F.3d 149, 156 (4th Cir.2009), cert. denied, — U.S.-, 130 S.Ct. 1720, 176 L.Ed.2d 185 (2010).

Contrary to the dissent,¹⁶ Falcon does not say that Plaintiffs must show a common policy of proven discrimination at the class action stage, rather than just a common policy alleged to be discriminatory. 457 U.S. at 158-59, 102 S.Ct. 2364. Of course, as we have already explained, named plaintiffs must do more than merely allege discriminatory practices against themselves. Id. As we demonstrate below, Plaintiffs here do so, making showings through their expert testimony and statistical evidence, which further distinguishes the certification decision here from that decision in Falcon. In maintaining otherwise in its discussion, the dissent confuses merits decisions such as Teamsters, 431 U.S. at 336, 97 S.Ct. 1843, and class action certification decisions.¹⁷ It also ignores our previous statement examining Falcon footnote 15 in which we explained that “[w]e understand footnote fifteen of Falcon to present a demonstrative example rather than a limited exception to the overall skepticism toward broad discrimination class actions.” Staton, 327 F.3d at 955. In other words, as we have previously described Falcon, the Supreme Court “does not generally ban all broad classes but rather precludes a class action that, on the basis of one form of discrimination against one or a handful of plaintiffs, seeks to adjudicate all forms of discrimination against all members of a group protected by Title VII, § 1981, or a similar statute.” Id.

The dissent also largely ignores Supreme Court guidance by failing to recognize that Falcon addressed the claim that the allegations of an employee subject to discrimination in promotion decisions “fairly encompassed” the claims of non-employees allegedly subject to discrimination in a discrete hiring process. Falcon, 457 U.S. at 158, 102 S.Ct. 2364. Such a determination under Rule 23 is clearly distinct from an inquiry where, as here, a class consists entirely of individuals actually employed under the same corporate policies.¹⁸ Critically, in Falcon, the individual plaintiffs claim was based on promotion practices for his allegations and hiring practices for the putative class members, yet Falcon’s “complaint contained no factual allegations concerning petitioner’s hiring practices,” id. at 150, 102 S.Ct. 2364, despite the fact that the class members’ claims depended on pattern or practice in hiring, id. at 159, 102 S.Ct. 2364. In contrast, where the individual plaintiffs seek to prove their own cases through pattern or practice methods, they are necessarily dependent on proving facts relevant to others of the same protected group subject to the same policy, class action or no class action. See Watson, 487 U.S. at 994-95, 108 S.Ct. 2777; see also Nagareda, Aggregate Proof, supra, at 150 (“The terms ‘pattern’ and ‘practice’ themselves imply an aggregate perspective.”).

Finally, Plaintiffs here are unlike the plaintiff in Falcon, who failed to “otherwise [jsupport [his] allegation that the company ha[d] a policy of discrimination” except by claiming that he himself had been denied a promotion on discriminatory grounds. Falcon, 457 U.S. at 157-58, 102 S.Ct. 2364. As we detail below, Plaintiffs here have introduced “significant proof’ of Wal-Mart’s policies, and their effects on the certified class, and have introduced evidence of far more than “the validity of [their] own claim[s].” Id. Even if the dissent were correct in creating a “significant proof’ standard (or burden, or requirement), which it is not, it would not apply to Plaintiffs in this case. “Subjective decisionmaking processes” are exactly what the Plaintiffs allege here and what the Supreme Court’s hypothetical expressed concern with in Falcon. Id. at 159 n. 15, 102 S.Ct. 2364. In addition, it is not clear that such a standard, if it existed, need apply to Plaintiffs — everyone of whom was or is an employee- — at all because they are not situated as the Falcon plaintiffs, who were both employees seeking promotion and job applicants pursuing a position and thus needed to show different facts and allege diverging legal theories.

Again unlike in Falcon, and as discussed in detail below, the district court here did not presume or fail “to evaluate carefully the legitimacy [of Plaintiffs’ claims to be] proper class representative^],” id. at 160, 102 S.Ct. 2364, but rather found Rule 23 satisfied only after undertaking the “rigorous analysis” the Supreme Court requires, see Dukes, 222 F.R.D. at 143-69. The specific recognition of Falcon's “rigorous analysis” requirement guided the district court’s analysis throughout its lengthy order, and the court made determinations that each of Rule 23’s requirements were satisfied. Id. at 143-44, 166-68 (applying Falcon). The dissent’s attenuated claim that the district court abused its discretion by failing to require a “specific presentation identifying the questions of law or fact that were common to the claims of respondent and of the members of the class he sought to represent,” see Falcon, 457 U.S. at 158, 102 S.Ct. 2364, is vitiated by the twenty-four pages in which the district court exhaustively performed exactly that analysis, explicitly considering questions of law and fact that were common to the members of the class and the named representatives,¹⁹ see Dukes, 222 F.R.D. at 145-69. In short, and as we will now explain, at the certification stage, it is difficult for us to envision a more rigorous analysis than the one the district court conducted.

III. Certification of This Class

Unsurprisingly, the class in this case is broad and diverse, encompassing both salaried and hourly employees in a range of positions, who are or were employed at one or more of Wal-Mart’s 3,400 stores across the country. The district court found that the large class is united by a complex array of company-wide practices, which Plaintiffs contend discriminate against women.

A. Rule 23(a)

At the outset of considering the district court’s certification order in light of the above-clarified standard, we note two facets of the district court’s review.

First, this case reached us after a significantly more searching review than cases in other circuit courts explaining the Rule 23 standard. In Szabo, for example, the district court “assumed that whatever [the plaintiff] allege[d] must be true,” and it “[p]roeeed[ed] as if class certification under Rule 23 were governed by the same principles as evaluating the sufficiency of the complaint under Rule 12(b)(6).” 249 F.3d at 674-75. In Gariety, the case came to the Fourth Circuit after the district court had concluded that “the fact that the plaintiffs have asserted that the Keystone market was efficient is enough at the certification stage to find the market efficient.” 368 F.3d at 361 (emphasis added); see also Vallario v. Vandehey, 554 F.3d 1259, 1265-66 (10th Cir.2009) (district court abused its discretion in thinking it was barred from any consideration of action’s merits); IPO, 471 F.3d at 30 (district court finding that only “plaintiffs — who have the burden of proof at class certification — must make ‘some showing ’ ”).

In other words, previous circuit courts addressing similar issues have been faced with district courts making the same mistake that we described a small number of district courts in this circuit have made in citing Blackie out of context: refusing to consider any issue overlapping with the merits, and assuming the plaintiffs’ substantive allegations as true for the Rule 23 inquiry. As we demonstrate below, such review is entirely dissimilar from the district court’s review in this case.

Second, we also note that resolving this case requires a close reading of the district court’s lengthy opinion to determine the standard the district court applied in deciding to certify the class. Contrary to the claims of Wal-Mart and the dissent, such a reading of the district court’s individual determinations on Rule 23(a) shows that the district court actually weighed evidence and made findings sufficient under the standard we have described above.

In broadly discussing the standard of review, the district court stated the following:

A party seeking to certify a class must demonstrate that it has met all four requirements of Federal Rule of Civil Procedure 23(a), and at least one of the requirements of Rule 23(b). Rule 23(a) requires that all of the following four factors be met .... In short, the class must satisfy the requirements of numerosity, commonality, typicality, and adequacy. ...

The party seeking certification must provide facts sufficient to satisfy Rule 23(a) and (b) requirements. In turn, the district court must conduct a rigorous analysis to determine that the prerequisites of Rule 23 have been met. Gen. Tel. Co. v. Falcon., 457 U.S. 147, 161, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982). If a court is not fully satisfied, certification should be refused.... See Fed. R.Civ.P. 23 advisory committee’s note to 2003 amends.

... “[AJlthough some inquiry into the substance of a case may be necessary to ascertain satisfaction of the commonality and typicality requirements of Rule 23(a), it is improper to advance a decision on the merits to the certification stage.” Moore v. Hughes Helicopters, Inc., 708 F.2d 475, 480 (9th Cir.1983) (citation omitted); see also Nelson v. United States Steel Corp., 709 F.2d 675, 679-80 (11th Cir.1983) (plaintiffs’ burden at class certification “entails more than the simple assertion of [commonality and typicality] but less than a prima facie showing of liability”) (citation omitted).

Dukes, 222 F.R.D. at 143-44 (some citations omitted) (last alteration in original) (emphasis added). Unlike some other district courts, which have assumed facts in a complaint, the district court here spent significant effort explaining a nuanced standard that clearly comprehended the need to look well beyond the pleadings in the case.

Most importantly, as we will now explain, the district court’s standard led to a review that complies with Falcon and with our herein-described standards that a district court must follow when deciding whether to certify a class.

1. Numerosity

Rule 23(a)(1) requires that the class be “so numerous that joinder of all members is impracticable.” Wal-Mart does not contest this point.

2. Commonality

Rule 23(a)(2) requires that “there are questions of law or fact common to the class.” Commonality focuses on the relationship of common facts and legal issues among class members. See, e.g., Conte & Newberg, supra, § 3:10, at 271. We noted in Hanlon, 150 F.3d at 1019:

Rule 23(a)(2) has been construed permissively. All questions of fact and law need not be common to satisfy the rule. The existence of shared legal issues with divergent factual predicates is sufficient, as is a common core of salient facts coupled with disparate legal remedies within the class.

The commonality test is “qualitative rather than quantitative” — one significant issue common to the class may be sufficient to warrant certification, see, e.g., Savino v. Computer Credit, Inc., 173 F.R.D. 346, 352 (E.D.N.Y.1997), aff'd, 164 F.3d 81 (2d Cir. 1998); see also Conte & Newberg, supra, § 3:10, at 272-74, a standard that dovetails with our discussion above, requiring plaintiffs to show common questions or issues.

The district court found that Plaintiffs had provided evidence sufficient to support their contention that significant factual and legal questions are common to all class members. After analyzing Plaintiffs’ evidence, the district court stated:

Plaintiffs have exceeded the permissive and minimal burden of establishing commonality by providing: (1) significant evidence of company-wide corporate practices and policies, which include (a) excessive subjectivity in personnel decisions, (b) gender stereotyping, and (c) maintenance of a strong corporate culture; (2) statistical evidence of gender disparities caused by discrimination; and (3) anecdotal evidence of gender bias. Together, this evidence raises an inference that Wal-Mart engages in discriminatory practices in compensation and promotion that affect all plaintiffs in a common manner.

Dukes, 222 F.R.D. at 166. Discussing commonality, the district court alluded to the determinations Rule 23 requires in certifying a class. The court stated, “Rule 23(a)(2) requires that common questions of law or fact exist among class members.” Id. at 144. Recognizing this requirement of Rule 23, the district court went on to make reasoned determinations that commonality was met, concluding that “th[e] evidence more than satisfies plaintiffs’ burden to demonstrate commonality under Rule 23(a)(2).” Id. at 145. The district court then detailed the four categories of commonality evidence to which we now turn. Id. We conclude, as explained in more detail below, that it was within the district court’s discretion, and in line with Falcon and the standard we have clarified today, to determine that the commonality prerequisite to class certification was satisfied. See Fed.R.Civ.P. 23(a)(2).²⁰

a. Evidence of a Common Policy of Discrimination

Pursuant to Falcon, 457 U.S. at 161, 102 S.Ct. 2364, Plaintiffs presented four categories of commonality evidence that the district court subjected to a “rigorous analysis”: (1) facts supporting the existence of company-wide policies and practices that, in part through their subjectivity, provide a potential conduit for discrimination; (2) expert opinions supporting the existence of company-wide policies and practices that likely include a culture of gender stereotyping; (3) expert statistical evidence of class-wide gender disparities attributable to discrimination; and (4) anecdotal evidence from class members throughout the country of discriminatory attitudes held or tolerated by management. See Dukes, 222 F.R.D. at 145-66. Wal-Mart contends this evidence is insufficient to suggest a common factual or legal question related to the existence of discrimination.

(1) Factual Evidence

As factual evidence, Plaintiffs presented evidence of the following: (1) uniform personnel and management structure across stores; (2) Wal-Mart headquarters’s extensive oversight of store operations, company-wide policies governing pay and promotion decisions, and a strong, centralized corporate culture; and (3) consistent gender-related disparities in every domestic region of the company. Such evidence supports Plaintiffs’ contention that WalMart operates a highly centralized company that promotes policies common to all stores and maintains a single system of oversight. Wal-Mart does not challenge this evidence.

(2) Expert Opinion

Plaintiffs presented evidence from Dr. William Bielby, a sociologist, to interpret and explain the facts that suggest that Wal-Mart has and promotes a strong corporate culture — a culture that may include gender stereotyping. Dr. Bielby based his opinion on, among other things, Wal-Mart managers’ deposition testimony; organizational charts; correspondence, memos, reports, and presentations relating to personnel policy and practice, diversity, and equal employment opportunity issues; documents describing the culture and history of the company; and a large body of social science research on the impact of organizational policy and practice on workplace bias.

Dr. Bielby testified that he employed a social framework analysis to examine the distinctive features of Wal-Mart’s policies and practices and evaluated them “against what social science shows to be factors that create and sustain bias and those that minimize bias.”²¹ In Dr. Bielby’s opinion, “social science research demonstrates that gender stereotypes are especially likely to influence personnel decisions when they are based on subjective factors, because substantial decisionmaker discretion tends to allow people to seek out and retain stereotyping-confirming information and ignore or minimize information that defies stereotypes.” Id. at 153(internal quotation marks omitted). Dr. Bielby concluded that: (1) Wal-Mart’s centralized coordination, reinforced by a strong organizational culture, sustains uniformity in personnel policy and practice; (2) there are significant deficiencies in Wal-Mart’s equal employment policies and practices; and (3) Wal-Mart’s personnel policies and practices make pay and promotion decisions vulnerable to gender bias. See id. at 153— 54.

The district court reviewed Plaintiffs’ and Wal-Mart’s competing claims as to Wal-Mart’s uniform culture and determined that “the evidence indicates that in-store pay and promotion decisions are largely subjective and made within a substantial range of discretion by store or district level managers, and that this is a common feature which provides a wide enough conduit for gender bias to potentially seep into the system.” Id. at 152. Having evaluated this evidence in detail, the court determined “that given the evidence regarding strong uniform culture and policies, the degree and impact of this practice is a significant question of fact common to the class as a whole.” Id. at 153. Such a reasoned determination is what our standard for Rule 23 requires.

Wal-Mart vigorously challenges Dr. Bielby’s third conclusion as vague and imprecise because he concluded that WalMart is “vulnerable” to bias or gender stereotyping but failed to identify a specific discriminatory policy at Wal-Mart. Specifically, Wal-Mart contends that Dr. Bielby’s testimony does not meet the standards for expert testimony set forth in Federal Rule of Evidence 702 and Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579, 597, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993), which held that a trial court must act as a “gatekeeper” in determining whether to admit or exclude expert evidence.

Wal-Mart made an identical argument to the district court and the district court properly rejected it. A close reading of the district court’s order demonstrates its correct understanding of its role at the Rule 23(a)(2) certification stage — to make factual determinations regarding evidence as it relates to common questions of fact or law but not to decide which parties’ evidence is ultimately more persuasive as to liability. The court stated, “Dr. Bielby presents enough of a basis, both in his review of the scientific literature and on the facts of the case, to provide a foundation for his opinions.” Dukes, 222 F.R.D. at 154. The court correctly explained that whether the jury was ultimately persuaded by those opinions was a question on the merits. For the class certification, however, Dr. Bielby’s opinions, for which WalMart did not challenge the methodology, raised a question “of corporate uniformity and gender stereotyping that is common to all class members.” Id. We cannot say that considering Dr. Bielby’s opinions in this method was an abuse of discretion.

This conclusion is furthered by the fact that Wal-Mart did not (and does not) challenge Dr. Bielby’s methodology or contend that his findings lack relevance because they “do[ ] not relate to any issue in the case,” Daubert, 509 U.S. at 591, 113 S.Ct. 2786 (internal quotation marks omitted). Wal-Mart challenges only whether certain inferences can be persuasively drawn from his data. But because Daubert does not require a court to admit or exclude evidence based on its persuasiveness, but rather requires a court to admit or exclude evidence based on its scientific reliability and relevance, id. at 587-92, 113 S.Ct. 2786(relevance standard “is a liberal one,” under which evidence is relevant if it has “ ‘any tendency to make the existence of any fact that is of consequence to the determination of the action more probable or less probable than it would be without the evidence’ ” (quoting Fed.R.Evid. 401)), testing Dr. Bielby’s testimony for “Daubert reliability” would not have addressed Wal-Mart’s objections. It would have simply revealed what Wal-Mart itself has admitted and courts have long accepted: that properly analyzed social science data, like that offered by Dr. Bielby, may support a plaintiffs assertions that a claim is proper for class resolution. See Price Waterhouse v. Hopkins, 490 U.S. 228, 235-36, 255, 109 S.Ct. 1775, 104 L.Ed.2d 268 (1989) (considering similar evidence offered by expert social psychologist).

Accordingly, Wal-Mart’s contention that the district court was required to strike Dr. Bielby’s testimony under the Daubert test at the class certification stage, simply because the conclusion he reached seemed unpersuasive absent certain corroborating evidence, is misplaced.²² See Daubert, 509 U.S. at 595, 113 S.Ct. 2786 (“The focus, of course, must be solely on principles and methodology, not on the conclusions that they generate.”). While a jury may ultimately agree with Wal-Mart that, in the absence of a specific discriminatory policy promulgated by Wal-Mart, it is not more likely than not, based solely on Dr. Bielby’s analysis, that Wal-Mart engaged in actual gender discrimination, that question must be left to the merits stage of the litigation (and presumably will not have to be decided as there will be other evidence). At the class certification stage, it is enough that Dr. Bielby presented scientifically reliable evidence tending to show that a common question of fact — i.e., “Does Wal-Mart’s policy of decentralized, subjective employment decision making operate to discriminate against female employees?” — exists with respect to all members of the class.²³ This he did and, thus, we find no error in the district court’s acceptance of Dr. Bielby’s evidence to support a finding of commonality.

(3) Statistical Evidence

It is well established that plaintiffs may demonstrate commonality by presenting statistical evidence, which survives a “rigorous analysis,” sufficient to fairly raise a common question concerning whether there is class-wide discrimination. See Falcon, 457 U.S. at 159 n. 15, 161, 102 S.Ct. 2364; Caridad, 191 F.3d at 292;²⁴ see also Stastny v. S. Bell Tel. & Tel. Co., 628 F.2d 267, 278 (4th Cir.1980) (recognizing that statistical data showing comparable disparities experienced by protected employees may indicate a policy or practice that commonly affects the class members and raises a common question concerning whether the pattern or practice is discriminatory).

A careful reading of the district court’s treatment of the competing statistical evidence demonstrates that, in conducting its analysis, the district court followed the correct standard as explained in Falcon and our earlier cases, and clarified today.

Dr. Richard Drogin, Plaintiffs’ statistician, analyzed data at a regional level. He ran separate regression analyses for each of the forty-one regions²⁵ containing WalMart stores.²⁶ He concluded that “there are statistically significant disparities between men and women at Wal-Mart in terms of compensation and promotions, that these disparities are widespread across regions, and that they can be explained only by gender discrimination.” Dukes, 222 F.R.D. at 154. Dr. Marc Ben-dick, Plaintiffs’ labor economics expert, conducted a “benchmarking” study comparing Wal-Mart with twenty of its competitors, concluding Wal-Mart promotes a lower percentage of women than its competitors.²⁷ See id.

Wal-Mart challenges Dr. Drogin’s findings and faults his decision to conduct his research on the regional level, rather than analyze the data store-by-store.²⁸ However, the proper test of whether workforce statistics should be viewed at the macro (regional) or micro (store or sub-store) level depends largely on the similarity of the employment practices and the interchange of employees at the various facilities. See Kirkland v. N.Y. State Dep’t of Corr. Servs., 520 F.2d 420, 425 (2d Cir.1975) (recognizing that the focus of analysis depends on the nature of a defendant’s employment practices); 2 Barbara Lindemann & Paul Grossman, Employment Discrimination Law 1598, 1723 (3d ed.1996).

Here, Dr. Drogin explained that a store-by-store analysis would not capture: (1) the effect of district, regional, and company-wide control over Wal-Mart’s uniform compensation policies and procedures; (2) the dissemination of Wal-Mart’s uniform compensation policies and procedures resulting from the frequent movement of store managers; or (3) Wal-Mart’s strong corporate culture. Dukes, 222 F.R.D. at 157.

In conducting its rigorous analysis of these claims, the district court first restated its standard of review. Its decision makes clear that the district court made determinations that Plaintiffs’ statistics raised common questions of fact or law only after it rigorously analyzed them, probing significantly behind the pleadings and resolving facts necessary to make determinations on Rule 23(a)(2).

Discussing the proper standard for evaluating the statistics, the district court said it rejected a full-blown merits evaluation of the evidence but had to “view[ ] the statistical evidence and testimony through the proper lens of the standards applicable to a class certification motion.” Dukes, 222 F.R.D. at 155. Though noting it would not decide the actual merits of the claims, the court did “delve[ ] into the substance of the expert testimony ... to the extent necessary to determine if it [wa]s sufficiently probative of an inference of discrimination to create a common question as to the existence of a pattern and practice of gender discrimination at Wal-Mart.” Id. This is the precise inquiry that cases such as IPO have required, and it clearly meets the standard we outline above, particularly given the depth of the district court’s review, the evidentiary posture of the case as a Title VII pattern and practice case, and the underlying procedural standard where the district court here was reviewing evidence under Rule 23(a)(2), to raise a common question, rather than, like the cases Wal-Mart cites, Rule 23(b)(3), to determine predominance. See Hnot, 241 F.R.D. at 210.

In addition to formulating a review that complied with Falcon and our precedent, the district court, contrary to Wal-Mart’s claims, did not improperly rely on out-ofeireuit cases the Second Circuit overruled in IPO. Specifically, Wal-Mart cites Visa Check and Caridad as decisions IPO rejected, noting the district court’s supposed reliance on this reasoning renders its decision error. As an initial matter, the district court did not cite Visa Check at all. See Dukes, 222 F.R.D. 137 passim. While it did cite to Caridad, nothing about its citation to Caridad was erroneous.

The district court cited Caridad five times. Two instances of this reliance can be immediately set aside as unproblematic because they relied on Caridad’s holding that excessive subjectivity in corporate policies can contribute to a finding of commonality. Dukes, 222 F.R.D. at 149-50, 154. IPO neither addressed nor overruled this holding in Caridad, and it still remains “controlling authority” in the Second Circuit. Hnot, 241 F.R.D. at 210. A third citation to Caridad supported the district court’s statement that, “although some inquiry into the substance of a case may be necessary to ascertain satisfaction of the commonality and typicality requirements of Rule 23(a), it is improper to advance a decision on the merits to the class certification stage.” Dukes, 222 F.R.D. at 144(internal quotation marks omitted) (citing Caridad, 191 F.3d at 292). This statement is entirely consistent with Falcon, our precedent, and our ruling today. The two other citations to Caridad equally show that the district court did not adopt the “some showing” standard that the Second Circuit properly rejected. Instead, the district court refused to consider the merits — it would not conclusively find whether Wal-Mart had discriminated — but it analyzed the evidence to the extent, under Rule 23(a)(2), that it had to determine whether “there are questions of law or fact common to the class.” Fed. R.Civ.P. 23(a)(2); Dukes, 222 F.R.D. at 143.²⁹

Critically, the district court did not shy away from issues overlapping with the merits; rather it devoted fifteen pages of its opinion to probing the parties’ statistics. The district court merely refused to decide the underlying merits themselves and examined evidence only to the extent necessary to satisfy itself under Rule 23(a)(2) that Plaintiffs raised common questions. In doing so, it joined dozens of other district courts in this circuit that have engaged the proper analysis, all using different wording, but all probing behind the pleadings to make determinations on the Rule 23 requirements. See, e.g., In re Cooper Cos. Sec. Litig., 254 F.R.D. 628, 641 n. 7 (C.D.Cal.2009); Bishop v. PetroChem. Transp., LLC, 582 F.Supp.2d 1290, 1305 (E.D.Cal.2008); Alexander v. JBC Legal Group, P.C., 237 F.R.D. 628, 629(D.Mont.2006); Westways World Travel, Inc., 218 F.R.D. at 230.

Turning to the factual assertions in Plaintiffs’ evidence, the court made a preliminary determination, based on the following “largely uncontested” statistics:

[W]omen working in Wal-Mart stores are paid less than men in every region, that pay disparities exist in most job categories, that the salary gap widens over time even for men and women hired into the same jobs at the same time, that women take longer to enter into management positions, and that the higher one looks in the organization the lower the percentage of women.

Dukes, 222 F.R.D. at 155. Importantly, and as instructed by Falcon, these factual determinations were arrived at after looking beyond the pleadings to Plaintiffs’ expert’s deposition. Id.

Correctly noting that descriptive statistics do not address causation, the district court then analyzed not the pleadings, but Plaintiffs’ and Wal-Mart’s statistics, finding, “In short, all of Dr. Drogin’s regressions show that gender is a statistically significant variable in accounting for the salary differentials between female class members and male employees at WalMart stores.” Id. at 156, 102 S.Ct. 2364.

The court specifically analyzed whether aggregation of statistics for regional units was proper or whether Wal-Mart was correct to insist upon a store-level evaluation. The district court first stated the relevant sub-standard: “The proper test of whether workforce statistics should be viewed at the macro (regional) or micro (store or sub-store) level depends largely on the similarity of the employment practices, and the interchange of employees, at the various facilities.” Id. at 157, 102 S.Ct. 2364 (citation omitted). It continued, “Dr. Drogin contends that it is proper to conduct the analysis on a regional level because the subjective decision-making in compensation and promotions takes place within parameters and guidelines that are highly uniform.... Plaintiffs also have shown, as discussed above, that the primary salary decision-makers, Store Managers, experience frequent relocation among the various stores.” Id. at 157, 102 S.Ct. 2364.

On appeal, Wal-Mart contends that the district court erred by not finding WalMart’s statistical evidence more persuasive than Plaintiffs’ evidence because, according to Wal-Mart, its analysis was conducted store-by-store. However, contrary to Wal-Mart’s characterization of its analysis, and the dissent’s concerns regarding statistical aggregation,³⁰ Wal-Mart’s own research was not conducted at the individual store level. Dr. Joan Haworth, WalMart’s expert, did not conduct a store-by-store analysis; instead she reviewed data at the sub-store level by comparing departments to analyze the pay differential between male and female hourly employees.³¹ Moreover, our task here is to determine whether the district court abused its discretion in finding that, based on all the evidence presented, there existed common questions of fact sufficient to justify class certification. See Gonzales v. Free Speech Coal, 408 F.3d 613, 618 (9th Cir.2005); Armstrong, 275 F.3d at 867. We are not to re-examine the relative strength or persuasiveness of the commonality evidence ourselves. Thus, even if we were to find, based on an independent review- of the record, that Wal-Mart’s statistical evidence was more persuasive than Plaintiffs’- — which we do not, in any event- — -this alone would not allow us to find that the district court improperly relied on Dr. Drogin’s testimony as a valid component of its commonality analysis or that the district court erred in its ultimate conclusion that the commonality prerequisite was satisfied. That the jury might later find Wal-Mart’s statistical evidence more persuasive does not detract from the district court’s determination, after extensive review, that Dr. Drogin’s regional analysis raises common issues appropriate for class adjudication.

Here, again, the district court followed the Supreme Court’s guidance to thoughtfully “probe behind the pleadings,” Falcon, 457 U.S. at 160, 102 S.Ct. 2364, and did not abuse its discretion when it relied on Dr. Drogin’s use and interpretation of statistical data as a valid component of its determination that Plaintiffs raised common questions. It considered Wal-Mart’s challenges to Dr. Drogin’s statistics and made the specific determination that “the Court is not persuaded that Dr. Drogin’s aggregated statistical, analysis should be rejected because he did not choose to utilize the Chow test,” as Wal-Mart had urged. Dukes, 222 F.R.D. at 158. In other words, for the purposes of class certification, the district court reasonably made the determination to credit Plaintiffs’ statistics.

Wal-Mart also claims the district court erred in determining that Wal-Mart provided little or no proper legal or factual challenge to Dr. Drogin’s analysis,³² and that, contrary to Wal-Mart’s contention, Dr. Haworth’s Store Manager survey evidence — which was stricken for failing to satisfy the standards of Rules 702 and 703 of Evidence³³ — did not undermine or contradict Dr. Drogin’s evidence.

In rejecting the inclusion of Wal-Mart’s Store Manager surveys as a challenge to Dr. Drogin’s statistics because they were not based on a scientifically valid reasoning or methodology, Dukes II, 222 F.R.D. at 197-98 (discussing Rules 702 and 703), see Daubert, 509 U.S. at 592-93, 113 S.Ct. 2786, the court noted that even if the evidence were included, “[t]he survey would not provide sufficient additional weight to Defendant’s challenge to Dr. Drogin’s analysis to sway the Court from its conclusion that his testimony supports an inference of discrimination, and thus the existence of substantial questions common to the class,” Dukes II, 222 F.R.D. at 198 n. 9 (emphasis added). In so ruling, the district court went even further into the merits than necessary to make a reasoned, justifiable determination, after proper review under the correct standard, that Plaintiffs’ claims were appropriate for class adjudication due to common questions.

Thus, because Dr. Drogin adequately explained, and the district court rigorously analyzed, why his statistical method best reflected the alleged discrimination, the court did not abuse its discretion when it credited Dr. Drogin’s analysis of statistical evidence of common discrimination questions. Nor did the district court abuse its discretion when it concluded that Dr. Drogin’s analysis supported Plaintiffs’ contention that there is a common core of facts flowing from Wal-Mart’s corporate structure and policies that affects class members generally with regard to their discrimination claims. While Plaintiffs and Wal-Mart disagree on whose findings are more persuasive, the disagreement is not one of whether Plaintiffs have asserted “common questions of law or fact.” Falcon, 457 U.S. at 157, 159, 102 S.Ct. 2364. The disagreement is the common question, and deciding which side has been more persuasive is an issue for the next phase of the litigation. Requiring even more findings and further analysis from the district court would be to force a trial on the merits at the certification stage.

Finally, and discussed further below, the district court’s review of statistics showing discrimination regarding promotions was also not an abuse of discretion. Plaintiffs and Defendant disagree over whether Dr. Drogin’s analysis of internal promotion data was proper. Specifically, both sides agreed that Wal-Mart’s actual applicant flow data for promotions during the class period was limited and contained significant gaps. Dukes, 222 F.R.D. at 162. Dr. Drogin’s statistics estimated the applicant flow by “tabulating ‘the incumbents in historical feeder jobs for each promotion.’” Id. Wal-Mart argued that the data, “while limited, is nonetheless sufficient to justify an extrapolation for all job openings during the entire class period.” Id.

The district court, addressing this statistical dispute, found Plaintiffs’ statistics “sufficient to create an inference of discrimination.” Id. at 164. In doing so, the district court found Dr. Drogin’s reasoning and methodology valid and applicable in the case. While the court noted in passing that Dr. Drogin’s statistics were “reasonable,” the court also, and more appropriately, stated that “it is well recognized that where actual applicant flow data is inadequate or unavailable, other measures of applicant flow — including but not limited to ‘feeder pools’ — are deemed acceptable so long as they are used in a reliable manner.” Id. at 162-63. In doing so the district court cited an evidence treatise, Ninth Circuit precedent, and a district court case that had accepted the same type of feeder pool methodology in a dispute of the very same experts — Dr. Drogin and Dr. Haworth. Id. at 163 & n. 37 (citing Hemmings v. Tidyman’s Inc., 285 F.3d 1174, 1185-86 (9th Cir.2002); R. Paetzold and S. Willborn, The Statistics of Discrimination §§ 4.02-4.04 (2002); Stender v. Lucky Stores, Inc., 803 F.Supp. 259, 333-34 (N.D.Cal.1992)).

Thus, properly considering Dr. Drogin’s statistics, the court made a determination on the applicant pool data. It found that “Defendant’s assertion that its approach is necessarily superior does not withstand scrutiny. Rather, Defendant’s arguments, which go to the weight of the evidence [i.e., the persuasiveness on the merits], merely highlight the presence of a significant issue affecting all class members which supports, rather than defeats, granting class certification.” Id. at 164. This determination was thus made after a rigorous analysis of the parties’ statistical claims.

In short, the district court stated the legal standard, analyzed Plaintiffs’ and Wal-Mart’s competing claims to the propriety of aggregating statistics on a regional level and addressing Wal-Mart’s missing applicant flow data, noted Plaintiffs have “shown” reasons to accept their statistics, dismissed Wal-Mart’s statistical challenges, demonstrated these finding were supported by relevant Ninth Circuit precedent (and an identical dispute involving the same two experts), rejected WalMart’s reliance on a district court case purportedly explaining why the sub-store statistical analysis was proper, and, finally, determined that at the class certification stage all of this analysis was sufficient to support Dr. Drogin’s analysis and raise questions of law or fact common to the class. This searching analysis was solid, well founded and in no way an abuse of discretion.

(4) Anecdotal Evidence

Circumstantial and anecdotal evidence of discrimination is commonly used in Title VII “pattern and practice” cases to bolster statistical proof by bringing “the cold numbers convincingly to life.” Teamsters, 431 U.S. at 339, 97 S.Ct. 1843; see also Rudebusch v. Hughes, 313 F.3d 506, 517 (9th Cir.2002). Wal-Mart contends that the district court erred by con-eluding that the anecdotal evidence, presented by Plaintiffs in the form of sworn declarations, supported a finding of commonality.³⁴ Wal-Mart maintains that the declarations depict a handful of “widely divergent” events that cannot be deemed probative or representative of discrimination in pay or management-track promotions.

In their declarations, the potential class members testified to being paid less than similarly situated men, being denied or delayed in receiving promotions in a disproportionate manner when compared with similarly situated men, working in an atmosphere with a strong corporate culture of discrimination, and being subjected to various individual sexist acts. The district court credited this evidence.

Wal-Mart argues that 120 declarations cannot sufficiently represent a class of this size. However, we find no authority requiring or even suggesting that a plaintiff class submit a specific number of declarations for such evidence to have any value. Moreover, the district court did not state that this anecdotal evidence provided sufficient proof to establish commonality by itself, but merely noted such evidence provides support for Plaintiffs’ contention that commonality is present. See Dukes, 222 F.R.D. at 166 (“This anecdotal evidence, in combination with the other evidence previously discussed, further supports an inference that [WalMart’s] policies and procedures have the effect of discriminating against Plaintiffs in a common manner.”).³⁵ Because the combination of these declarations and Plaintiffs’ other evidence, discussed below, raise an inference of common discriminatory experiences, the district court did not abuse its discretion when it considered Plaintiffs’ anecdotal evidence.³⁶

Finally, in arguing against certification based on affidavits, Plaintiffs’ personal allegations, statistics, and expert testimony, the dissent intersperses references of Falcon immediately with discussion of Cooper v. Federal Reserve Bank, 467 U.S. 867, 879, 104 S.Ct. 2794, 81 L.Ed.2d 718 (1984), claiming “individual stories do not constitute significant proof that Wal-Mart has adopted a general policy of discrimination or that such a policy prevails” at WalMart. Dissent at 635. This criticism, and particularly its temporary suspension of consideration of the statistical evidence, is indicative of the dissent’s repeated attempts to go beyond the Supreme Court’s concern with the need for “more precise pleadings” in class actions, Falcon, 457 U.S. at 160-61, 102 S.Ct. 2364 (citing Johnson, 417 F.2d at 1125), and the Court’s command that “sometimes it may be necessary for the court to probe behind the pleadings before coming to rest on the certification question,” id. at 160, 102 S.Ct. 2364, in an effort to reframe the question as one on whether Plaintiffs can actually succeed on the merits. Cf. Eisen, 417 U.S. at 177-78, 94 S.Ct. 2140(explaining that the plaintiffs’ likelihood of success on the merits is not part of deciding whether certification is proper).

Contrary to the dissent’s proposed standard, Plaintiffs here need not “establish a prima facie case” on the merits. Dissent at 643. Class certification in Cooper had already taken place and, in fact, the case had gone through a trial. 467 U.S. at 872, 104 S.Ct. 2794. The decision for the Supreme Court was whether the class’s loss at trial precluded a class member from maintaining his own, separate civil action subsequent to the loss. Id. at 869, 104 S.Ct. 2794. This is an entirely different posture than the present case presents, and the standard for winning a claim under McDonnell Douglas Corp. v. Green, 411 U.S. 792, 93 S.Ct. 1817, 36 L.Ed.2d 668 (1973), is, of course, different than that for certifying a class action. See Cooper, 467 U.S. at 875-76, 104 S.Ct. 2794; Falcon, 457 U.S. at 160-61, 102 S.Ct. 2364.

b. Subjective Decision Making

As discussed above, the district court found substantial evidence suggesting common pay and promotion policies among Wal-Mart’s many stores. See Dukes, 222 F.R.D. at 148-51. The court also reasoned that Wal-Mart’s decision to permit its managers to utilize subjectivity in interpreting those policies offers additional support for a commonality finding. See id. Relying on Sperling v. Hoffmann-La Roche, Inc., 924 F.Supp. 1346 (D.N.J.1996), Wal-Mart challenges the latter conclusion, contending that managers’ discretionary authority does not support a finding of commonality because “[decentralized, discretionary decisionmaking is not inherently discriminatory.”

It is well established that subjective decision making is a “ready mechanism! ] for discrimination” and that courts should scrutinize it carefully. Sengupta v. Morrison-Knudsen Co., 804 F.2d 1072,1075 (9th Cir.1986) (internal quotation marks omitted). Wal-Mart is correct that discretionary decision making by itself is insufficient to meet Plaintiffs’ burden. The district court recognized this, noting that managerial discretion is but one of several factors that supported a finding of commonality. See Dukes, 222 F.R.D. at 149-50 (“And while the presence of excessive subjectivity, alone, does not necessarily create a common question of fact, where, as here, such subjectivity is part of a consistent corporate policy and supported by other evidence giving rise to an inference of discrimination, courts have not hesitated to find that commonality is satisfied.”).

Wal-Mart is incorrect, however, that decentralized, subjective decision making cannot contribute to a common question of fact regarding the existence of discrimination. See Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. 2364. Indeed, courts from around the country have found “[ajllegations of similar discriminatory employment practices, such as the use of entirely subjective personnel processes that operate to discriminate, [sufficient to] satisfy the commonality and typicality requirements of Rule 23(a).” Shipes v. Trinity Indus., 987 F.2d 311, 316 (5th Cir.1993); see also supra note 7 and cases cited therein.

Plaintiffs produced substantial evidence of Wal-Mart’s centralized firm-wide culture and policies, see Dukes, 222 F.R.D. at 151-54, thus providing a nexus between the subjective decision making and the considerable statistical evidence demonstrating a pattern of lower pay and fewer promotions for female employees. See id. at 154-65; see also Reid v. Lockheed Martin Aeronautics Co., 205 F.R.D. 655, 670-72 (N.D.Ga.2001) (subjective decision making may give rise to an inference of discrimination where evidence shows a nexus .between decision making and discrimination). Therefore, for the reasons stated above, there was no abuse of discretion in determining that Wal-Mart’s subjective decisionmaking policy provides support for Plaintiffs’ contention that commonality exists among potential class members.

c. Commonality Conclusion

The district court’s analysis of Rule 23(a)(2) complies with the standard the Supreme Court has set down and we have explained today for a district court adjudicating a motion for class certification. Plaintiffs’ factual evidence, expert opinions, statistical evidence, and anecdotal evidence provide sufficient support to raise the common question whether Wal-Mart’s female employees nationwide were subjected to a single set of corporate policies (not merely a number of independent discriminatory acts) that may have worked to unlawfully discriminate against them in violation of Title VII. Evidence of Wal-Mart’s subjective decision-making policies suggests a common legal or factual question regarding whether Wal-Mart’s policies or practices are discriminatory. Many other courts have reached the same conclusion based on similar evidence. See, e.g., Staton, 327 F.3d at 955; Shipes, 987 F.2d at 316; Cox v. Am. Cast Iron Pipe Co., 784 F.2d 1546, 1557 (11th Cir.1986); Segar v. Smith, 738 F.2d 1249, 1276 (D.C.Cir.1984).

3. Typicality

As an initial matter, Plaintiffs contend that Wal-Mart has waived a chailenge to the district court’s typicality finding by failing to offer specific objections to the district court’s typicality determination. However, because Wal-Mart refers, somewhat obliquely, to the typicality factor in its opening brief and because typicality and commonality are similar and tend to merge, see Falcon, 457 U.S. at 157 n. 13, 102 S.Ct. 2364,³⁷ we conclude that WalMart did not waive its opportunity to challenge the district court’s findings with regard to typicality.

Thus, although Wal-Mart did not raise a specific challenge, it nevertheless raised a general objection to the district court’s conclusion that Plaintiffs’ evidence satisfies the typicality requirement. As discussed below, to satisfy the typicality pre-requisite, Plaintiffs must demonstrate that their claims and their class representatives are sufficiently typical of the class.

a. Plaintiffs’ Claims Are Sufficiently Typical

Rule 23(a)(3) requires that “the claims or defenses of the representative parties are typical of the claims or defenses of the class.” Fed.R.Civ.P. 23(a)(3). We stated in Hanlon that, “[u]nder the rule’s permissive standards, representative claims are ‘typical’ if they are reasonably coextensive with those of absent class members; they need not be substantially identical.” 150 F.3d at 1020; see also Staton, 327 F.3d at 957(stating that “a class representative [from each job category] for each type of discrimination claim alleged ... is not necessary”).

Thus, we must consider whether the injury allegedly suffered by the named plaintiffs and the rest of the class resulted from the same allegedly discriminatory practice. See Staton, 327 F.3d at 957. Even though individual employees in different stores with different managers may have received different levels of pay or may have been denied promotion or promoted at different rates, because the discrimination they claim to have suffered occurred through alleged common practices — e.g., excessively subjective decision making in a corporate culture of uniformity and gender stereotyping — the district court did not abuse its discretion by finding that their claims are sufficiently typical to satisfy Rule 23(a)(3).

b. Plaintiffs’ Representatives Are Sufficiently Typical of the Class

Typicality requires that the named plaintiffs be members of the class they represent. See Falcon, 457 U.S. at 156, 102 S.Ct. 2364. There is no dispute that the class representatives are “typical” of the hourly class members, because almost all of the class representatives hold hourly positions. Instead, Wal-Mart contends that the class representatives are not typical of all female in-store managers beeáuse only one of the class representatives holds a salaried management position, and she holds a somewhat low-level position.

However, because all female employees faced the same alleged discrimination, the lack of a class representative for each management category does not undermine Plaintiffs’ certification goal. See Hartman v. Duffey, 19 F.3d 1459, 1471 (D.C.Cir. 1994) (An employee can challenge discrimination in “different job categories where the primary practices used to discriminate in the different categories are themselves similar. While it may be prudent to have the class divided into sub-classes represented by a named plaintiff from each of the differing job categories, it would not be necessary to the validity of the class certification to do so.”); Paxton v. Union Nat’l Bank, 688 F.2d 552, 562 (8th Cir.1982) (“Typicality is not defeated because of the varied promotional opportunities at issue, or the differing qualifications of the plaintiffs and class members.”).

In addition, because the range of managers in the proposed class is limited to managers working in Wal-Mart’s stores and excludes those at regional or national offices, it is not an excessively diverse class; a named plaintiff occupying a lower-level, salaried, in-store management position is sufficient to satisfy the “permissive” typicality requirement. Staton, 327 F.3d at 957(recognizing that, “ ‘[u]nder the rule’s permissive standards,’ ” plaintiffs are not required to offer a class representative for each type of discrimination claim alleged (quoting Hanlon, 150 F.3d at 1020)).

Because Plaintiffs’ claims and Plaintiffs’ representatives are sufficiently typical of the class, the district court acted within its discretion when it found that Plaintiffs satisfied the typicality prerequisite.

4. Adequate Representation

Rule 23(a)(4) permits certification of a class action only if “the representative parties will fairly and adequately protect the interests of the class.” Fed.R.Civ.P. 23(a)(4). This factor requires: (1) that the proposed representative Plaintiffs do not have conflicts of interest with the proposed class, and (2) that Plaintiffs are represented by qualified and competent counsel. See Hanlon, 150 F.3d at 1020; see also Molski v. Gleich, 318 F.3d 937, 955 (9th Cir.2003).

Before the district court, Wal-Mart argued that Plaintiffs cannot satisfy this factor because of a conflict of interest between female in-store managers who are both plaintiff class members and decision-making agents of Wal-Mart. Relying on Staton, the district court recognized that courts need not deny certification of an employment class simply because the class includes both supervisory and non-supervisory employees. See Dukes, 222 F.R.D. at 168; see also Staton, 327 F.3d at 958-59. This decision was not an abuse of discretion. Finally, because Wal-Mart does not challenge the district court’s finding that Plaintiffs’ class representatives and counsel are adequate, we need not analyze this factor.

5. Commonality, Typicality, and Manageability of Promotion Claims

The district court certified Plaintiffs’ promotion claims for injunctive and declaratory relief and punitive damages, but concluded that “manageability concerns” required that “any lost pay remedy on Plaintiffs’ promotion claim would be limited to that subset of the class for whom objective applicant data exists.” Dukes, 222 F.R.D. at 183. Wal-Mart objects to the certification of Plaintiffs’ promotion claims, even with this limitation, on commonality and manageability grounds. Plaintiffs, on the other hand, object to the district court’s limitation of eligibility for back pay to positions for which objective applicant data exist.

Title VII pattern and practice class actions frequently include both salary and promotion claims. See, e.g., Bazemore v. Friday, 478 U.S. 385, 406, 106 S.Ct. 3000, 92 L.Ed.2d 315 (1986) (per curiam); Cooper, 467 U.S. at 870, 104 S.Ct. 2794. As discussed above, the district court found that Plaintiffs here have provided evidence sufficient to support their contention that company-wide corporate practices and policies — including excessive subjectivity in personnel decisions, gender stereotyping, and maintenance of a strong corporate culture — affected both compensation and promotion of all Plaintiffs in a common manner. Dukes, 222 F.R.D. at 166. Although there may be some variation among individual promotion decisions, variation does not prevent class certification of common issues of fact. See Hanlon, 150 F.3d at 1019 (“Rule 23(a)(2) has been construed permissively. All questions of fact and law need not be common to satisfy the rule.”). Given its findings regarding WalMart’s company-wide promotion policies and minimum guidelines, see Dukes, 222 F.R.D. at 148-49, the district court did not abuse its discretion by concluding that there was a sufficient common factual basis among putative class members’ promotion claims to satisfy Rule 23(a)(2).

We also affirm the district court’s conclusion that a back pay remedy for promotion claims would be manageable only “with respect to those positions for which objective applicant data is available to document class member interest.” Dukes, 222 F.R.D. at 183. Wal-Mart’s extensive database containing information on each employee individually with respect to job history, seniority, job review ratings, and many other factors, see id. at 183-85, would enable a determination of each class member’s qualifications for promotion without the need for potentially unmanageable individualized hearings. The district court did not abuse its discretion by limiting eligibility for back pay, however, because allowing class members to demonstrate post hoc their previous interest in promotions might indeed be unmanageable. See id. at 181-82.

6. Rule 23(a) Conclusion

Based on the evidence before the district court, which it rigorously analyzed pursuant to Falcon, 457 U.S. at 161, 102 S.Ct. 2364, and the standard herein described, the district court did not abuse its discretion when it found that the Rule 23(a) elements were satisfied.

B. Rule 23(b)

As mentioned earlier, Plaintiffs moved to certify the class under Rule 23(b)(2), which requires showing that “the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief ... is appropriate respecting the class as a whole.” Fed.R.Civ.P. 23(b)(2).³⁸ The district court agreed with Plaintiffs. See Dukes, 222 F.R.D. at 170-71. Wal-Mart argues the district court merely “paid lip service” to Rule 23(b)(2) and erred in certifying the class under Rule 23(b)(2) because claims for monetary relief predominate over claims for injunctive and declaratory relief.

Rule 23(b)(2) is not appropriate for all classes and “does not extend to cases in which the appropriate final relief relates exclusively or predominantly to money damages.” Fed.R.Civ.P. 23(b)(2) advisory committee’s note to 1966 amends., 39 F.R.D. at 69, 102; see also Zinser, 253 F.3d at 1195 (“Class certification under Rule 23(b)(2) is appropriate only where the primary relief sought is declaratory or injunctive.”). An interpretation of Rule 23(b)(2) that prevented any claim for monetary relief would render this advisory committee requirement redundant or irrelevant.

We first turn to the appropriate standard for determining when monetary relief “predominates” over declaratory and injunctive relief and therefore precludes certification of a Rule 23(b)(2) class. We have previously joined the Second Circuit in adopting a test that focuses on the plaintiffs’ subjective intent in bringing a lawsuit. See Molski, 318 F.3d at 950; Robinson v. Metro-North Commuter R.R. Co., 267 F.3d 147, 164 (2d Cir.2001). In contrast, several other circuits use the “incidental damages standard” that was first enunciated by the Fifth Circuit in Allison v. Citgo Petroleum Corp., 151 F.3d 402, 415-16 (5th Cir.1998). See, e.g., Reeb v. Ohio Dep’t of Rehab. & Corr., 435 F.3d 639, 646-51 (6th Cir.2006); Murray v. Auslander, 244 F.3d 807, 812 (11th Cir. 2001); Lemon v. Int’l Union of Operating Eng’rs Local No. 139, 216 F.3d 577, 580-81 (7th Cir.2000). Under the Allison approach, monetary relief predominates over other forms of relief “unless it is incidental to requested injunctive or declaratory relief.” See Allison, 151 F.3d at 415.

We see no need to employ either approach, which are both, essentially, glosses on the text of the Advisory Committee’s Note’s statement that Rule 23(b)(2) “does not extend to cases in which the appropriate final relief relates exclusively or predominantly to money damages.” Merriam-Webster defines “predominant” as “having superior strength, influence, or authority: prevailing.” Merriam-Webster’s Collegiate Dictionary 978 (11th ed.2004). To be certified under Rule 23(b)(2), therefore, a class must seek only monetary damages that are not “superior [in] strength, influence, or authority” to injunctive and declaratory relief.

An analysis of a plaintiffs subjective intent in bringing a suit, as required by the standard set forth in Molski, is, at best, an incomplete method for answering this question. By eschewing consideration of the practical impact of a request for monetary relief on the litigation itself, the sole emphasis on the plaintiffs intent ignores important indicators of the “strength, influence, [and] authority” of a request for specific monetary relief. In short, Molski’s focus on subjective intent and its concomitant failure to consider the pragmatic impact of a request for monetary relief render it fatally flawed.

The Molski approach is troubling for the additional reason that it requires courts to engage in a nebulous and imprecise inquiry into the plaintiffs’ intent in bringing a particular suit. Only in those cases in which a request for injunctive relief is obviously a ruse will this inquiry provide a clear answer. More often than not, we suspect that the answer will be equivocal and, therefore, essentially an entirely discretionary one.

Although the standard set forth in Allison is an objective one that does consider the practical effect of a request for monetary damages, it suffers from a different deficiency. By requiring monetary relief to be no more than “incidental” to injunctive or declaratory relief, the Allison approach is in direct conflict with the text of the Advisory Committee’s Note, which forbids certification under Rule 23(b)(2) if monetary relief is the “predominant” form of relief. See 151 F.3d at 412-16. “Predominant” is not synonymous with “more than incidental.” As a result, the Allison approach would preclude the certification of some Rule 23(b)(2) classes that the drafters of the Rules intended to allow.

Allison’s standard also “usurps the district courts’ authority granted by Rule 23 ... to rigorously analyze the case, probe behind the pleadings if necessary, and exercise its own discretion within the framework of the rules in determining whether the action is to be so maintained.” Id. at 431(Dennis, J., dissenting); see also Robinson, 2Í61 F.3d at 165 (describing Allison’ s incidental damages test as a “one-size-fits-all approach”). Such a limitation is in direct conflict with the Federal Rules’ intention that district courts be afforded discretion. The predominance test suggested by the Advisory Committee’s Note, by contrast, is consistent with the wide discretion given to district courts in making the class certification decision and accomplishing Title VII’s remedial objectives.

To the extent Molski required the district court to inquire only into the intent of the plaintiffs and focus primarily on determining whether reasonable plaintiffs would bring suit to obtain injunctive or declaratory relief even in the absence of a possible monetary recovery, see Molski, 318 F.3d at 950 & n. 15, it is overruled. In light of the inadequacy of both the Allison and Molski approaches, we adopt instead the standard that Rule 23(b)(2)’s drafters straightforwardly indicated: Rule 23(b)(2) certification is not appropriate where monetary relief is “predominant” over injunctive relief or declaratory relief. To determine whether monetary relief predominates, a district court should consider, on a case-by-case basis, the objective “effect of the relief sought” on the litigation. See Allison, 151 F.3d at 416. Factors such as whether the monetary relief sought determines the key procedures that will be used, whether it introduces new and significant legal and factual issues, whether it requires individualized hearings, and whether its size and nature — as measured by recovery per class member — raise particular due process and manageability concerns would all be relevant, though no single factor would be determinative.

Under this standard, as discussed more fully below, the district court’s decision to include claims for back pay in a class certified under Rule 23(b)(2) was not an abuse of its discretion. On the other hand, the district court did abuse its discretion by failing to analyze whether certifying Plaintiffs’ punitive damages claims under Rule 23(b)(2) caused monetary damages to predominate, notwithstanding its decision to require notice and an opportunity for Plaintiffs to opt-out of the punitive damages claims.

1. Wal-Mart’s Evidence Does Not Undermine Plaintiffs’ Claim That Injunctive and Declaratory Relief Predominate

Wal-Mart first asserts that the district court “failed to even evaluate” Rule 23(b)’s requirement that the challenged conduct be generally applicable to the class. WalMart’s contention that its “unrebutted” statistics demonstrate that there is no evidence of pervasive discrimination that would justify injunctive relief and that, therefore, the “challenged conduct” does not affect all members, is simply not persuasive. As explained above, Wal-Mart’s evidence was rebutted by Plaintiffs to the extent that Plaintiffs’ evidence and theories remain viable at this pre-merits analysis stage. Further, the issue before us is whether monetary relief predominates, not whether Plaintiffs will ultimately prevail.

2. The Large Size of the Class Does Not Undermine Plaintiffs’ Claim That Injunctive and Declaratory Relief Predominate

Wal-Mart contends that monetary claims necessarily predominate because this case involves claims that may amount to billions of dollars. However, such a large amount is principally a function of Wal-Mart’s size, and the predominance test turns on the primary goal and nature of the litigation — not the theoretical or possible size of the total damages award. A comparison between the amount of monetary damages available for each plaintiff and the importance of injunctive and declaratory relief for each is far more relevant to establishing predominance than the total size of a potential monetary award for the class as a whole. As the district court stated, “focusing on the potential size of a punitive damage award would have the perverse effect of making it more difficult to certify a class the more egregious the defendant’s conduct or the larger the defendant. Such a result hardly squares with the remedial purposes of Title VII.” Dukes, 222 F.R.D. at 171. Because WalMart has not shown that the size of the monetary request undermines Plaintiffs’ claim that injunctive and declaratory relief predominate, we find that Wal-Mart’s argument fails.

3. Request for Back Pay Does Not Undermine Plaintiffs’ Claim That Injunctive and Declaratory Relief Predominate

Wal-Mart asserts that Plaintiffs’ request for back pay weighs against certification because it proves that claims for monetary relief predominate. The district court reasoned that back pay “is recoverable as an equitable, make-whole remedy in employment class actions notwithstanding its monetary nature.” Dukes, 222 F.R.D. at 170. Wal-Mart contends that the district court erred by failing to recognize that back pay, whether “equitable” or not, is still a form of monetary relief.

Wal-Mart’s argument is without merit. Although the circuits have adopted different tests for determining when monetary relief predominates, every circuit to have addressed the issue has acknowledged that Rule 23(b)(2) does allow for some claims for monetary relief.³⁹ More to the point, it is equally well accepted, even by circuits that are generally restrictive in certifying classes seeking monetary damages under Rule 23(b)(2), that a request for back pay in a Title VII case is fully compatible with the certification of a Rule 23(b)(2) class. See, e.g., Allison, 151 F.3d at 415; see also Thorn, 445 F.3d at 331; Cooper, 390 F.3d at 720; Coleman, 296 F.3d at 449. As the Fifth Circuit stated in Allison, “final injunctive relief [was] appropriate and the defendant’s liability for back pay [was] rooted in grounds applicable to all members of the defined class. Under these circumstances the award of back pay, as one element of the equitable remedy, conflicts in no way with the limitations of Rule 23(b)(2).” 151 F.3d at 415 (internal quotation marks omitted).⁴⁰

Courts have provided two justifications for the conclusion that a request for back pay is consistent with certification under Rule 23(b)(2). First, appellate courts have noted that, “in the Title VII context, awards of backpay do not predominate over the injunctive remedies available because the ‘calculation of back pay generally involves [relatively unjcomplicated factual determinations and few [] individualized issues.’ ” Thom, 445 F.3d at 331-32(alter-ations in original) (quoting Coleman, 296 F.3d at 449). The second rationale is that back pay is “an integral component of Title VII’s ‘make whole’ remedial scheme,” see Allison, 151 F.3d at 415, a scheme to which the drafters of the Federal Rules of Civil Procedure clearly intended Rule 23(b)(2) to apply. Fed.R.Civ.P. 23(b)(2) advisory committee’s note to 1966 amends., 39 F.R.D. at 102 (stating that the principal category of cases certifiable under Rule 23(b)(2) are “actions in the civil-rights field where a party is charged with discriminating unlawfully against a class”).

We find this reasoning persuasive and therefore join the consensus view that a request for back pay in a Title VII case is fully consistent with the certification of a Rule 23(b)(2) class action.⁴¹ Accordingly, we hold that the district court did not abuse its discretion when it concluded, like many courts before it, that this discrimination class action was certifiable under Rule 23(b)(2) notwithstanding Plaintiffs’ prayer for back pay relief.

4. Monetary Relief May Predominate With Respect to Plaintiffs’ Bifurcated Punitive Damages Claims

In determining whether “appropriate final relief relates exclusively or predominantly to money damages,” Fed.R.Civ.P. 23(b)(2), advisory committee’s note to 1996 amends., 39 F.R.D. at 102, claims for punitive damages weigh against certification under Rule 23(b)(2) because punitive damages are, of course, neither injunctive nor declaratory relief. The view that punitive damages can never be awarded consistent •with Rule 23(b)(2), however, has not been adopted by this circuit.⁴² On the other hand, this court also has not approved class certification under Rule 23(b)(2) involving the potential for substantial punitive damages.

To decide whether certification under Rule 23(b)(2) is appropriate, per the standard described in Part II, a district court must squarely face and resolve the question of whether the monetary damages sought by the plaintiff class predominate over the injunctive and declaratory relief. If so, then the court may either deny certification under Rule 23(b)(2) or bifurcate the proceedings by certifying a Rule 23(b)(2) class for equitable relief and a separate Rule 23(b)(3) class for damages. See Molski, 318 F.3d at 951 n. 16 (citing Jefferson v. Ingersoll Int'l Inc., 195 F.3d 894, 897-98 (7th Cir.1999)).⁴³

Here, the district court certified Plaintiffs’ punitive damages claims as a separate class, but did so under Rule 23(b)(2) rather than Rule 23(b)(3), exercising its discretion to impose additional requirements for notice and the opportunity to opt-out of this separate Rule 23(b)(2) class. See Dukes, 222 F.R.D. at 173 (“[N]otice and the opportunity to opt-out can be provided in a(b)(2) class action” and “shall be provided to the plaintiff class with respect to Plaintiffs’ claim for punitive damages.”). As the district court noted, its discretion to require notice and the opportunity to opt-out in a Rule 23(b)(2) class action is well established. Id. (citing In re Monumental Life Ins. Co., 365 F.3d at 417; Molski, 318 F.3d at 951 n. 16; Robinson, 267 F.3d at 165-67; Jefferson, 195 F.3d at 898-99).⁴⁴

Notwithstanding its decision to require notice and an opportunity for opt-out, the district court abused its discretion by certifying the punitive damages claims under Rule 23(b)(2) without first undertaking an analysis of whether certification of the claim for punitive damages (in addition to injunctive and declaratory relief as well as back pay) rendered the final relief “predominantly” related to monetary damages. See Fed.R.Civ.P. 23(b)(2), advisory committee’s note to 1996 amends., 39 F.R.D. at 102. Although notice and an opportunity for Plaintiffs to opt-out of the punitive damages claims alleviated some of the concerns regarding cohesiveness and due process that substantial punitive damages claims potentially raise, requiring these procedures does not remove the district court’s obligation to determine whether each of the requirements of certification under either Rule 23(b)(2) or (b)(3) is met.

On remand, the district court must determine whether certification under Rule 23(b)(2) of the punitive damages claims would cause monetary relief to predominate. As discussed above, the district court should not limit its inquiry to the former Molski factors, but should also consider any other factors relevant to whether monetary relief predominates when determining if certification under Rule 23(b)(2) is appropriate.⁴⁵ We note at least four factors present in this case that are relevant to the question whether monetary relief predominates, but the district court need not limit its inquiry to these factors alone.

First, the inclusion of a punitive damages request means that the key issue in this case, Wal-Mart’s liability, will be decided by a jury, rather than a judge. This significant procedural change weighs in favor of finding that monetary relief would predominate if the punitive damages claims are certified, although it is not dis-positive.⁴⁶

Second, Plaintiffs’ request for punitive damages introduces a new and substantial factual issue. To recover punitive damages, Plaintiffs must show not only that Wal-Mart engaged in a pattern or practice of discrimination, but also that it did so “with malice or with reckless indifference to the federally protected rights of’ Plaintiffs. 42 U.S.C. § 1981a(b)(1). This additional factual question will likely require the Plaintiffs to introduce significant evidence and legal argument that would not have otherwise been necessary; the need for such extra evidence and argument weighs in favor of a finding that monetary relief predominates.

Third, the size of a potential punitive damages award, measured on an individual basis, could be quite significant. Title VII permits a punitive damage award of up to $300,000 per employee. See id. § 1981a(b)(3). Such a large potential award raises due process and manageability concerns. Although the district court’s decision to provide notice and opt-out to class members alleviates some of these concerns, the size of the potential award per class member in this case militates in favor of a finding that monetary relief predominates, triggering the need for other safeguards applicable when a class is certified under Rule 23(b)(3), rather than Rule 23(b)(2).

Finally, we note that, unlike in other punitive and compensatory damages cases, this case does not require individualized punitive damages determinations. Plaintiffs’ theory of liability is a class-wide theory that is based on a company policy that allegedly affects all class members in a similar way. See Allison, 151 F.3d at 417(leaving open the possibility that a punitive damage class could be certified under 23(b)(2) where the “plaintiffs challenge broad policies and practices” and “contend that each plaintiff was affected by th[o]se polices and practices in the same way”). While this factor counsels against a finding that punitive damages predominate, it also is not necessarily dispositive.

In sum, we hold that the district court abused its discretion when it certified a Rule 23(b)(2) class including punitive damages without first undertaking a comprehensive analysis of whether the inclusion of such damages in this case causes monetary relief to predominate. To allow for further pertinent fact-finding, we remand the certification of the bifurcated punitive damages claims to the district court to consider whether certification is proper under Rule 23(b)(2). We decline to prejudge the outcome of this determination because the question of whether monetary relief predominates should be answered by the district court in the first instance, consistent with the Federal Rules’ intention to vest district courts with significant discretion.

If the district court concludes that the punitive damage class in this action cannot be certified under Rule 23(b)(2), it should also consider whether class certification of the punitive damages claims is appropriate under Rule 23(b)(3). We and a number of other courts of appeals have endorsed such “hybrid certification” of Rule 23(b)(2) and Rule 23(b)(3) classes in one action, particularly in civil rights cases that may involve significant monetary damages. See, e.g., Molski, 318 F.3d at 951 n. 16; Jefferson, 195 F.3d at 898; Eubanks, 110 F.3d at 96.

Under this hybrid approach, the highly cohesive Rule 23(b)(2) phase of the proceedings, including liability, can be adjudicated without the costly class notice and opt-out process required under Rule 23(b)(3). In order to protect the due process interests of absent class members, however, notice and opt-out is required for the Rule 23(b)(3) punitive damages proceedings. See Fed.R.Civ.P. 23(c)(2), advisory committee’s note to 2003 amends. (“If a Rule 23(b)(3) class is certified in conjunction with a(b)(2) class, the (c)(2)(B) notice requirements must be satisfied as to the (b)(3) class.” (emphasis added)); Eubanks, 110 F.3d at 96 (“[Hybrid certification] effectively grant[s] (b)(3) protections ... at the monetary relief stage.”); Diaz v. Hillsborough County Hosp. Auth., 165 F.R.D. 689, 695 (M.D.Fla.1996) (explaining that in Stage I, the court will resolve liability using Rule 23(b)(2) procedures and, if liability is established, adjudicate damages using “opt out” procedures in Rule 23(b)(3)). This procedure retains the benefits of Rule 23(b)(2) and Rule 23(b)(3) for all parties, and “promotes both ease of administration and the underlying principles of Rule 23.” See Williams v. Local No. 19, Sheet Metal Workers Int’l Ass’n, 59 F.R.D. 49, 56 (E.D.Pa.1973) (explaining that the liability issue will be litigated first and, if the plaintiffs are successful, notice will be given for the damages proceeding).

We do not express a view on whether the punitive damages claims in this case meet the Rule 23(b)(3) requirements or, if so, how the district court should manage the class proceedings in order to comply with Seventh Amendment principles. We believe, however, that this might be a case in which “[djivided certification also is worth consideration.” Jefferson, 195 F.3d at 898.

5. Class Certification May Not be Proper as to Class Members Who Were Not Wal-Mart Employees as of the Date Plaintiffs’ Complaint Was Filed

Wal-Mart’s final contention is that, because a substantial number of the putative class members no longer work for WalMart — and, thus, no longer have standing to seek injunctive or declaratory relief— such relief cannot possibly predominate over monetary relief for purposes of certifying this class under Rule 23(b)(2).

We agree with Wal-Mart to this extent: those putative class members who were no longer Wal-Mart employees at the time Plaintiffs’ complaint was filed do not have standing to pursue injunctive or declaratory relief. See Walsh v. Nev. Dep’t of Human Res., 471 F.3d 1033, 1037 (9th Cir.2006) (recognizing that former employees lack standing to seek injunctive relief because they “would not stand to benefit from an injunction requiring the anti-discriminatory policies [to cease] at [their] former place of work”); ACLU of Nev. v. Lomax, 471 F.3d 1010, 1015 (9th Cir.2006) (“When evaluating whether [the standing] elements are present, we must look at the facts ‘as they exist at the time the complaint was filed.’ ” (quoting Lujan v. Defenders of Wildlife, 504 U.S. 555, 569 n. 4, 112 S.Ct. 2130, 119 L.Ed.2d 351 (1992))). Under these circumstances, it is difficult to say that monetary relief does not predominate with respect to claims by plaintiffs who lack standing to seek injunctive or declaratory relief.

However, this does not mean that former employees are ineligible to receive any form of relief. Although women who were not employed by Wal-Mart as of June 8, 2001, the date on which the complaint was filed, do not have standing to seek injunctive or declaratory relief, they may be eligible to receive back pay and punitive damages. The district court may, in its discretion, certify a separate Rule 23(b)(3) class of former employees for back pay and punitive damages.

Putative class members who were still Wal-Mart employees as of June 8, 2001, do have standing to seek the injunctive and declaratory relief requested in the complaint.⁴⁷ See Lomax, 471 F.3d at 1015. We are satisfied that class certification under Rule 23(b)(2) was not an abuse of discretion, at least as to those Plaintiffs’ claims, for the reasons explained in Parts II.B.1-II.B.4.

In summary, we affirm the district court’s certification of a Rule 23(b)(2) class insofar as the class consists of current employees (as of the date the complaint was filed), with respect to claims for injunctive relief, declaratory relief, and back pay. On remand, the district court should analyze whether certification under Rule 23(b)(2) or Rule 23(b)(3) is appropriate for the punitive damages claims and whether an additional class or classes may be appropriate under Rule 23(b)(3) with respect to the claims of former employees. The court may, if appropriate, certify an additional class or classes under Rule 23(b)(3).

IV. Class Action Can Proceed In a Way that is Both Manageable and In Accordance With Due Process

The district court was cognizant of the large size of the class when it concluded that the class size was not unmanageable. See Dukes, 222 F.R.D. at 173. Indeed, the district court acknowledged that, “[wjhile courts possess wide discretion to flexibly respond to manageability issues that may arise during the course of a class action, this Court must be confident that such issues will not be of such a magnitude as to defy its ability to oversee this case in a responsible and reasonable manner.” Id. (citation omitted). After “giv[ing] these matters considerable thought and deliberation,” the district court concluded that, with one minor exception,⁴⁸ “the size of the class would not present undue obstacles to managing” this class action. Id.

To demonstrate the manageability of the class action, the district court outlined a trial plan based, in large part, on how other courts have handled similarly large and complex class action suits.⁴⁹ WalMart, supported by a number of business-related amici,⁵⁰ contend that at least some aspects of this trial plan violate their due process rights, as well as section 706(g)(2) of Title VII,⁵¹ the Rules Enabling Act,⁵² and the Supreme Court’s decision in Teamsters, 431 U.S. at 359-60, 97 S.Ct. 1843.⁵³

At this stage, we express no opinion regarding Wal-Mart’s objections to the district court’s tentative trial plan (or that trial plan itself), but simply note that, because there are a range of possibilities— which may or may not include the district court’s proposed course of action — that would allow this class action to proceed in a manner that is both manageable and in accordance with due process, manageability concerns present no bar to class certification here.

For example, in Hilao v. Estate of Marcos, 103 F.3d 767, 782-87 (9th Cir.1996), the district court employed the following procedure to determine the amount of compensatory damages due the plaintiffs in a large class action:⁵⁴

In all, 10,059 claims were received. The district court ruled 518 of these claims to be facially invalid, leaving 9,541 claims. From these, a list of 137 claims was randomly selected by computer. This number of randomly selected claims was chosen on the basis of the testimony of James Dannemiller, an expert on statistics, who testified that the examination of a random sample of 137 claims would achieve “a 95 percent statistical probability that the same percentage determined to be valid among the examined claims would be applicable to the totality of claims filed.”....

The district court then appointed Sol Schreiber as a special master (and a court-appointed expert under Rule 706 of the Federal Rules of Evidence). Schreiber supervised the taking of depositions ... of the 137 randomly selected claimants....

Id. at 782 (subheading omitted). “Schreiber then reviewed the claim[s],” and “[h]e recommended that 6 claims of the 137 in the sample be found not valid.” Id. at 782-83. “Schreiber then recommended the amount of damages to be awarded to the 131 [remaining] claimants.” Id. at 783.

Based on his recommendation that 6 of the 137 claims in the random sample (4.37%) be rejected as invalid, he recommended the application of a five-per-cent invalidity rate to the remaining claims.

He recommended that the award to the class be determined by multiplying the number of valid remaining claims ... by the average award recommended for the ... claims....

By adding the recommended awards ... Schreiber arrived at a recommendation for a total compensatory damage award....

A jury trial on compensatory damages was [then] held ... [Hilao’s statistical expert] testified that the selection of the random sample met the standards of inferential statistics, that the successful efforts to locate and obtain testimony from the claimants in the random sample “were of the highest standards” in his profession, that the procedures followed conformed to the standards of inferential statistics, and that the injuries of the random-sample claimants were representative of the class as a whole. Testimony from the 137 random-sample claimants and their witnesses was introduced. Schreiber testified as to his recommendations, and his report was supplied to the jury. The jury was instructed that it could accept, modify or reject Sehreiber’s recommendations and that it could independently, on the basis of the evidence of the random-sample claimants, reach its own judgment as to the actual damages of those claimants and of the aggregate damages suffered by the class as a whole.

The jury deliberated for five days before reaching a verdict. Contrary to the master’s recommendations, the jury found against only two of the 137 claimants in the random sample. As to the sample claims, the jury generally adopted the master’s recommendations, although it did not follow his recommendations in 46 instances. As to the claims of the remaining class members, the jury adopted the awards recommended by the master. The district court subsequently entered judgment for 135 of the 137 claimants in the sample in the amounts awarded by the jury, and for the remaining plaintiffs ... in the amounts awarded by the jury, to be divided pro rata.

Id. at 783-84 (subheadings and footnotes omitted).

On appeal, the Hilao court was presented with some of the same objections to its trial plan as Wal-Mart presents here.⁵⁵

After a lengthy discussion, however, the Hilao court rejected these challenges and approved of the trial plan, addressing the due process issue as follows:

While the district court’s methodology in determining valid claims is unorthodox, it can be justified by the extraordinarily unusual nature of this case. “ ‘Due process,’ unlike some legal rules, is not a technical conception with a fixed content unrelated to time, place and circumstances.” Cafeteria and Restaurant Workers Union, Local 473 v. McElroy, 367 U.S. 886, 895, 81 S.Ct. 1743, 6 L.Ed.2d 1230 ... (1961).

The interest of the [defendant] that is affected is at best an interest in not paying damages for any invalid claims.... The statistical method used by the district court obviously presents a somewhat greater risk of error in comparison to an adversarial adjudication of each claim, since the former method requires a probabilistic prediction (albeit an extremely accurate one) of how many of the total claims are invalid.... Hilao’s interest in the use of the statistical method, on the other hand, is enormous, since adversarial resolution of each class member’s claim would pose insurmountable practical hurdles. The “ancillary” interest of the judiciary in the procedure is obviously also substantial, since 9,541 individual adversarial determinations of claim validity would clog the docket of the district court for years. Under the balancing test set forth in Mathews [v. Eldridge, 424 U.S. 319, 96 S.Ct. 893, 47 L.Ed.2d 18 (1976),] and [Connecticut v.] Doehr [, 501 U.S. 1, 111 S.Ct. 2105, 115 L.Ed.2d 1 (1991) ], the procedure used by the district court did not violate due process.

Hilao, 103 F.3d at 786-87 (footnote and citations omitted).

Because we see no reason why a similar procedure to that used in Hilao could not be employed in this case,⁵⁶ we conclude that there exists at least one method of managing this large class action that, albeit somewhat imperfect, nonetheless protects the due process rights of all involved parties.⁵⁷ Accordingly, we find no manageability-based reason to find this otherwise-certifiable class unsuited to class certification.

CONCLUSION

For the reasons set forth above, we hold that the district court acted within its broad discretion in concluding that it would be better to handle some parts of this case as a class action instead of clogging the federal courts with innumerable individual suits litigating the same issues repeatedly. The district court did not abuse its discretion in finding the requirements of Rule 23 satisfied with respect to those Plaintiffs who were still Wal-Mart employees on June 8, 2001, and with respect to claims for injunctive and declaratory relief and back pay. Given the tentative nature of the district court’s trial plan, we decline to address Wal-Mart’s due process and manageability challenges to that plan. We note, however, that the district court has the discretion to modify or decertify the class should it become unmanageable. Although the size of this class action is large, mere size does not render a case unmanageable.

We deny Plaintiffs’ cross-appeal, because the district court did not abuse its discretion when it found that back pay for promotions may be limited to those Plaintiffs for whom proof of qualification and interest exists. Finally, we must reiterate that our findings relate only to class action procedural questions; we neither analyze nor reach the merits of Plaintiffs’ allegations of gender discrimination.

AFFIRMED in part; REMANDED in part. Each party to bear its own costs on appeal.

Notes

1 . We have jurisdiction under 28 U.S.C. § 1292(e).

2 . The action was originally filed on June, 8, 2001.

3 . Ten times the dissent points out the large class size, referring to the "1.5 million” women alleging discrimination as a reason to reject certification. However, at oral argument, counsel for Wal-Mart conceded that if past employees are excluded the class to be certified would be less than 1.5 million, perhaps two-thirds less than the figure the dissent cites. Further, the dissent emphasizes that this is the largest class that we have ever certified "based on a small number of incidents.” Dissent at 635 n. 11 (all references in this Opinion to the dissent refer to Judge Ikuta’s dissent). As an initial matter, we do not find 120 claims of illegal sex discrimination a small number. Nevertheless, given that the class is suing by far the largest employer in the United States, we are unsurprised that Plaintiffs are seeking to represent such a large class. Of course, this should not change the analysis the Supreme Court required in General Telephone Co. of the Southwest v. Falcon, 457 U.S. 147, 157-59, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982), nor should it necessitate that we take an even harder look at the merits, essentially conducting a trial, as the dissent proposes.

4 . As the district court acknowledged, Dukes, 222 F.R.D. at 143, although federal courts are no longer permitted to engage in “conditional certification,” see Fed.R.Civ.P. 23 advisory committee’s note to 2003 amends., district courts retain the authority to amend or decertify a class if, based on information not available or circumstances not anticipated when the class was certified, the court finds that either is warranted.

5 . After a trial, the district court in Falcon made converse findings regarding Falcon, who was the named plaintiff, and the class he sought to represent. The district court found that the defendant, General Telephone Company of the Southwest, had not discriminated against Falcon when it hired him, but had discriminated against him when it failed to promote him. Falcon, 457 U.S. at 152, 102 S.Ct. 2364. But the district court also found the defendant had not discriminated against Mexican-Americans generally in pro motion decisions, but had discriminated against them in hiring. Id. This disconnect created a potential typicality problem under Rule 23(a)(3).

6 . We recognize that these words imply slightly different approaches, and we agree with the IPO court's thoughtful discussion of resisting use of the word " ‘findings' because the word usually implies that a district judge is resolving a disputed issue of fact," and the Rule 23 inquiry, though it may require findings in some cases, "is really a mixed question of fact and law.” 471 F.3d at 40. We cannot, however, agree that the difference in usage is significant enough that circuit courts using these various words employ substantially different standards. Cf. New Motor Vehicles, 522 F.3d at 24.

7 . Many other decisions have similarly shown this circuit’s proper, narrow reading of the "no merits inquiry” passage from Eisen to prohibit only merits inquiries unnecessary for the class certification decision. See, e.g., McElmurry v. U.S. Bank Nat'l Ass'n, 495 F.3d 1136, 1141-42(9th Cir.2007) (rejecting the appellants’ broad claim relying on misunderstood language from Eisen); Staton, 327 F.3d at 954(citing Eisen in explaining that a merits inquiry is improper at the class certification stage though the district court was "well within” its discretion in considering overlapping issues); Valentino v. Carter-Wallace, Inc., 97 F.3d 1227, 1232 (9th Cir.1996) (discussing Eisen as prohibiting "a separate hearing to consider the merits of the plaintiffs’ claims when determining class certification”); Hanon, 976 F.2d at 509(describing Eisen as holding "a motion for class certification is not the appropriate point at which to resolve the merits of a plaintiff's claim”); Wright v. Schock, 742 F.2d 541, 544-45 (9th Cir.1984) (understanding Eisen narrowly to allow a summary judgment motion before the class certification decision); Moore, 708 F.2d at 480 ("Although some inquiry into the substance of a case may be necessary to ascertain satisfaction of the commonality and typicality requirements of Rule 23(a), it is improper to advance a decision on the merits to the class certification stage.”); Jordan v. County of Los Angeles, 669 F.2d 1311, 1321-22 (9th Cir.), (characterizing Eisen as confirming the narrow rule that "the class representative need not demonstrate a likelihood of success on the merits in order to maintain a class action”), vacated on other grounds, 459 U.S. 810, 103 S.Ct. 35, 74 L.Ed.2d 48 (1982).

8 . That district courts must consider questions overlapping with the merits, if doing so is necessary to find the Rule 23 requirements met, is the understanding of the majority of district court orders considering the issue in this circuit. See Colvin v. Citigroup Global Mitts., Inc., No. C 09-00238, 2009 WL 1657331, at *1-2 (N.D.Cal. June 11, 2009) (hearing issues overlapping with merits before class certification motion); In re First Am. Coip. ERISA Litig., 258 F.R.D. 610, 616 (C.D.Cal.2009) (citing Eisen but also noting Falcon's statement that class determinations are often "enmeshed” with merits issues); In re Cooper Cos. Sec. Litig., 254 F.R.D. 628, 641 n. 7 (C.D.Cal.2009) (noting the obligation to consider evidence relating to the Rule 23 determination, but stating that "[a]t the same time, the inclusion of evidence that may speak to the merits of a case does not mean that a district court should mistake class action certification for summary judgment”); Ballew v. Matrixx Initiatives, Inc., No. CV-07-267, 2008 WL 4831481, at *2 (E.D.Wash. Oct. 31, 2008) ("It is true that at this stage of the proceedings the Court is not resolving whether Plaintiff can establish that she can prove her case. Rather, the narrow question before the Court is whether, under Fed.R.Civ.P. 23, a class action is a proper vehicle for litigating the products liability claims brought by Plaintiff.”); Bishop v. Petro-Chem. Transp., LLC, 582 F.Supp.2d 1290, 1305 (E.D.CaI.2008) ("[T]he Court has evaluated the evidence of the Motor Carrier exemption, not to determine whether plaintiff will ultimately prevail at trial, but to determine, again, whether the proposed class members are similarly situated.”); Lindquist v. Fanners Ins. Co. of Ariz., No. CV 06-597, 2008 WL 343299, at *6 (D.Ariz. Feb. 6, 2008) ("While preliminary inquiry into the merits is sometimes appropriate to the extent necessary to evaluate class certification issues that happen to be intertwined with facts relating to the underlying merits ... the general rule followed by federal courts [is] that trial courts may not decide the merits of the case or consider the likelihood of success on the merits in determining whether the requirements of Rule 23 class certification are met.”); Alexander v. JBC Legal Group, P.C., 237 F.R.D. 628, 630 (D.Mont. 2006) ("It is improper for this Court to consider the merits of the complaint beyond what is necessary to decide whether the requirements of Rule 23 have been met.”); In re Seagate Tech. II Sec. Litig., 843 F.Supp. 1341, 1367 (N.D.Cal.1994) (noting that "district courts must examine some of the merits” because "the adequacy of the representation cannot be ascertained without examining the composition of the class”); Hernandez v. Alexander, 152 F.R.D. 192, 194 (D.Nev.1993) ("The determination of whether to certify a class does not permit or require a preliminary inquiry into the merits. Only the class certification requirements of Rule 23 need be considered at the class certification stage. However, before certifying a class action the trial court must be satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been met.” (citations to Eisen and Falcon omitted)); Lim v. Citizens Sav. & Loan Ass’n, 430 F.Supp. 802, 805-06 (N.D.Cal.1976) ("Although the Eisen rule has evolved into a wedge for broad class certification by plaintiffs, a close reading of the Eisen case indicates the Supreme Court's contrary intention. ...”); El Concilio v. Modesto Elementary Sch. Dist., No. C-76-2479, 1978 WL 56, at *1 (N.D.Cal. May 11, 1978) (“While the Court is foreclosed from a preliminary inquiry into the merits in order to determine whether it may be maintained as a class action, it must determine whether plaintiffs have met their burden of establishing the existence of a claim of unlawful employment practices affecting an ascertainable class of persons of which they are members.” (internal quotation marks and citation omitted)); Nguyen Da Yen v. Kissinger, 70 F.R.D. 656, 661 (N.D.Cal.1976) ("In determining whether a matter should proceed as a class action the court is required to make findings concerning each essential element of the class action rule.”).

9 . District courts in this circuit have quoted this passage somewhat out of context on a number of occasions. See Mateo v. V.F. Corp., No. C 08-05313, 2009 WL 3561539, at *4 (N.D.Cal. Oct. 27, 2009) (citing Blackie though noting that the defense’s arguments addressed only the merits and not the Rule 23 inquiry); Perez v. Safety-Kleen Sys., Inc., 253 F.R.D. 508, 511 (N.D.Cal.2008) (noting that "the court must accept the substantive allegations contained in plaintiffs’ complaints as true”); Nw. Fruit Co. v. A. Levy & J. Zentner Co., 116 F.R.D. 384, 388 (E.D.Cal.1986) (same); see also Steig D. Olson, "Chipping Away”: The Misguided Trend Toward Resolving Merits Disputes as Part of the Class Certification Calculus, 43 U.S.F. L.Rev. 935, 949 n. 89 (2009) (understanding Blackie as requiring acceptance of a plaintiff’s claims as to not only merits questions, but Rule 23 requirements as well).

10 . IPO's "release valve” reads as follows:

[W]e reach the following conclusions: ... (5) a district judge has ample discretion to circumscribe both the extent of discovery concerning Rule 23 requirements and the extent of a hearing to determine whether such requirements are met in order to assure that a class certification motion does not become a pretext for a partial trial of the merits.

471 F.3d at 41.

11 . The plaintiffs in Hnot pled both disparate impact and disparate treatment theories of discrimination under Title VII. See Hnot v. Willis Group Holdings Ltd., 228 F.R.D. 476, 486 (S.D.N.Y.2005).

12 . More specifically, as the cases demonstrate, is the standard for evidence under Rule 23(a)(2), commonality, versus Rule 23(b)(3), predominance.

13 . Even commentators who clearly recognize this difference often lapse into speaking about Rule 23 as requiring a single showing, minimizing the differences between Rule 23(a)(2) and Rule 23(b)(3). See, e.g., Nagareda, Class Actions, supra, at 622 ("To task the court at the class certification stage here with the malting of a 'definitive assessment' of compliance with Rule 23 would be to call, as a practical matter, for an assessment of which side is right on the merits with regard to the existence of a company-wide policy.”).

14 . See also Olson, supra, at 947(noting that class certification must be particularly predictive under Rule 23(b)(3)).

15 . The dissent chides us for ignoring the “ 'significant proof requirement” from Falcon, an approach the dissent finds troubling, particularly because Falcon is "directly on point.” Dissent at 632-33. Putting aside the fact that, unlike here, the problem in Falcon was a proposed class consisting of members of the general public who had sought employment, represented by an employee who had sought promotion, alleging different theories of discrimination, see Falcon, 457 U.S. at 157-58, 102 S.Ct. 2364, ignoring relevant Supreme Court guidance would indeed be troubling if that was in fact what we were doing. But the dissent mischaracterizes the role that "significant proof” played in the Falcon analysis. The statement was a hypothetical in clear dicta. Id. at 159 n. 15, 102 S.Ct. 2364. It is important not to apply Supreme Court dicta “to create a new rule — one [the Court] never envisioned.” Lopez-Rodriguez v. Holdr, 560 F.3d 1098, 1099 (9th Cir.2009) (order) (Bea, J., dissenting from denial of reh’g en banc). Further, here, we are not taking Supreme Court statements and "blandly shrug[ging] them off because they were not a holding.” Newdow v. U.S. Congress, 328 F.3d 466, 480 (9th Cir.2003) (order) (O'Scannlain, J., dissenting from denial of reh'g en banc). Instead, we are considering the statements and avoiding creating a new rule of law, noting that we should be particularly wary of conceiving such a rule from dicta when the facts underlying the dicta are easily distin- ' guished from the present case and when the dicta relied upon was in the form of a hypothetical.

16 . The dissent reads Falcon to require a showing of evidence "sufficient to carry plaintiffs’ burden of adducing significant proof." Dissent at 641. However, the dissent’s encapsulating all of Falcon into this one phrase in dicta ignores other key distinctions on which the Court was actually relying. Falcon does not hold that a plaintiff must prove a widespread policy of discrimination to obtain class certification nor, of course, does it allow a district court to certify a class based on a plaintiff’s mere allegations of discriminatory practices. 457 U.S. at 157-58, 102 S.Ct. 2364. Rather, Falcon only says that an individual who, on his own behalf, does not allege any overall policy cannot, by proving his own case, establish on the merits an inference relevant to the group, thus remedying his failure to otherwise make a showing of commonality. Id. Such a statement is a far cry from suggesting, as does the dissent, see Dissent at 633-34, 634-36, that a class cannot be certified unless the plaintiffs prove a discriminatory overall policy — in which case there would be no reason to have a trial at all.

17 . The dissent’s claim that Plaintiffs must prove discrimination at this stage, Dissent at 633-34, conflates the class certification and merits phases of the litigation. See Brown, 576 F.3d at 153 ("[Ajllegations of a practice of disparate treatment in the exercise of unbridled discretion raises questions of law and fact common to all subject black employees.” (emphasis added) (internal quotation marks omitted)); Shipes v. Trinity Indus., 987 F.2d 311, 316 (5th Cir.1993) ("The threshold requirements of commonality and typicality are not high.... Allegations of similar discriminatory employment practices, such as the use of entirely subjective personnel processes that operate to discriminate, satisfy the commonality and typicality requirements of Rule 23(a).” (emphasis added)).

18 . Indeed, the record provides significant evidence of central control. The district court found that, "[hjaving reviewed the extensive evidence submitted by the parties, ... WalMart's systems for compensating and promoting in-store employees are sufficiently similar across regions and stores to support a finding that the manner in which these systems affect the class raises issues that are common to all class members.” Dukes, 222 F.R.D. at 149. Yet the dissent seems to assume that central control was not a pervasive feature of WalMart’s baseline promotion policies, or that if it was, such control undercuts Plaintiffs’ case. Dissent at 6253 (rejecting Plaintiffs’ claim that " 'the subjective decision-making in compensation and promotions takes place within parameters and guidelines that are highly uniform, and within a strong corporate culture,' ” because "this argument is contrary to the thrust of plaintiffs’ legal theory”) (citing Dukes, 222 F.R.D. at 157). Such a view ignores the precise contours of Plaintiffs’ claims. Questioning the degree of central control over the framework policies is particularly surprising given the district court's findings, see Dukes, 222 F.R.D. at 145-54, and additional evidence in the record that, for example, Wal-Mart posted employees in the "Rising Star Program” who were eligible for promotion on a wall in the Bentonville Home Office.

19 . These common issues of law and fact, necessary to bridge the gap from the class representatives to the entire class, are discussed in Part III.A. For now, it is sufficient to note that they include an analysis of similar promotion and compensation policies Wal-Mart applied to all employees, a dominant corporate culture Wal-Mart maintained throughout its stores, and statistical evidence showing class-wide gender disparities in compensation. Dukes, 222 F.R.D. at 145-69.

20 . Of course, returning to the standard discussed above, if the district court had rejected Wal-Mart’s arguments regarding commonality solely because they overlapped with "merits issues,” that would have been error. However, as we explain in the following sections, the district court did not do this but, instead, conducted a "rigorous analysis” of the conflicting evidence presented on the commonality question and ultimately concluded that Plaintiffs raised "questions of law or fact common to the class.” Fed.R.Civ.P. 23(a)(2); see Falcon, 457 U.S. at 161, 102 S.Ct. 2364; Hanon, 976 F.2d at 509.

21 . For a description of the "social framework analysis,” see Melissa Hart & Paul M. Secunda, A Matter of Context: Social Framework Evidence in Employment Discrimination Class Actions, 78 Fordham L.Rev. 37, 41-55 (2009).

22 . We are not convinced by the dissent’s argument that Daubert has exactly the same application at the class certification stage as it does to expert testimony relevant at trial. Dissent at 6255. However, even assuming it did, the district court here was not in error. Thus we need not resolve this issue here.

In accepting Dr. Bielby’s social framework analysis, the district court stated:

The Court is further guided by Daubert v. Metrell Dow Pharm., Inc. (Daubert II), 43 F.3d 1311, 1316 (9th Cir.1995), in which the Ninth Circuit stated that scientific knowledge ’’does not mean absolute certainty,” and that expert testimony should be admitted when "the proffered testimony is 'based on scientifically valid principles.' ” Id., quoting Daubert I, 509 U.S. 579, 113 S.Ct. 2786, 125 L.Ed.2d 469. The Ninth Circuit continued: "Our task, then, is to analyze not what the experts say, but what basis they have for saying it.” Daubert II, 43 F.3d at 1316. The Court is satisfied that Dr. Bi[e]lby’s opinion — while subject to critique — is based on valid principles. Thus, it is sufficiently probative to assist the Court in evaluating the class certification requirements at issue in this case. Accordingly, Defendant’s motion to strike Dr. Bilby’s declaration is denied.

Dukes v. Wal-Mart, Inc., 222 F.R.D. 189, 192 (N.D.Cal.2004) ("Dukes II”).

On this review, regardless of whether full Daubert review is required at the class certification stage, we cannot say that admitting Dr. Bielby’s social framework analysis was an abuse of discretion. The district court specifically considered Daubert’s applicability to Dr. Bielby and was within its discretion to limit its inquiry. As the district court observed, Dr. Bielby's testimony could be admissible even without reaching “definitive[]” conclusions, because tentative rather than “conclusive determination[s]” are in “the nature of this particular field of science.” Dukes, 222 F.R.D. at 154.

Other cases support this conclusion. The Third Circuit, citing IPO, has noted that when the plaintiffs’ ability to prove their case "is genuinely disputed, the district court must resolve it after considering all relevant evidence. Here [in Hydrogen Peroxide ], the District Court apparently believed it was barred from resolving” these expert disputes. Hydrogen Peroxide, 552 F.3d at 325. Unlike Hydrogen Peroxide and other cases Wal-Mart cites, in the present case the district court heard, rather than excluded, the bulk of the relevant evidence. As a general rule, " '[district courts are not required to hold a Daubert hearing before ruling on the admissibility of scientific evidence,’ ” Millenkamp v. Davisco Foods Int’l, Inc., 562 F.3d 971, 979 (9th Cir. 2009) (quoting Jaros v. E.I. DuPont (In re Hanford Nuclear Reservation Litig.), 292 F.3d 1124, 1138 (9th Cir.2002)). Plaintiffs clearly established foundation for Dr. Bielby’s testimony and statistics. It was not an abuse of discretion for the district court not to exclude them, under Daubert or otherwise.

23 . As discussed below in Part II.A.2.b, this court and many others have held that “delegation to supervisors, pursuant to company-wide policies, of discretionary authority without sufficient oversight ... gives rise to common questions of fact warranting certification of the proposed class.” Caridad, 191 F.3d at 291, overruled on other grounds by IPO, 471 F.3d at 39-42; see Hnot, 241 F.R.D. at 210(explaining that IPO did not overrule this aspect of Caridad); see also Staton, 327 F.3d at 955-56 (rejecting argument that "decision-making at Boeing is too decentralized to permit a class that combines plaintiffs from disparate locales”); Shipes v. Trinity Indus., 987 F.2d 311, 316 (5th Cir.1993) (upholding commonality finding where all of company's plants “utilized the same subjective criteria in making personnel decisions”); Cox v. Am. Cast Iron Pipe Co., 784 F.2d 1546, 1557 (11th Cir. 1986) (holding that " '[a]llegations of similar discriminatory employment practices, such as ... [the] use of entirely subjective personnel processes that operated to discriminate, would satisfy the commonality and typicality requirements of Rule 23(a)' ” (alterations in original) (quoting Carpenter v. Stephen F. Austin State Univ., 706 F.2d 608, 617 (5th Cir. 1983))); Segar v. Smith, 738 F.2d 1249, 1276 (D.C.Cir.1984) (explaining that "subjective criteria may well serve as a veil of seeming legitimacy behind which illegal discrimination is operating”).

24 . See Hnot, 241 F.R.D. at 210 ("Caridad held, in part, that class challenges to subjective employment practices, like other disparate treatment and disparate impact claims, may satisfy the commonality requirement. In re IPO did not even address this aspect of Caridad. ... [I]t did not question Caridad's holding that statistical evidence can demonstrate commonality in a challenge to subjective employment practices, which remains controlling authority.” (citation omitted)).

25 . Each region contains approximately 80 to 85 stores.

26 . Regression analyses, in general terms, provide estimates of the effect of independent variables on a single dependent variable. See Hemmings v. Tidyman's Inc., 285 F.3d 1174, 1183-84 & n. 9 (9th Cir.2002). The purpose of this methodology is to estimate the extent to which a particular independent variable (in this case, gender) has influenced the dependent variables of compensation and promotion. See id.; see also Rudebusch v. Hughes, 313 F.3d 506, 511-12 (9th Cir.2002). As long as the analyses include enough relevant non-discriminatory independent variables (e.g., education, experience, performance, etc.), the results will indicate whether any salary disparities are attributable to gender (thereby raising an inference of discrimination) or whether the disparities are attributable to other factors (and thereby refuting such an inference). See Hemmings, 285 F.3d at 1183-84 & n. 9; see also EEOC v. Gen. Tel. Co. of Nw., Inc., 885 F.2d 575, 577 n. 3 (9th Cir.1989) ("A regression analysis is a common statistical tool .'.. designed to isolate the influence of one particular factor — [e.g.,] sex — on a dependent variable — [e.g.,] salary.” (internal quotation marks omitted)).

27 . “Specifically, Dr. Bendick compared, or 'benchmarked,’ Wal-Mart against twenty other [similar] general merchandise retailers by comparing workforce data provided by the companies to the Equal Employment Opportunity Commission.” Dukes, 222 F.R.D. at 164. Dr. Bendick analyzed the data "to determine the extent to which women in the relevant market sought promotion, so that an inference could be made that roughly the same percentage of women would have [sought promotion] at Wal-Mart if given the opportunity.” See id. As Dr. Bendick explained, "The logic in benchmarking is that, if retail chains comparable to Wal-Mart are successfully employing women at some rate, then women are presumably available, interested, and qualified to hold comparable positions at Wal-Mart at a similar rate.” Id.

28 . This argument is unsurprising, and is the statistical argument similarly situated defendants make as a matter of course.

29 . The district court’s other two citations to Caridad supported its statements that, first, ''[djefendant's arguments seek to engage the Court in a merits evaluation of the expert opinions. The Court rejects this approach, and views the statistical evidence and testimony through the proper lens of the standards applicable to a class certification motion.” Dukes, 222 F.R.D. at 155 & n. 21 (citing Caridad, 191 F.3d at 292). Second, "[tjhe ultimate question of whether subjective decision-making and a uniform culture contribute to a nation-wide pattern of gender discrimination will, of course, be for a jury to decide. At this stage, however, these factors are apparent enough to support Dr. Drogin’s regional approach as at least a reasonable means of conducting a statistical analysis.” Id. at 159 & n. 29 (citing Caridad, 191 F.3d at 292-93). These are determinations of the type we require for Rule 23(a)(2) certification. Though we have used different language in clarifying the standard today, the district court examined the evidence and made a determination, over Wal-Mart’s evidence, that Plaintiffs’ evidence raised questions of law or fact common to the class. That is what Rule 23(a)(2), the Supreme Court, and we require.

30 . Contrary to the dissent's characterization, the district court was cognizant of the possibility of aggregation problems in the data. Dukes, 222 F.R.D. at 155-59 & n. 22. The dissent's invocation of "Simpson's Paradox” to discount Plaintiffs’ statistical evidence is not compelling, particularly at the class certification stage. See Dissent at 637 n.12. First, Simpson’s Paradox is a problem of incorrect causation inferences as much as it is the result of an aggregation problem, see Judea Pearl, Causality: Models, Reasoning, and Inference 130 (2000) (noting that the study the dissent cites, while not "defective,” highlights "that no adjustment is guaranteed to give an unbiased estimate of causal effects, direct or indirect, absent a careful examination of the causal assumptions that ensure identification”), and the district court considered potential problems and causation inferences in its discussion of the competing regression analyses. See Dukes, 222 F.R.D. at 155-56 & n. 22. Second, at trial, Wal-Mart is free to further argue the unpersuasiveness of Plaintiffs' statistics, but lacking evidence showing a likelihood of a Simpson’s Paradox, and given the many causal factors the statistical analyses addressed, see id.; Marios G. Pavlides & Michael D. Perlman, How Likely is a Simpson’s Paradox, 63 Am. Statistician 226, 229-30 (2009), we find the district court properly considered Plaintiffs' statistics probative after a rigorous analysis. See Falcon, 457 U.S. at 161, 102 S.Ct. 2364; Dukes, 222 F.R.D. at 156-59.

31 . This means that Dr. Haworth ran separate regression analyses for: (1) each of the specialty departments in the store, (2) each grocery department in the store, and (3) the store’s remaining departments. She did not run regression analyses to examine pay differential between male and female salaried employees.

32 . For example, Wal-Mart maintains that the district court erred by not requiring Dr. Drogin to perform a "Chow test” to determine whether data could be properly aggregated. We have not found a single case suggesting or requiring use of such a test.

33 . In addition to her sub-store analysis, Dr. Haworth conducted a survey of store managers. After reviewing the survey and its methodology, the district court concluded that the Store Manager survey was biased both "on its face” and in the way that it was conducted. Dukes II, 222 F.R.D. at 196-97(noting that the survey's results "are not the ‘product of reliable principles and methods,' and therefore are not the type of evidence that would be 'reasonably relied upon by experts' ” (quoting Fed.R.Evid. 702, 703)). Dr. Haworth’s disaggregated analysis created pools too small to yield any meaningful results. Wal-Mart has not appealed this issue. Accordingly, this evidence is not properly before us. See Kohler v. Inter-Tel Techs., 244 F.3d 1167, 1179 n. 8 (9th Cir.2001) (recognizing that the appellant waived a claim by failing to raise it in her briefs).

34 . Plaintiffs submitted declarations from each of the class representatives, as well as 114 declarations from putative class members around the country. See Dukes, 222 F.R.D. at 165.

35 . Of course, the district court made no finding that Plaintiffs' anecdotal declarations alone would raise an inference of common discriminatory experiences. Therefore, contrary to the dissent's mischaracterization of our holding, Dissent at 6244-45, this case does not present the opportunity for us to consider, let alone affirm, such a hypothetical finding.

36 . The dissent’s accusation that both we and the district court are playing the "proverbial shell game,” is unpersuasive given the weight of the evidence the district court considered. See Dissent at 640-41. Not surprisingly, the dissent’s implication that a single form of evidence must, by itself, be able to support a commonality finding to satisfy the standard of significance set forth in Falcon, cites no authority for its proposition. Id. Even under a preponderance of the evidence standard, see, e.g., Teamsters Local 445 Freight Div. Pension Fund v. Bombardier Inc., 546 F.3d 196, 202 (2d Cir.2008), the district court's review here would have been sufficient, finding Plaintiffs' "evidence more than satisfies plaintiffs’ burden to demonstrate commonality under Rule 23(a)(2).” See Dukes, 222 F.R.D. at 143, 145 ("The party seeking certification must provide facts sufficient to satisfy Rule 23(a) and (b) requirements. In turn, the district court must conduct a rigorous analysis to determine that the prerequisites of Rule 23 have been met. If a court is not fully satisfied, certification should be refused.” (citations omitted)).

37 . Although the “commonality and typicality requirements of Rule 23(a) tend to merge,” see Falcon, 457 U.S. at 157 n. 13, 102 S.Ct. 2364, each factor serves a discrete purpose. Commonality examines the relationship of facts and legal issues common to class members, while typicality focuses on the relationship of facts and issues between the class and its representatives. See Newberg on Class Actions § 3:13, at 317.

38 . The purported class need only satisfy one of Rule 23(b)’s prongs to be sustainable. See Zinser, 253 F.3d at 1186.

39 . See, e.g., Thorn v. Jefferson-Pilot Life Ins. Co., 445 F.3d 311, 331 (4th Cir.2006) (awards of back pay do not predominate over the injunctive remedies because their calculation generally involves relatively uncomplicated factual determinations); In re Monumental Life Ins. Co., 365 F.3d 408, 418 (5th Cir.2004) (same); Cooper v. Southern Co., 390 F.3d 695, 720 (11th Cir.2004), overruled on other grounds by Ash v. Tyson Foods, Inc., 546 U.S. 454, 457-58, 126 S.Ct. 1195, 163 L.Ed.2d 1053 (2006) (noting that back pay can be awarded in a case certified under Rule 23(b)(2)); see also Reeb, 435 F.3d at 650 (same); Coleman v. GMAC, 296 F.3d 443, 449 (6th Cir.2002) (same); Robinson, 267 F.3d at 164(holding compensatory or punitive damages sometimes permissible under Rule 23(b)(2)); Thomas v. Albright, 139 F.3d 227 (D.C.Cir.1998) (upholding a settlement in which claims for compensatory damages were included in a class certified under Rule 23(b)(2)).

40 . Although some circuits have held that a request for back pay technically weighs on the monetary side of the scale, even though it is also an equitable form of relief, none has held that this prevents requests for back pay from being certified under Rule 23(b)(2). See, e.g., Thom, 445 F.3d at 332 ("Rule 23(b)(2) class certification is proper in the Title VII context not because backpay is an equitable form of relief, but because injunctive or declaratory relief predominates despite the presence of a request for back pay.”); Eubanks v. Billington, 110 F.3d 87, 95 (D.C.Cir.1997); see also In re Monumental Life Ins. Co., 365 F.3d at 418("[E]quitable monetary remedies are less likely to predominate over a class’s claim for injunctive relief, but this has more to do with the uniform character of the relief rather than with its label.”).

We need not decide whether to adopt the view that a request for equitable relief such as back pay weighs against certification because here, even assuming without deciding that it does, Plaintiffs' request for back pay does not predominate over their request for the injunctive and declaratory relief and therefore does not prevent certification under Rule 23(b)(2).

41 . The dissent claims that "the majority misses the point” in citing Coleman for the Sixth Circuit's statement “that back pay is a permissible remedy in a Rule 23(b)(2) class,” 296 F.3d at 449, because “the issue is not that plaintiffs seek back pay,” but whether "individualized treatment makes class-wide relief improper.” Dissent at 650 n. 28. While it is true that the Sixth Circuit rejected certification of the class in Coleman, it did so by distinguishing the plaintiff's claim for back pay and compensatory damages from previous cases that upheld certification of a class under Rule 23(b)(2) that sought back pay but not compensatory damages. Id. at 449 (“[C]alculation of back pay generally involves less complicated factual determinations and fewer individualized issues.”). Further, the Coleman court took issue with the plaintiff's own calculations of compensatory damages, which she sought to calculate "by reference to the markup [on a car loan] charged to each individual member of the class.” Id. Given that Plaintiffs here are not seeking compensatory damages, and that, unlike here, there was no way in Coleman to calculate damages other than an individualized framework that, in fact, the plaintiff proposed, we find Coleman instructive for its statement "that back pay is a permissible remedy in a Rule 23(b)(2) class.” Id. (citing Alexander v. Aero Lodge No. 735, 565 F.2d 1364, 1372 (6th Cir.1977) (“A request for back pay does not preclude certification under [23(b)(2) ].”))

The dissent's claim of error is equally misguided to Thorn where the concern with individualized inquiries related to the subjective state of mind requirement for the Fourth Circuit's accrual rule for a statute of limitations. See Thorn, 445 F.3d at 317, 320. The Fourth Circuit addressed an earlier ruling in apparent conflict by explaining the previous case had dealt with constructive notice and was thus amenable to class resolution, but under the subjective, actual notice claim presented in Thom there was no way around individualized inquiries. Id. at 325. Here, back pay and other claims can be calculated on a class-wide basis, and because this is a pattern and practice case, discrimination must be shown at a group level. See Domingo v. New England Fish Co., 727 F.2d 1429, 1444 (9th Cir. 1984) (per curiam); infra notes 49 and 53. Further, the Thom court, in discussing Title VII cases, stated that “in the Title VII context, awards of back pay do not predominate over the injunctive remedies available because the calculation of back pay generally involves relatively uncomplicated factual determinations and few individualized issues.... Rule 23(b)(2) class certification is proper in the Title VII context not because back pay is an equitable form of relief, but because injunctive or declaratory relief predominates despite the presence of a request for back pay.” Thom, 445 F.3d at 331-32 (brackets, citation, and internal quotation marks omitted). Again, given these statements and distinguishing contexts, we find Thom supportive of our conclusion that a request for back pay in a Title VII case is fully consistent with the certification of a Rule 23(b)(2) class action.

42 . Neither Williams v. Owens-Illinois, Inc., 665 F.2d 918, 928-29 (9th Cir.1982), nor Zinser, 253 F.3d at 1195-96, supports Wal-Mart’s argument that this circuit will not certify a class action that involves punitive damages. Rather, this court merely held that it was not an abuse of discretion to deny class certification based on the specific facts presented in those cases. See Williams, 665 F.2d at 929(holding that damages requests were not incidental to the request for injunctive relief where requested compensatory damages were not clearly compatible with class injunctive relief); Zinser, 253 F.3d at 1195-96 (same, holding that a request for medical monitoring against manufacturer of pacemaker is not certifiable under Rule 23(b)(2) in the specific circumstances of the case, where the class primarily sought establishment of a reserve fund for past and future damages, compensation for future medical treatment, and other compensatory and punitive damages).

43 . Relying on Rule 23(c)(4), our own precedent also generally allows class treatment of common issues even when not all issues may be treated on a class basis. See Valentino v. Carter-Wallace, Inc., 97 F.3d 1227, 1234 (9th Cir.1996); see also Augustin v. Jablonsky (In re Nassau County Strip Search Cases), 461 F.3d 219, 226 (2d Cir.2006) (same).

44 . The dissent argues that Ticor Title Insurance Co. v. Brown, 511 U.S. 117, 114 S.Ct. 1359, 128 L.Ed.2d 33 (1994) (per curiam), precludes district courts from granting notice and opt-out rights in class actions certified under Rule 23(b)(2). Dissent at 6270-71. Ti-cor involved the res judicata effect of a prior case certified under Rule 23(b)(2) without an opportunity for opt-out, so the Court's attention, quite understandably, was not focused on the extent of the district court’s discretion to provide for such measures. Ticor s reference to typical Rule 23(b)(2) procedures also was contained in an order dismissing a grant of certiorari and therefore was not even dictum, let alone a holding. See id.

45 . The dissent largely ignores the limits herein imposed on the district court’s certification order, particularly regarding the mode of proof described below. Dissent at 647-48, 651-52.

46 . We do not mean to suggest that the issue of Wal-Mart's liability could not ultimately be decided by a jury. If the district court concludes that the punitive damages class can be certified as a Rule 23(b)(3) class and adopts a hybrid approach, then a jury would have to decide the liability issue for both the Rule 23(b)(2) and the (b)(3) class.

47 . For this reason, the dissent’s observation, Dissent at 629-30, that four of the lead plaintiffs have left Wal-Mart since filing their complaint, is irrelevant. Their standing was established as of that date and they were under no obligation to continue to work for WalMart throughout several years of subsequent litigation in order to maintain their legal rights.

48 . This one exception related to Plaintiffs’ promotion claim. The district court determined that it would be unmanageable to fashion a remedy for the subset of the class for whom objective applicant data did not exist. See Dukes, 222 F.R.D. at 183. As discussed above, the district court did not abuse its discretion in its analysis and resolution of this issue.

49 . The trial plan described by the district court involved two stages. In Stage I, Plaintiffs would attempt to prove that Wal-Mart engaged in a pattern and practice of discrimination against the class via its company-wide employment policies and that the pattern or practice “was undertaken maliciously or recklessly in the face of a perceived risk that defendant’s actions would violate federal law.” Dukes, 222 F.R.D. at 173. If Plaintiffs prevailed in Stage I, the case would move to Stage II, the remedy phase. The first task in Stage II would be to fashion class-wide injunctive relief. The second task would be to calculate and distribute the back pay award. As to Plaintiffs’ promotional claim, a formula would be used to calculate the "lump sum” in back pay that Wal-Mart owes to the class (a procedure similar to that employed in Domingo, 727 F.2d at 1444-45). As to Plaintiffs’ equal pay claim, the court would examine Wal-Mart’s employment records to determine which class members were victims of this form of discrimination (and how much in back pay each is owed) to determine a second "lump sum” owed by Wal-Mart. Dukes, 222 F.R.D. at 174-86. A separate procedure would then be used to distribute these lump sums to those class members entitled to share in them — a stage in which Wal-Mart would no longer have an interest. Id. at 179 n. 49.

50 . The court was favored with an extraordinary variety of amicus briefs that were both thoughtful and helpful to it in its deliberations.

51 . This section states that "[n]o order of the court shall require ... the payment to [a person] of any back pay, if such individual ... was refused employment or advancement or was suspended or discharged for any reason other than [unlawful] discrimination” and that, "[o]n a claim in which an individual proves a violation under section 2000e-2(m) of this title and a respondent demonstrates that the respondent would have taken the same action in the absence of the impermissible motivating factor, the court ... shall not award damages.” Title VII, § 706(g)(2) (codified at 42 U.S.C. § 2000e-5(g)(2)).

52 . This statute states that the Federal Rules of Civil Procedure, including Rule 23 regarding class actions, "shall not abridge, enlarge or modify any substantive right. All laws in conflict with such rules shall be of no further force or effect after such rules have taken effect." 28 U.S.C. § 2072(b).

53 . The dissent also argues that Wal-Mart "has a statutory right, recognized by the Supreme Court, to prove that its actions against individual employees were not discriminatory,” which the dissent understands to mean that the district “court must allow up to 1.5 million individual determinations of liability.” Dissent at 629-30, 644. However, as the dissent recognizes, this case is at least largely one for prospective relief, which is available once a pattern and practice of discriminatory conduct is proven. The dissent ignores that the pattern and practice has to be proven on a group basis. Teamsters, 431 U.S. at 336, 97 S.Ct. 1843(Plaintiffs have "to prove more than the mere occurrence of isolated or 'accidental' or sporadic discriminatory acts. [They] ha[ve] to establish by a preponderance of the evidence that [gender] discrimination was the company's standard operating procedure — the regular rather than the unusual practice.”). It is both uneconomical and inefficient to do so in individual actions. In fact, since Teamsters, we have affirmatively held that the district court may award class-wide relief at the damages phase of a Title VII class action without individualized hearings. See Domingo, 727 F.2d at 1444 (“The facts of this case justify a departure from an individualized remedy for each claimant ... [because the defendant's] lack of objective hiring criteria and use of word-of-mouth recruitment ... makes it difficult to determine precisely which of the claimants would have been given a better job absent discrimination, but it is clear that many should have.” (citations omitted)). The dissent’s view would thus essentially preclude class members from either obtaining relief that is available or proving the prima facie portion of the damages case under Teamsters.

54 . Hilao was a 10,000+ plaintiff class action filed by Philippine nationals and their descendants who were allegedly victims of torture, summary execution, and "disappearance” at the hands of Ferdinand E. Marcos, the Philippines’ former president.

55 . For example, the defendant in Hilao argued that the trial plan “violated its rights to due process because individual questions apply to each subset of claims, i.e., whether the action was justified, the degree of injury, proximate cause, etc.” 103 F.3d at 785 (internal quotation marks omitted).

56 . We note that this procedure would allow Wal-Mart to present individual defenses in the randomly selected "sample cases,” thus revealing the approximate percentage of class members whose unequal pay or nonpromotion was due to something other than gender discrimination. The "invalid claim rate” revealed by this process would, as it did in Hilao, come very close to the invalid claim rate one would expect to find among the entire class.

57 . We do not suggest that this is the only conceivable way in which this class action could lawfully progress. Indeed, the district court may want to consider whether a more limited "test case” procedure similar to that employed in In re TMI Litig. Consol. Proceedings, 927 F.Supp. 834, 837 & n. 5 (M.D.Pa. 1996), would aid the parties in evaluating the strength of their respective claims.

And, of course, the option proposed by the district court may also remain viable; indeed, it appears that a number of circuits have approved of similar trial plans in discrimination cases. See, e.g., Segar, 738 F.2d at 1291 (explaining why a similar trial plan did not violate § 706(g)(2) of Title VII and commenting that, "[i]f effective relief for the victims of discrimination necessarily entails the risk that a few nonvictims might also benefit from the relief, then the employer, as a proven discriminator, must bear that risk”); see also Shipes, 987 F.2d at 316-19; Catlett v. Mo. Highway & Transp. Comm’n, 828 F.2d 1260, 1266-67 (8th Cir.1987). We point to the Hilao procedure above solely because this circuit has already considered and approved of that procedure.

---

Concurrence Opinion

GRABER, Circuit Judge,

concurring:

The majority and the dissent have written scholarly and complete explanations of them positions. What the length of their opinions may mask is the simplicity of the majority’s unremarkable holding:

Current female employees may maintain a Rule 23(b)(2) class action against their employer, seeking injunctive and declaratory relief and back pay on behalf of all the current female employees, when they challenge as discriminatory the effects of their employer’s company-wide policies.

If the employer had 500 female employees, I doubt that any of my colleagues would question the certification of such a class. Certification does not become an abuse of discretion merely because the class has 500,000 members.

I therefore concur fully in the majority opinion.

IKUTA, Circuit Judge, with whom Chief Judge KOZINSKI and Judges RYMER, SILVERMAN, and BEA join, dissenting:

No court has ever certified a class like this one, until now. And with good reason. In this case, six women who have worked in thirteen of Wal-Mart’s 3,400 stores seek to represent every woman who has worked in those stores over the course of the last decade — a class estimated in 2001 to include more than 1.5 million women. According to the plaintiffs’ theory, Wal-Mart has a corporate policy of discrimination which it implements through the discretionary decisions of store managers, resulting in class-wide injury in violation of Title VII of the Civil Rights Act of 1964.

But while the six plaintiffs allege they have suffered discrimination at the hands of a few individual store managers, they fail to present “[significant proof’ of a discriminatory policy or practice of WalMart that would make it possible to conclude that 1.5 million members of the proposed class suffered similar discrimination. See Gen. Tel. Co. of Sw. v. Falcon, 457 U.S. 147, 159 n. 15, 102 S.Ct. 2364, 72 L.Ed.2d 740 (1982). Without evidence of a company-wide discriminatory policy implemented by managers through their discretionary decisions, or other evidence of a discriminatory company-wide practice, there is nothing to bind these purported 1.5 million claims together in a single action. See id. at 158-59, 102 S.Ct. 2364.

Then there is the problem of individual hearings. Under Title VII, Wal-Mart has the right to raise affirmative defenses as to each class member’s claim. This means the court must allow up to 1.5 million individual determinations of liability. On its face, a class action of this sort makes no sense.

In certifying the class despite these obstacles, the district court abused its discretion. In affirming most of the district court’s determinations, the majority compounds the error, and creates a precedent for class action suits that departs from the language and intent of Rule 23 of the Federal Rules of Civil Procedure, ignores Supreme Court mandates, and neglects the rights of defendants. I respectfully dissent.

I

A brief review of the facts highlights the obstacles to certifying this class. WalMart is the largest private employer in the world, Dukes v. Wal-Mart Stores, Inc. (Dukes I), 222 F.R.D. 137, 141 (N.D.Cal. 2004), and the largest retail chain in the United States. At year-end 2001, when this suit was filed, Wal-Mart employed over 930,000 retail employees in hundreds of different jobs at roughly 3,400 stores nationwide.

Wal-Mart’s corporate structure is complex. The company divides its retail operation into seven divisions, six for WalMart and one for Sam’s Club.¹ Those divisions are split into 41 separate regions, each of which is overseen by a regional vice president. The regions are further divided into roughly 400 individual districts, which are headed by district managers. Each of Wal-Mart’s regions consists of 80 to 85 stores, with Wal-Mart employing anywhere from 80 to 500 people per store.

Individual stores are run by store managers who are responsible for hiring and promoting the hourly employees in their respective stores. At each store, assistant managers report to store managers; these management positions are salaried, and employees may be promoted to these positions after first working as entry-level, non-salaried management trainees. WalMart’s retail employees work hourly in 53 different departments and 170 different job classifications; these positions include cashiers, associates, team leads, and department managers.

Wal-Mart gives its managers substantial discretion in both pay and promotion decisions. Store managers make pay decisions for hourly employees, subject only to a general pay structure. Id. at 146. Store managers “are allowed to depart from the minimum start rates, within a two dollar per hour range, without being constrained by objective criteria and with limited oversight.” Id. at 146^17. Wal-Mart also allows store managers to increase pay for exceptional performance. Id. at 147. District managers set the pay for salaried employees. They have “discretion to set pay rates with little guidance and limited oversight.” Id.

Managers likewise have substantial discretion in making promotion decisions. Wal-Mart has corporate guidelines regarding entrance requirements for its management trainee program, but individual store managers may apply their own subjective criteria in selecting employees who will be invited to enter the program. Id. at 148. Decisions regarding advancement into the highest store-level managerial positions (assistant manager, co-manager, and store manager) are “similarly based on subjective assessments beyond adherence to corporate minimum guidelines.” Id. Discretion is “further compounded by the fact that the company does not monitor the promotion decisions being made or otherwise systematically review the grounds on which candidates are selected for promotion.” Id. at 149.

This action was brought by six women on behalf of all women who were employed in any retail store position throughout Wal-Mart. Two of these six women still work at Wal-Mart. Of the other four, one was terminated,² and three left Wal-Mart before the complaint was even filed.³ In the aggregate, these six plaintiffs worked at ten stores in California, one store in Texas, one store in Oklahoma, and one store in Missouri. Three plaintiffs allege that they expressed interest in, or applied for, various promotions at the store in which they were then employed, but did not receive them. These three women allege that men were promoted instead. Only one of these plaintiffs, Christine Kwapnoski, has held a management position. Two plaintiffs allege that similarly situated men received higher pay. One plaintiff alleges that a less-qualified man received the same pay as she did.

Together, these six women assert that they are adequate class representatives of a class comprising “[a]ll women employed at any Wal-Mart domestic retail store at any time since December 26, 1998 who have been or may be subjected to WalMart’s challenged pay and management track promotions policies and practices.” Id. at 141-42. The class includes every woman working in Wal-Mart’s retail stores, which by definition encompasses salaried store managers who were responsible for the allegedly discriminatory hiring decisions,⁴ assistant managers and co-managers who were also responsible for much of the alleged discriminatory treatment, and hourly workers whose in-store positions range from personnel clerk to deli manager. At the time the suit was filed, the class was estimated to include 1.5 million women.⁵

II

Before it can certify a class, a court must ensure that the proposed class meets the standards set forth in Rule 23 of the Federal Rules of Civil Procedure. Rule 23 is a procedural rule that permits courts to aggregate the legal claims of multiple parties when it is efficient and fair to do so.⁶ It does not alter the substance of claims or the plaintiffs’ burden to prove their claims. See 28 U.S.C. § 2072(b).

We review the district court’s class certification decisions for abuse of discretion. Staton v. Boeing Co., 327 F.3d 938, 953 (9th Cir.2003). Abuse of discretion occurs when the district court commits a legal error, or when it “relies upon an improper factor, omits consideration of a factor entitled to substantial weight, or mulls the correct mix of factors but makes a clear error of judgment in assaying them.” Vinole v. Countrywide Home Loans, Inc., 571 F.3d 935, 939(9th Cir.2009) (citation and internal quotation marks omitted).

Here, the district court abused its discretion in two ways. First, it failed to follow the Supreme Court’s direction to “evaluate carefully the legitimacy of the named plaintiffs plea that he is a proper class representative under Rule 23(a),” Falcon, 457 U.S. at 160, 102 S.Ct. 2364, and to ensure “after a rigorous analysis” that the prerequisites of Rule 23(a) have been met, id. at 161, 102 S.Ct. 2364. Second, the district court erred in ignoring Wal-Mart’s statutory right to raise defenses to liability for back pay and punitive damages under Title VII, see 42 U.S.C. § 2000e-5(g)(2); Rules Enabling Act, 28 U.S.C. § 2072(b), and therefore abused its discretion in holding that the proposed class could be certified under Rule 23(b)(2). I would reverse and remand this action to the district court to correct these errors, which I address below in turn.

A

In Falcon, the Supreme Court held that when a small number of plaintiffs allege company-wide discrimination and attempt to certify a company-wide class, the district court must undertake a “rigorous analysis” to determine whether the Rule 23(a) commonality and typicality factors exist.⁷ Falcon, 457 U.S. at 157, 161, 102 S.Ct. 2364; see Staton, 327 F.3d at 953-54(calling a proposed class of 15,000 members throughout six facilities an “ambitious” claim for commonality that was “therefore especially worthy of scrutiny”). As Falcon explained, the plaintiffs own experience of discrimination does not allow the district court to infer that “discriminatory treatment is typical of [the employer’s employment] practices.” 457 U.S. at 158, 102 S.Ct. 2364; see E. Tex. Motor Freight Sys., Inc. v. Rodriguez, 131 U.S. 395, 403-04, 97 S.Ct. 1891, 52 L.Ed.2d 153 (1977). Indeed, the Court warned, “if one allegation of specific discriminatory treatment were sufficient to support an across-the-board attack, every Title VII case would be a potential company-wide class action. ” Falcon, 157 U.S. at 159, 102 S.Ct. 2361. Instead, to maintain a company-wide class action based on discrimination, the plaintiff must bridge the “wide gap” between: (1) the plaintiff’s own discriminatory treatment; and (2) the existence of a class that has suffered the same injury as the plaintiff as a result of a company-wide discriminatory policy. Id. at 157, 102 S.Ct. 2364.

This principle is simple common sense. A female employee in a store in California, for example, may have a valid claim that her supervisor discriminated against her when making decisions regarding promotion opportunities. But this individual claim would not by itself entitle the California employee to bring the alleged discrimination claims of female employees in a store in Wyoming. Absent evidence that the Wyoming employees had been subject to the same sort of discrimination and could bring the same sort of claims, such a proposed class would clearly fail the Rule 23(a) commonality and typicality requirements. A fortiori, a female employee in California could not represent a class of all female employees in Wal-Mart absent similar proof to bridge the gap between her claim and the existence of company-wide discrimination.

The Supreme Court has identified two situations in which a plaintiff could potentially bridge this gap. The first is where a plaintiff alleges that an employer “used a biased testing procedure” to evaluate all members of the proposed class. Id. at 159 n. 15. This makes sense because if the employer used a biased testing procedure throughout the company, then every employee subject to the test would have a similar claim.

The Court’s second example applies directly to this case: a plaintiff “conceivably” could bring a company-wide Title VII action if the plaintiff adduced “[significant proof that an employer operated under a general policy of discrimination” and the discriminatory policy was implemented through “entirely subjective decision making processes” in a manner that affected all members of the class. Id. This example also makes sense. As the majority acknowledges, Maj. Op. at 612, subjective decision making by itself is not a discriminatory practice. See Watson v. Fort Worth Bank & Trust, 487 U.S. 977, 988, 990, 108 S.Ct. 2777, 101 L.Ed.2d 827 (1988) (“It is true, to be sure, that an employer’s policy of leaving promotion decisions to the unchecked discretion of lower level supervisors should itself raise no inference of discriminatory conduct.”). But if an employer has a company-wide policy of discrimination against women, and the employer implements this policy through company-wide subjective decision making, then the plaintiff could reasonably claim that all female employees were subject to the same discriminatory employment practice. As in our prior example, if the female employee in the California store could offer proof of a company-wide policy of discrimination against women, she would be able to allege that female employees in stores throughout the country suffered a similar discriminatory injury as she did.

Although Falcon did not spell out what constitutes “significant proof’ of the existence of a general policy of discrimination, it did make one thing clear: Evidence of discrete instances of discrimination are insufficient to sustain an inference of an employer’s general policy and do not rise to the level of “significant proof.” See 457 U.S. at 158, 102 S.Ct. 2364(holding that evidence that the plaintiff “was passed over for promotion when several less deserving whites were advanced” may support a claim of individual discrimination, but “would not necessarily justify” a company-wide claim of discrimination). This is the standard that must be applied to the claims here.

B

The majority offers several reasons for concluding it is not bound by Falcon’s “significant proof’ requirement. First, the majority claims that this requirement has little weight because it was a “hypothetical in clear dicta.”⁸ Maj. Op. at 595 n.15. This contention has no merit. We are bound by the Supreme Court’s instructions whether they are stated in the first sentence of an opinion or in the final footnote. Falcon’s analysis of Rule 23(a) is controlling here, given that it is the sole Supreme Court case addressing Rule 23(a) in the Title VII discrimination context and is directly on point in this case. Moreover, the significant proof requirement is clearly a reasonable interpretation of Rule 23(a); a plaintiff cannot bring a company-wide discrimination class action unless there is sufficient evidence that the alleged discriminatory injury actually affected employees throughout the company.

Second, the majority attempts to cabin Falcon to its facts, asserting that the “significant proof’ requirement applies only when the plaintiff is an employee of the company and other members of the purported class are job applicants. In such a case, the majority asserts, the plaintiff and the purported members of the class have “distinct legal theories of recovery.” Maj. Op. at 595. The majority’s efforts to distinguish Falcon on this ground are unpersuasive. Falcon’s determination that the district court erred in certifying a class of existing employees and job applicants was based on the general principle that a plaintiffs claims must “fairly encompass” those of the class to meet the Rule 23(a) commonality and typicality requirements. See Falcon, 457 U.S. at 156, 102 S.Ct. 2364. In elaborating further on this principle, the Supreme Court explained that plaintiffs who seek to bridge the gap between their claims and those of the class members by alleging a general company-wide policy of discrimination must adduce “significant proof’ that such a policy exists. Id. at 159 n. 15, 102 S.Ct. 3014. The Supreme Court did not limit this principle to classes combining plaintiffs with hiring and promotion claims.

Finally, the majority claims that Falcon’s “significant proof’ requirement conflates the class certification and merits phases of the litigation. Maj. Op. at 596 n.17. This is incorrect. Falcon does not require plaintiffs to prove the merits of their claim; here, for example, plaintiffs need not establish a prima facie case of discrimination that can survive the employer’s rebuttal evidence, nor need they prove their claims for individual relief. See Int’l Bhd. of Teamsters v. United States, 431 U.S. 324, 360, 97 S.Ct. 1843, 52 L.Ed.2d 396 (1977). But Falcon does require the plaintiffs to put forth some significant proof, i.e., evidence beyond allegations, that a general policy of discrimination exists. Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. 2364. More important, as Falcon states (and as the majority acknowledges), courts at the class certification stage must “probe behind the pleadings” to examine evidence which goes to the requirements of Rule 23, even if such evidence also relates to the underlying merits of the case. Id. at 160, 102 S.Ct. 2364. In this context, the degree of overlap between the merits determination and the determination that the class meets the Rule 23 requirements is largely irrelevant. Id. Rather, Falcon’s principle that a handful of individual incidents of discrimination are insufficient for class certification is applicable both at the merits stage and the class certification stage. See Cooper v. Fed. Reserve Bank, 467 U.S. 867, 876-77, 104 S.Ct. 2794, 81 L.Ed.2d 718 (1984). If plaintiffs cannot demonstrate that the proposed class was subject to a general policy of discrimination, then the class action is not an efficient mechanism for pursuing relief, and the district court may not certify the class.

C

We must consider plaintiffs’ motion for class certification under Rule 23(a) in light of Falcon’s requirements. In this case, the plaintiffs asked the district court to certify a class of 1.5 million women based on: (i) 120 anecdotes; (ii) statistical evidence; and (iii) expert testimony.⁹ In light of Falcon, the district court’s responsibility to conduct a rigorous scrutiny of this evidence was clear. As explained below, the evidence does not come close to meeting the Falcon requirements for demonstrating commonality and typicality.

1

On its face, 120 anecdotes, or one anecdote for every 12,500 class members, does not support plaintiffs’ claim that Wal-Mart had a company-wide policy of discrimination. The affidavits describe the affiants’ experiences in, at most, 235 of Wal-Mart’s 3,400 stores, meaning that the affidavits provide no information about working conditions in over 3,100 stores. A single affidavit from a single store in Michigan tells little about whether there is discrimination at each of the other 72 stores in Michigan, let alone the rest of the company. Cooper, 467 U.S. at 877-78, 104 S.Ct. 2794; see Cooper v. S. Co., 390 F.3d 695, 714-15 (11th Cir.2004) (holding that evidence of discrimination in one part of a company does not necessarily give rise to an inference of discrimination in a different part of the company), overruled on other grounds by Ash v. Tyson Foods, Inc., 546 U.S. 454, 457-58, 126 S.Ct. 1195, 163 L.Ed.2d 1053 (2006) (per curiam).

Nor are the 120 affidavits geographically representative of Wal-Mart as a whole. See Cooper, 390 F.3d at 719(upholding the district court’s determination that, “given the sheer size and geographically dispersed nature of the defendants’ workforce, the anecdotal evidence — disturbing as some of it may have been — was inadequate to establish discrimination class-wide”). Here, more than half of the 120 women who submitted affidavits are concentrated in only six states — Alabama, California, Florida, Missouri, Texas, and Wisconsin — -where less than a third of Wal-Mart stores are located. There are no more than one or two affidavits relating to Wal-Mart operations in half the states, and 14 states are not represented at all.¹⁰

The Supreme Court has held that a court may find a company-wide policy of discrimination where plaintiffs have offered a substantial number of affidavits compared to the size of the class, along with sufficient statistical evidence.¹¹ Teamsters, 431 U.S. at 339, 97 S.Ct. 1843. In Teamsters, for example, 40 anecdotes of discrimination out of a class of some 334 employees (one out of every eight class members) were held probative of a pattern or practice of discrimination. See United States v. T.I.M.E.-D.C., Inc., 517 F.2d 299, 308 (5th Cir.1975), overruled by Teamsters, 431 U.S. 324, 97 S.Ct. 1843, 52 L.Ed.2d 396. These 40 anecdotes were from employees spread throughout the company and represented seven of the company’s ten largest operations. See id. at 315 & n. 30.

On the other hand, the Supreme Court has concluded that a handful of discriminatory incidents is insufficient evidence of a company-wide policy of discrimination to justify certification of a company-wide class. Falcon, 457 U.S. at 159, 102 S.Ct. 2364. A plaintiffs individual experience of discriminatory treatment does not itself raise the inference that such treatment is typical of the employer’s practices. Id. at 158, 102 S.Ct. 2364; see Cooper, 467 U.S. at 877-78, 104 S.Ct. 2794 (“[A] class plaintiffs attempt to prove the existence of a company-wide policy, or even a consistent practice within a given department, may fail even though discrimination against one or two individuals has been proved.”).

Here, 120 affidavits compared to 1.5 million members of the class amount to nothing more than evidence of “isolated or sporadic” incidents of discrimination, Cooper, 467 U.S. at 876, 104 S.Ct. 2794, which are insufficient to show a company-wide policy of discrimination. See Falcon, 457 U.S. at 159, 102 S.Ct. 2364. The affiants claim discrimination in different forms, at the hands of different people, in different stores, in different parts of the country, at different times, and under a constellation of facts unique to each individual. Whether or not these 120 employees have actionable discrimination claims against the store and management where they worked, their individual stories do not constitute significant proof that Wal-Mart has adopted a general policy of discrimination or that such a policy prevails in the remaining 3,100 stores. In relying on such scanty anecdotal evidence to support class certification, the district court failed to “probe behind the pleadings” to determine whether the six plaintiffs actually did “possess the same interest and suffer the same injury” as the proposed class members. Id. at 160, 102 S.Ct. 2364.

The majority makes the same mistake. Discarding Wal-Mart’s objection that the district court abused its discretion in considering the 120 affidavits to be an adequate basis for certifying the class, the majority states, “we find no authority requiring or even suggesting that a plaintiff class submit a specific number of declarations for such evidence to have any value.” Maj. Op. at 610. This is a strawman argument. The Supreme Court has not specified a threshold number of affidavits that a plaintiff alleging company-wide discrimination must have in hand. But if the plaintiffs affidavits do not raise the inference that the employer’s practices are “motivated by a policy of [gender] discrimination” pervading the employer’s company, they do not support class certification. Falcon, 457 U.S. at 158, 102 S.Ct. 2364. Here, the majority’s conclusion that one declaration for every 12,500 women, amounting to a handful of declarations scattered among 3,400 stores in fifty states, “raise[s] an inference of common discriminatory experiences,” Maj. Op. at 611, does not pass even the straight-faced test.

2

The plaintiffs’ statistical evidence is no better. Plaintiffs relied most heavily on the statistical study produced by Dr. Richard Drogin, a retired professor of statistics from California State University, Hayward. Drogin conducted a regional analysis comparing the percentage of women promoted into management positions at Wal-Mart with the percentage of women in the available pool of hourly workers, and concluded that women were underrepresented in management in almost every one of Wal-Mart’s 41 regions. Drogin also compared the earnings of women to the earnings of men at Wal-Mart and concluded that in each region, Wal-Mart pays women less than men in comparable hourly positions.

Wal-Mart objected to this evidence on the ground that data aggregated at the regional or national level did not demonstrate there was a general policy of discrimination throughout Wal-Mart’s 3,400 individual retail stores. The district court dismissed these objections. Rejecting what it characterized as Wal-Mart’s effort “to engage the Court in a merits evaluation of the expert opinions,” the district court stated it would “delve[] into the substance of the expert testimony only to the extent necessary to determine if it is sufficiently probative of an inference of discrimination to create a common question as to the existence of a pattern and practice of gender discrimination at WalMart.” Dukes I, 222 F.R.D. at 155. Applying this standard, the court held that Drogin’s regional analysis was not “lacking in probative value,” was “at least a reasonable means of conducting a statistical analysis,” id. at 159, and therefore was sufficient to “create an inference of discrimination” for class certification purposes, id. at 164.

The district court’s superficial examination of Drogin’s statistics constituted legal error. The plaintiffs’ class proposal hinged on its proof that Wal-Mart had a general policy of discrimination; absent convincing proof on that point, the plaintiffs could not bridge the gap between their discrimination claims and the purported claims of the class. Accordingly, the district court was obliged to scrutinize plaintiffs’ evidence and make a reasoned determination as to whether it constituted significant proof that Wal-Mart had a general policy of discrimination, and thus whether the requirements of Rule 23(a) had been met. See Falcon, 457 U.S. at 161, 102 S.Ct. 2364.

The district court expressly rejected this responsibility. Instead of rigorously analyzing whether the plaintiffs’ evidence was significant proof of a general policy of discrimination, the district court made it Wal-Mart’s burden to prove that Drogin’s statistics were no longer probative or were “fatally flawed.” See Dukes I, 222 F.R.D. at 161-62; see also id. at 159(coneluding that Wal-Mart failed to persuade the court “that Dr. Drogin’s regional analysis should be rejected ... because it is lacking in probative value”); id. at 160(coneluding that “[Wal-Mart] has not discredited or nullified Dr. Drogin’s results” with respect to Wal-Mart’s objection that Drogin’s methodology failed to account for a variety of relevant factors). The majority affirms this error by simply labeling the district court’s analysis as “rigorous” and insisting (without analysis) that the district court adequately probed behind the pleadings.

Had the district court properly analyzed the evidence, it could not have concluded that Drogin’s statistics constituted significant proof of plaintiffs’ theory. Information about disparities at the regional and national level does not establish the existence of disparities at individual stores, let alone raise the inference that a company-wide policy of discrimination is implemented by discretionary decisions at the store and district level. As Wal-Mart’s statistical expert, Dr. Joan Haworth, explained, the statistical disparities at the regional level could be due to decisions made at only a small percentage of Wal Mart stores.¹² Indeed, Haworth concluded that, when data is considered at the store level, over 90 percent of Wal-Mart’s stores showed no statistical difference in the hourly pay rates between men and women associates with similar work-related characteristics. Although Drogin argued that it was proper to conduct an analysis on a regional level because “the subjective decision-making in compensation and promotions takes place within parameters and guidelines that are highly uniform, and within a strong corporate culture,” Dukes I, 222 F.R.D. at 157, this argument makes little sense. Among other things, this argument is contrary to the thrust of plaintiffs’ legal theory, namely, that the decisions of individual managers regarding pay and promotions are subjective and not subject to uniform “parameters and guidelines.” It is also contrary to the district court’s own determination that “in-store pay and promotion decisions are largely subjective and made within a substantial range of discretion by store or district level managers.” Id. at 152. More important, because statistical disparities may exist at the regional level even if the majority (or all) of the store managers made nondiscriminatory decisions at the store level, Drogin’s data is not sufficiently probative of conditions at the store level to constitute significant proof of the existence of a class subjected to a policy of company-wide discrimination.¹³

The majority disposes of Wal-Mart’s objection that regional data is not indicative of a general discriminatory policy at individual stores by criticizing Wal-Mart’s statistics, which aggregated data at the sub-store rather than the store level, an approach the district court had previously allowed.

Dukes I, 222 F.R.D. at 157 n. 25. But the quality of the defendant’s statistics is not the issue here; rather, it is the plaintiffs’ burden to produce significant proof of a policy of discrimination across Wal-Mart that manifested itself in similar ways throughout the company. See Falcon, 457 U.S. at 160, 102 S.Ct. 2364. Plaintiffs’ statistics lack probative value even without Wal-Mart’s opposing data.

If scientific, technical, or other specialized knowledge will assist the trier of fact to understand the evidence or to determine a fact in issue, a witness qualified as an expert by knowledge, skill, experience, training, or education, may testify thereto in the form of an opinion or otherwise, if (1) the testimony is based upon sufficient facts or data, (2) the testimony is the product of reliable principles and methods, and (3) the witness has applied the principles and methods reliably to the facts of the case.

3

Finally, the district court committed legal error by failing to test the reliability of the expert opinion of William Bielby, Ph. D., as required by Federal Rule of Evidence 702¹⁴ and Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579, 589, 113 S.Ct. 2786, 125 L.Ed.2d 469 (1993). In brief, Bielby presented his theory that subjective decision making, such as that which occurs at Wal-Mart stores, is “susceptible to unconscious discriminatory impulses.” ¹⁵ Wal-Mart moved to strike Bielby’s testimony on the ground that his opinion was not reliable. According to Wal-Mart, Bielby failed to apply the same standards to his work in preparing his testimony that he applied to his work as a scientist. Wal-Mart alleged, inter alia, that Bielby had misrepresented aspects of the literature upon which he relied, made unsustainable extrapolations from that literature, failed to consider evidence that tended to undermine his theory, and failed to test his data.

Instead of evaluating the reliability of Bielby’s report in light of Wal-Mart’s arguments, the district court denied WalMart’s motion on the ground that “courts should not even apply the full Daubert ‘gatekeeper’ standard” at the class certification stage. Dukes II, 222 F.R.D. at 191. Rather, the district court held that “[t]he appropriate question at this stage of the litigation is not whether Dr. Bielby can make a conclusive determination, but whether [his opinion] could add probative value to the inference of discrimination that plaintiffs allege.” Dukes I, 222 F.R.D. at 154. Concluding that the report was not “so flawed that it lacks sufficient probative value to be considered in assessing commonality,” the district court relied on Bielby’s report in making its determination that plaintiffs met the Rule 23(a) requirements. Dukes II, 222 F.R.D. at 192.

In so ruling, the district court misunderstood Daubert and the level of inquiry required at the class certification stage. The purpose of conducting a Daubert inquiry is to ensure that proffered expert testimony is relevant and reliable. Kumho Tire Co. v. Carmichael, 526 U.S. 137, 152, 119 S.Ct. 1167, 143 L.Ed.2d 238 (1999). “It is to make certain that an expert, whether basing testimony upon professional studies or personal experience, employs in the courtroom the same level of intellectual rigor that characterizes the practice of an expert in the relevant field.” Id. To this end, a court must make “a preliminary assessment of whether the reasoning or methodology underlying the testimony is scientifically valid and of whether that reasoning or methodology properly can be applied to the facts in issue.” Daubert, 509 U.S. at 592-93, 113 S.Ct. 2786. Courts are obliged to address allegations that an expert’s opinion is not supported by the data, and may reject “opinion evidence that is connected to existing data only by the ipse dixit of the expert.” Gen. Elec. Co. v. Joiner, 522 U.S. 136, 146, 118 S.Ct. 512, 139 L.Ed.2d 508 (1997).

These principles are equally applicable in the class certification context. The district court must “assess all of the relevant evidence admitted at the class certification stage and determine whether each Rule 23 requirement has been met, just as the judge would resolve a dispute about any other threshold prerequisite for continuing a lawsuit.” In re Initial Pub. Offerings Sec. Litig., 471 F.3d 24, 42 (2d Cir.2006). Like any other evidence, expert evidence introduced to “establish a component of a Rule 23 requirement” must be reliable; it is not enough that the expert testimony is “not fatally flawed.” Id.; see also Am. Honda Motor Co. v. Allen, 600 F.3d 813, 815-16, (7th Cir.2010) (holding that the court must conduct a Daubert analysis on key expert testimony even at the class certification stage); In re Hydrogen Peroxide Antitrust Litig., 552 F.3d 305, 323 (3d Cir.2008) (“Expert opinion with respect to class certification, like any matter relevant to a Rule 23 requirement, calls for rigorous analysis.”). A court cannot reasonably certify a class based on evidence that is not “based upon sufficient facts or data, ... the product of reliable principles and methods,” and the product of reliable application of the methods to the facts. Fed.R.Evid. 702. Although a district court has considerable discretion under Daubert, there is no basis for the district court’s conclusion that it can forego a reliability inquiry altogether. The district court abused its discretion in doing so.¹⁶

The majority’s discussion of this claim misses the mark entirely. First, instead of engaging the question whether the district court properly tested the reliability of Bielby’s testimony, the majority simply concludes that “Dr. Bielby presented scientifically reliable evidence.” Maj. Op. at 603. This not only assumes the answer to the very question at issue, but is wrong on its face: The district court explicitly did not consider whether the evidence was “scientifically reliable,” holding instead that the evidence was not “so flawed that it lacks sufficient probative value to be considered in assessing” whether class certification was appropriate. Dukes II, 222 F.R.D. at 192.¹⁷

The majority focuses its discussion on what it characterizes as Wal-Mart’s objection to the persuasiveness of Bielby’s testimony. According to the majority, WalMart “did not (and does not) challenge Dr. Bielby’s methodology.” Maj. Op. at 602. In fact, Wal-Mart never challenged the persuasiveness of Bielby’s testimony, but rather challenged Bielby’s methodology, a challenge that is proper under Daubert. See Daubert, 509 U.S. at 592-93, 113 S.Ct. 2786; see also Joiner, 522 U.S. at 144, 118 S.Ct. 512. By missing this point, the majority erroneously approves the district court’s reliance on an arguably unreliable expert opinion.

D

When plaintiffs’ evidence is subjected to the rigorous inquiry required by Falcon, it is inadequate to bridge the gap between the six plaintiffs’ claims of individual discrimination and a class-wide claim of company-wide discrimination. None of plaintiffs’ evidence is probative of company-wide discrimination. Every piece of evidence merely purports to support another. While plaintiffs’ anecdotes do not show company-wide discrimination, plaintiffs argue they support the statistical evidence. The statistics are not probative of a company-wide policy of discrimination, but plaintiffs allege they may be “attributable” to such a policy when viewed in connection with Wal-Mart’s uniform corporate policies. Maj. Op. at 600-01. The uniform corporate policies are not themselves discriminatory but, according to plaintiffs, provide a potential “conduit” for discrimination. Maj. Op. at 600. The expert opinions do not point to discrimination on a company-wide basis, but merely “support ] the existence of company-wide policies and practices that likely include a culture of gender stereotyping.” Maj. Op. at 600 (emphasis added). And Wal-Mart’s corporate policy of subjective decision making is not discriminatory in itself but, plaintiffs urge, may be evidence of company-wide discrimination in light of the statistical evidence and anecdotes. Maj. Op. at 611. Like the proverbial shell game, the plaintiffs’ circular presentation cannot conceal the fact that they have failed to offer any significant proof of a company-wide policy of discrimination, no matter which shell is lifted.

By taking the district court’s determination at face value, the majority erroneously accepts a chain of weak inferences as sufficient to carry plaintiffs’ burden of adducing significant proof that Wal-Mart “operated under a general policy of discrimination” that affected all members of the class. Falcon, 457 U.S. at 159 n. 15, 102 S.Ct. 2364. Any reasonable scrutiny of the evidence in this case compels the conclusion that although the six plaintiffs here may have individualized claims of discrimination, they cannot represent a class of 1.5 million past and present employees. The district court abused its discretion in holding that the plaintiffs met the commonality and typicality requirements of Rule 23(a).

Ill

Even assuming the district court was correct in concluding that the proposed class satisfied the Rule 23(a)-standard, the district court made two crucial errors when it certified the class under Rule 23(b)(2). First, the district court failed to consider whether it could protect the parties’ substantive rights in the class action context. Although the plaintiffs were bringing Title VII claims, the district court erroneously concluded that it had no obligation to allow Wal-Mart to raise the statutory defenses provided by Title VII. This was an error: A court must ensure that its certification of a class does not affect the substantive rights of either party. See Rules Enabling Act, 28 U.S.C. § 2072(b);¹⁸ Ortiz v. Fibreboard Corp., 527 U.S. 815, 845, 119 S.Ct. 2295, 144 L.Ed.2d 715 (1999) (“[N]o reading of the Rule can ignore the Act’s mandate that rules of procedure shall not abridge, enlarge or modify any substantive right.” (internal quotation marks omitted) (quoting Am-chem Prods., Inc. v. Windsor, 521 U.S. 591, 613, 117 S.Ct. 2231, 138 L.Ed.2d 689 (1997))). This mistake led to a second legal error: Because the district court turned a blind eye to Wal-Mart’s right to individual hearings, the district court failed to realize that the class did not fit the historical model of Rule 23(b)(2), and therefore could not be certified under that section of the rule.

A

The parties’ substantive rights in this ease are defined by Title VII. The majority acknowledges that the legal and factual frameworks of different causes of action can impact the outcome of a court’s Rule 23 determination. Maj. Op. at 580, 591-92. Yet the majority fails to address the specific legal and factual framework of Title VII or consider how it impacts the certification of plaintiffs’ proposed class. It is therefore necessary to provide some background regarding the structure of claims and defenses under Title VII before discussing the majority and district court’s error in certifying the proposed class under Rule 23(b)(2).

1

Under Title VII, the plaintiff has the ultimate burden of establishing that the employer is engaged in an unlawful employment practice described in 42 U.S.C. § 2000e-2. If the plaintiff prevails in a Title VII action, the court “may enjoin the respondent from engaging in such unlawful employment practice, and order such affirmative action as may be appropriate, which may include, but is not limited to, reinstatement or hiring of employees, with or without back pay ..., or any other equitable relief as the court deems appropriate.” 42 U.S.C. § 2000e-5(g)(1). In cases where the plaintiff alleges that the employer engaged in intentional discrimination, the plaintiff may also seek compensatory and punitive damages. § 1981a(a)(l). Punitive damages are available only if the plaintiff can show the employer acted “with malice or with reckless indifference to the federally protected rights of an aggrieved individual.” § 1981a(b)(1); e.g., Landgraf v. USI Film Prods., 511 U.S. 244, 254, 114 S.Ct. 1483 (1994).

Title VII affords defendant-employers certain affirmative defenses. If the employer can prove that an adverse employment action against an individual employee was “for any reason other than discrimination on account of race, color, religion, sex, or national origin or in violation of section 2000e-3(a) of this title,” a court shall not order the “hiring, reinstatement, or promotion of an individual as an employee, or the payment to him of any back pay.” § 2000e-5(g)(2)(A). Even if the employee proves that the adverse employment action was motivated in part by a discriminatory motive, see § 2000e-2(m), a court cannot award damages or back pay if the employer can prove that it “would have taken the same action in the absence of the impermissible motivating factor,” § 2000e-5(g)(2)(B)(ii). See Costa v. Desert Palace, Inc., 299 F.3d 838, 848 (9th Cir.2002) (explaining that in a mixed motive case, an employer “may assert an affirmative defense to bar certain types of relief by showing the absence of ‘but for’ causation”), aff'd, 539 U.S. 90, 123 S.Ct. 2148, 156 L.Ed.2d 84. If the employer is successful in proving an affirmative defense, Title VII limits the court to granting declaratory or injunctive relief, as well as certain attorneys’ fees and costs. 42 U.S.C. § 2000e-5(g)(2)(B)(ii).

These defenses are available to a defendant regardless of whether the plaintiffs allege discrimination against a single individual or against a class. See 28 U.S.C. § 2072(b); Teamsters, 431 U.S. at 360, 97 S.Ct. 1843(citing Franks v. Bowman Trans. Co., 424 U.S. 747, 96 S.Ct. 1251, 47 L.Ed.2d 444 (1976)). An employer cannot be deprived of its statutory right to raise individualized defenses to claims for monetary relief merely because plaintiffs characterize their claim as a pattern or practice of discrimination and bring the suit on behalf of a class. Instead, the Supreme Court has created a framework for litigating pattern-or-practice claims that provides employers the opportunity to put forth individual defenses. See Teamsters, 431 U.S. at 360-62, 97 S.Ct. 1843.¹⁹

The first phase of a Title VII pattern-or-practice lawsuit focuses on company-wide practices. The party bringing the suit must first attempt “to establish by a preponderance of the evidence that ... discrimination was the company’s standard operating procedure[,] the regular rather than the unusual practice.” Id. at 336, 97 S.Ct. 1843. In this phase, the plaintiff need not show that each member of the class “was a victim of the employer’s discriminatory policy.” Id. at 360, 97 S.Ct. 1843. If the plaintiff “establish[es] a prima facie case that such a policy existed[,]” the burden shifts to the employer to show there was no such pattern of discrimination at a company-wide level. Id. If the employer fails to rebut the inference of discrimination arising from the plaintiffs’ prima facie case, the district court may “conclude that a violation has occurred” and order prospective relief. Id. at 361, 97 S.Ct. 1843.

But if plaintiffs seek individual relief such as reinstatement or back pay, the district court must conduct additional proceedings in the second phase of the trial “to determine the scope of individual relief.” Id. In this second phase, “[t]he proof of the pattern or practice supports an inference that any particular employment decision, during the period in which the discriminatory policy was in force, was made in pursuit of that policy.” Id. at 362, 97 S.Ct. 1843. Therefore, individual class members may establish a prima facie case merely by showing that an individual “unsuccessfully applied for a job and therefore was a potential victim of the proved discrimination.” Id. (footnote omitted). The burden then shifts to the employer to raise its affirmative defenses and to “demonstrate that the individual applicant was denied an employment opportunity for lawful reasons.” Id. Teamsters acknowledged that the district court’s task of determining individual relief would “not be a simple one,” because “the court will have to make a substantial number of individual determinations in deciding which of the minority employees were actual victims of the company’s discriminatory practices.”²⁰ Id. at 371-72, 97 S.Ct. 1843.

As should be clear, claims for monetary relief cannot be resolved until this second phase. See, e.g., Reeb v. Ohio Dep’t of Rehab. & Corr., 435 F.3d 639, 651 (6th Cir.2006); Lemon v. Int’l Union of Operating Eng’rs, Local No. 139, 216 F.3d 577, 581 (7th Cir.2000); Allison, 151 F.3d at 417. This applies equally to punitive damages. Because the Supreme Court has indicated that punitive damages must bear a “reasonable relationship to compensatory damages,” BMW of N. Am., Inc. v. Gore, 517 U.S. 559, 580, 116 S.Ct. 1589, 134 L.Ed.2d 809 (1996) (internal quotation marks omitted), “punitive damages must be determined after proof of liability to individual plaintiffs at the second stage of a pattern or practice case, not upon the mere finding of general liability to the class at the first stage,” Allison, 151 F.3d at 418; see also 42 U.S.C. § 1981a(b)(1)(an employer may avoid punitive damages by showing lack of requisite mens rea as to each aggrieved individual); Kolstad v. Am. Dental Ass’n, 527 U.S. 526, 545-46, 119 S.Ct. 2118,144 L.Ed.2d 494 (1999) (holding that an employer may avoid punitive damages under § 1981a if it has made good-faith efforts to prevent discrimination in the workplace); Rodriguez-Torres v. Caribbean Forms Mfr., Inc., 399 F.3d 52, 64 (1st Cir.2005) (holding that an employer’s good faith efforts to implement an anti-discrimination policy in the workplace is an affirmative defense to an award of punitive damages under § 1981a).

In sum, a determination that an employer has engaged in a pattern or practice of discrimination in violation of Title VII entitles the class to prospective relief and creates a presumption of liability for individual relief. The employer then has a statutory right, recognized by the Supreme Court, to prove that its actions against individual employees were not discriminatory. The Rules Enabling Act prohibits a district court from depriving an employer of this right simply to facilitate the certification of a class action.²¹ See 28 U.S.C. § 2072(b).

2

In this case, notwithstanding the requirements of Title VII, the district court rejected Wal-Mart’s arguments that it was entitled to raise a defense with respect to each class member’s claim for back pay and punitive damages. Dukes I, 222 F.R.D. at 173-74. The district court did not provide any reason for this decision, stating only that “holding individual hearings for the number of women potentially entitled to backpay in this case is impractical on its face, and thus the traditional Teamsters mini-hearing approach is not feasible here.” Id. at 176.

It seems obvious that the district court’s determination that it could not certify the class in compliance with Teamsters compels the conclusion that it could not certify the class at all. Nevertheless, the district court determined it could bypass Supreme Court precedent. Relying on cases the court characterized as allowing back pay awards to be calculated on a class-wide basis, see Dukes I, 222 F.R.D. at 176-78, 180, 184, the district court held that once plaintiffs proved that Wal-Mart engaged in a pattern or practice of discrimination, liability as to individual class members could be determined by means of a formula for employees denied promotions, id. at 176-78, or by analyzing Wal-Mart’s personnel data to identify employees entitled to back pay, id. at 184-86. Further, the district court rejected Wal-Mart’s argument that punitive damages could not be awarded on a class-wide basis, holding that it could “limit recovery of any punitive damages to those class members who actually recover an award of lost pay,” as determined by a formula. Id. at 172.

In adopting the approach described above, the district court deprived WalMart of its statutory right to raise its defenses and violated the mandate of the Rules Enabling Act. If Wal-Mart can prove that it took an adverse employment action with respect to a particular plaintiff for non-discriminatory reasons, or that it “would have taken the same action in the absence of the impermissible motivating factor,” 42 U.S.C. § 2000e-5(g)(2)(A), (B), it cannot be held liable for back pay or punitive damages as to those individual class members. See Teamsters, 431 U.S. at 361, 97 S.Ct. 1843.

The cases relied on by the district court, see Dukes I, 222 F.R.D. at 176-78, 180, 184, do not justify its refusal to provide individual hearings, but rather provide that defendants are entitled to raise statutory defenses to liability. See Domingo, 727 F.2d at 1445; see also McKenzie v. Sawyer, 684 F.2d 62, 77-78 (D.C.Cir.1982), abrogated on other grounds by Berger v. Iron Workers Reinforced Rodmen, Local 201, 170 F.3d 1111 (D.C.Cir.1999); Hameed v. Int’l Ass’n of Bridge, Structural & Ornamental Iron Workers Local Union No. 396, 637 F.2d 506, 520 (8th Cir.1980) (“If a black applicant was rejected for a nondiscriminatory reason, that applicant would not be a member of the class of potential discriminatees.”). The district court could not properly rely on cases that preceded Teamsters, see, e.g., Stewart v. Gen. Motors Corp., 542 F.2d 445, 452-53 (7th Cir.1976), nor those from other circuits that are contrary to Teamsters, see, e.g., EEOC v. O & G Spring & Wire Forms Specialty Co., 38 F.3d 872, 879-80 & n. 9 (7th Cir.1994); Segar v. Smith, 738 F.2d 1249, 1291-92 (D.C.Cir.1984).

B

The district court’s crucial misunderstanding of Title YII and Teamsters led to a second mistake: it failed to undertake a proper analysis of whether the proposed class should be certified under Rule 23(b)(2) or Rule 23(b)(3). The majority makes the same error, with nary an acknowledgment that the Supreme Court has provided general guidelines for making this determination. See, e.g., Ortiz, 527 U.S. at 845-46, 119 S.Ct. 2295; Amchem, 521 U.S. at 625, 117 S.Ct. 2231. Again, to understand how the district court and majority went astray, it is first necessary to describe the basic analytical structure for categorizing classes under Rule 23(b).

1

Rule 23(b) establishes three different types of classes. In certifying a proposed class, courts must determine whether a proposed class fits into the historical models on which Rule 23(b)(1) or (2) were based, or is a more “adventuresome” class action, Amchem, 521 U.S. at 614, 117 S.Ct. 2231, that must be certified under Rule 23(b)(3). See Ortiz, 527 U.S. at 832-37, 119 S.Ct. 2295.

Courts considering whether to certify a class under Rule 23(b)(1) or (2) must take the “prudent course” of staying “close to the historical model” and “traditional paradigm” as understood by the Advisory Committee in drafting the rule. OHiz, 527 U.S. at 842, 864, 119 S.Ct. 2295. The Advisory Committee intended Rule 23(b)(1) classes to be “limited fund” actions, a type of action where multiple plaintiffs seek to divide a limited fund among claimants. Id. at 838-41, 119 S.Ct. 2295. Rule 23(b)(2) allows a court to certify a class action when “the party opposing the class has acted or refused to act on grounds that apply generally to the class, so that final injunctive relief or corresponding declaratory relief is appropriate respecting the class as a whole.” Fed. R.Civ.P. 23(b)(2). A review of the historical sources relied upon by Ortiz reveals that the Advisory Committee based Rule 23(b)(2) on “various actions in the civil-rights field where a party is charged with discriminating unlawfully against a class, usually one whose members are incapable of Specific enumeration.” Fed.R.Civ.P. 23(b)(2) advisory committee’s note (1966). Specifically, the Committee cited, as the historical predicate of the Rule 23(b)(2) action, cases where plaintiffs endeavored to desegregate public accommodations through prospective declaratory and injunctive relief applicable to the class as a whole. Notably, these historical actions sought the correction of a specific discriminatory policy affecting all members of the class in the same way. See, e.g., Bailey v. Patterson, 323 F.2d 201, 206 (5th Cir. 1963).

None of the cases cited by the Advisory Committee included claims for monetary relief or individual relief of any kind. Fed. R.Civ.P. 23(b)(2) advisory committee’s note.²²

The Advisory Committee Notes explain that, “[t]he subdivision does not extend to cases in which the appropriate final relief relates exclusively or predominantly to money damages.” Id. In other words, Rule 23(b)(2) was designed for classes seeking class-wide injunctive relief to remedy a common injury to the class as a whole, not for classes seeking individual damages, back pay, or other individual relief.

While Rule 23(b)(1) and (2) are linked to well-understood historical models, Rule 23(b)(3) was designed for those cases in which “class-action treatment is not as clearly called for.” Amchem, 521 U.S. at 615, 117 S.Ct. 2231 (internal quotation marks omitted); see also Ortiz, 527 U.S. at 842, 119 S.Ct. 2295(explaining that the “Advisory Committee looked cautiously at the potential for creativity” under Rule 23(b)(1) and (2) and was “not forward looking as it was in anticipating innovations under Rule 23(b)(3)”); Fed.R.Civ.P. 23(b)(3) advisory committee’s note. Specifically, Rule 23(b)(3) provides for class actions that potentially allow plaintiffs to seek monetary relief on a class-wide basis. Fed.R.Civ.P. 23(b)(3) advisory committee’s note; Benjamin Kaplan, Continuing Work of the Civil Committee: 1966 Amendments of the Federal Rules of Civil Procedure (I), 81 Harv. L.Rev. 356, 393 (1967).

Because these more “adventuresome” class actions, Amchem, 521 U.S. at 614, 117 S.Ct. 2231, raise due process concerns, the Advisory Committee imposed two safeguards on courts certifying such a class. First, a district court must “find[ ] that the questions of law or fact common to class members predominate over any questions affecting only individual members.” Fed. R.Civ.P. 23(b)(3). This inquiry ensures that deviation from “the usual rule that litigation is conducted by and on behalf of the individual named parties only” is justified. Califano v. Yamasaki, 442 U.S. 682, 700-01, 99 S.Ct. 2545, 61 L.Ed.2d 176 (1979). Second, a district court must give prospective class members the opportunity to be excluded from the class, which is commonly known as the right to “opt out.” Fed.R.Civ.P. 23(c)(2)(B)(v). This option protects “the interests of the individuals in pursuing their own litigations[, which] may be so strong here as to warrant denial of a class action altogether.” Fed. R. Civ. Pro. 23(c)(2) advisory committee’s note. The opt-out option also ensures that the class is binding on those members of the class who choose not to opt out. Eisen v. Carlisle & Jacquelin, 417 U.S. 156, 176, 94 S.Ct. 2140, 40 L.Ed.2d 732 (1974).

This right to opt out of a class action, however, is exclusive to classes certified under Rule 23(b)(3). On its face, Rule 23 does not provide absent class members the right to request exclusion in classes certified under Rule 23(b)(2), and it is clear from the Rule’s structure that the Advisory Committee did not intend for such a right to exist under Rule 23(b)(2).²³ Moreover, the Supreme Court has interpreted Rule 23(b)(2) as precluding opt outs. Ticor Title Ins. Co. v. Brown, 511 U.S. 117, 121, 114 S.Ct. 1359, 128 L.Ed.2d 33 (1994). In Ticor, the Supreme Court dismissed its grant of certiorari on the question whether binding absent class members to a class-wide settlement would violate their constitutional due process rights, where the district court certified the class under Rules 23(b)(1)(A) and (b)(2), and the absent members had not had the chance to opt out of the class. Id. at 120-21, 114 S.Ct. 1359. The Court first noted that Rule 23(b)(3) permits opt outs, and Rules 23(b)(1) and (b)(2) do not. Id. at 121, 114 S.Ct. 1359. From this, the Court reasoned it could avoid the constitutional question raised in the petition for certiorari if it held that “in actions seeking monetary damages, classes can be certified only under Rule 23(b)(3),” not Rules 23(b)(1) and (b)(2). Id. The Court concluded that this interpretation of Rule 23 “is at least a substantial possibility.” Id.²⁴

2

In certifying this class under Rule 23(b)(2), the district court failed to take the “prudent course” required by Ortiz, 527 U.S. at 842, 119 S.Ct. 2295, and consider the history and structure of Rule 23(b). Had the district court undertaken such a review, it would have been compelled to conclude that the proposed class does not fit the historical model for Rule 23(b)(2) classes, and cannot be certified in that category. First, the proposed class raises an enormous number of individual questions. Although the plaintiffs’ action is in civil rights, the plaintiffs do not ask merely for class-wide injunctive relief (as in the cases cited by the Advisory Committee); they also seek individualized relief in the form of back pay and punitive damages. As explained above, this may result in up to 1.5 million individual questions about Wal-Mart’s liability for individual relief. A class that raises so many individual questions also raises the concerns the Advisory Committee expressly considered in drafting Rule 23(b)(3), and requires the court to ask whether “the questions of law or fact common to class members predominate over any questions affecting only individual members.” Fed.R.Civ.P. 23(b)(3). Given the facts in this case, it is far from certain that the proposed class could be certified under the more flexible model provided by Rule 23(b)(3), let alone meet the narrower historical model for Rule 23(b)(2).

Second, the class’s claims for monetary relief raise the very due process concerns considered by the Advisory Committee in drafting Rule 23(b)(3). The district court acknowledged that due process required that absent class members in this case be given the right to opt out of the class. See Dukes I, 222 F.R.D. at 172-73. Because Rule 23(b)(2) does not permit opt-outs, such recognition should have led the district court to determine that certification under Rule 23(b)(2) was inappropriate, and to instead consider whether the class met the criteria for certification under Rule 23(b)(3).

Given the Supreme Court’s direction to adhere to the historical models in considering which classes can be certified under Rule 23(b)(1) and (2), see Ortiz, 527 U.S. 815, 119 S.Ct. 2295, 144 L.Ed.2d 715; Amchem, 521 U.S. 591, 117 S.Ct. 2231, 138 L.Ed.2d 689, there was no reason for the district court here to sidestep the procedural requirements of Rule 23(b)(3) to certify this class under Rule 23(b)(2). See Eubanks v. Billington, 110 F.3d 87, 93 n. 9 (D.C.Cir.1997) (noting that when notice and opt-outs are required by the district court, “the arguments supporting certification ... under subdivision (b)(2) [as opposed to (b)(3) ] are surprisingly weak”). This is particularly true given that the district court’s primary justification for certification under Rule 23(b)(2) was that the suit was similar to the historical model contemplated by the Advisory Committee. See Dukes, 222 F.R.D. at 170. In certifying this class under Rule 23(b)(2), the district court abused its discretion.

C

In holding that the district court did not err in certifying the class under Rule 23(b)(2), the majority focuses not on the text, structure, and history of this rule, but rather on a single sentence in the Advisory Committee Notes which states that Rule 23(b)(2) “does not extend to cases in which the appropriate final relief relates exclusively or predominantly to money damages.” Maj. Op. at 616 (quoting Fed. R.Civ.P. 23(b)(2) advisory committee’s notes). The majority infers from this single sentence that Rule 23(b)(2) must therefore extend to cases in which the appropriate final relief does not relate “exclusively or predominantly to money damages.” Maj. Op. at 616. The majority then turns to a dictionary definition of the word “predominant” and derives a new test: “a class must seek only monetary damages that are not ‘superiorjjn] strength, influence, or authority’ to injunctive and declaratory relief.’ ” Maj. Op. at 616 (citing Merriam-Webster’s Collegiate Dictionary 978 (11th ed.2004)).²⁵ In furtherance of this test, the majority directs district courts to consider “the objective 'effect of the relief sought’ on the litigation,” by recourse to “[f|actors such as [1] whether the monetary relief sought determines the key procedures that will be used, [2] whether it introduces new and significant legal and factual issues, [3] whether it requires individualized hearings, and [4] whether its size and nature ... raise particular due process and manageability concerns.” Maj. Op. at 617.²⁶

But in upholding the district court’s certification with regard to back pay, the majority fails to apply these factors from its own test. Had it done so, it would have concluded that the district court erred in certifying the class under Rule 23(b)(2). First, the plaintiffs’ claim for back pay triggers the need for a key procedure (factor 1), namely the requirement for individualized hearings (factor 3). Second, the plaintiffs’ claims for individual relief introduced significant new legal and factual issues into the case (factor 2), including Wal-Mart’s defenses to liability, as well as due process concerns that necessitate opt-outs (factor 4). Third, the need for individualized hearings raises serious manageability concerns (factor 4).

The majority avoids these issues, however, and dismisses Wal-Mart’s arguments as mere objections to the district court’s “trial plan.” Maj. Op. at 624. The majority “express[es] no opinion regarding WalMart’s objections to the district court’s tentative trial plan” and concludes that there are many steps the district court could take “that would allow this class action to proceed in a manner that is both manageable and in accordance with due process,” Maj. Op. at 625 (citing Hilao v. Estate of Marcos, 103 F.3d 767, 782-87 (9th Cir.1996))²⁷

The majority’s failure to address these objections runs contrary to the Supreme Court’s mandate, Ortiz, 527 U.S. at 861, 119 S.Ct. 2295, and the majority’s own rule, Maj. Op. at 6217(stating that courts must consider “whether [the proposed class action] requires individualized hearings” when it determines whether a class may be certified under Rule 23(b)(2)).

Presumably because the application of its own test would lead to the conclusion that the district court erred, the majority does not apply its test (or remand back to the district court to apply it, as it did with the punitive damages class). Instead, the majority determines that the back-pay remedy does not predominate over the claims for injunctive and declaratory relief for two reasons: (1) the calculation of back pay in Title VII cases “generally involves [relatively uncomplicated factual determinations and few[] individualized issues,” and (2) back pay is a “ ‘make whole’ remedial scheme, a scheme to which the drafters of the Federal Rules of Civil Procedure clearly intended Rule 23(b)(2) to apply.” Maj. Op. at 619 (alterations in original) (citations omitted). As noted above, neither of these rationales is correct. In this case, back pay cannot be calculated until after Wal-Mart has had an opportunity to raise its defenses, which will involve the litigation of a multitude of individual issues. Moreover, while' the drafters of Rule 23 intended that classes involving demands for injunctive relief be certified under Rule 23(b)(2), the majority provides no evidence that the Advisory Committee thought Rule 23(b)(2) was appropriate for classes seeking back pay.²⁸

As to claims for punitive damages, the majority remands the case to the district court to determine under its new test whether the case could be certified under Rule 23(b)(2). Maj. Op at 622. Before doing so, however, it determines in a single sentence that punitive damages do not require individualized determinations because the plaintiffs allege that Wal-Mart’s policy “affects all class members in a similar way.” Maj. Op. at 622. This unprecedented holding, made with virtually no analysis, is wrong both as a matter of law and fact. As noted above, it conflicts with the statutory language of Title VII, see 42 U.S.C. § 2000e-5(g)(2)(B), and with the Supreme Court’s reasoning regarding punitive damages, see BMW, 517 U.S. at 581, 116 S.Ct. 1589. Moreover, it is factually wrong: even assuming there was significant proof of a company-wide policy of discrimination, plaintiffs do not allege that Wal-Mart’s alleged policy affected all class members in the same way for purposes of determining monetary relief. Plaintiffs acknowledge that Wal-Mart’s allegedly discriminatory practices have affected class members differently, stating that “different class members may have been under-compensated in different amounts; some employees may have been denied promotion to assistant manager while others may have been denied promotion to store manager.” The district court cannot reasonably determine whether an individual class member was injured, let alone the nature and amount of injury, until the second phase of individual hearings has been accomplished. Other circuits addressing the issue have agreed with this analysis. See, e.g., Reeb, 435 F.3d at 651; Lemon, 216 F.3d at 581; Allison, 151 F.3d at 417. By untethering punitive damage claims in a Title VII case from evidence regarding how the plaintiffs were actually injured, the majority opens the door to results that would be unfair to both claimants and defendants in future Title VII class actions.

IV

The class action device is a procedural mechanism to aggregate individual claims for purposes of judicial efficiency. Falcon, 457 U.S. at 159, 102 S.Ct. 2364. By allowing claims to go forward that ordinarily would not be litigated, the class action has proven to be an invaluable tool in litigating civil rights and other cases involving systemic discrimination. But, because Rule 23 creates a procedural device that constitutes “an exception to the usual rule that litigation is conducted by and on behalf of the individual named parties only,” Califano, 442 U.S. at 700-01, 99 S.Ct. 2545, courts must use it carefully, ensuring that the proposed class action complies with controlling law and does not jeopardize the parties’ substantive rights. See Gulf Oil Co. v. Bernard, 452 U.S. 89, 100, 101 S.Ct. 2193, 68 L.Ed.2d 693 (1981) (holding that, in considering a proposal to certify a class, the district court’s discretion is “bounded by the relevant provisions of the Federal Rules”). The “[m]ere invocation of the language of Rule 23 in Title VII suits is no mystical legal talisman guaranteeing class treatment.” Doninger v. Pac. Nw. Bell, Inc., 564 F.2d 1304, 1312 (9th Cir.1977).

The district court here failed in its duty to apply Rule 23 carefully and correctly. In its enthusiasm for allowing the class action to proceed, the district court sped forward, dodging substantive and procedural hurdles, and making an end-run around Title VII, the Rules Enabling Act, and Supreme Court precedent. It conducted no rigorous analysis to determine “the existence of a class of persons who have suffered the same injury” as the class representatives. Falcon, 457 U.S. at 157, 102 S.Ct. 2364. It certified the class even after determining that compliance with the Supreme Court’s framework in Teamsters was infeasible. Finally, the district court failed to consider the text, structure, and historical framework of Rule 23(b) that should have guided its analysis. The resultant class certification runs directly contrary to controlling precedent.

Despite these flaws, the majority affirms those legal errors with even less analysis and inquiry than the district court. Never before has such a low bar been set for certifying such a gargantuan class. The majority’s ruling provides scant limits to the types of classes that can be certified. Put simply, the door is now open to Title VII lawsuits targeting national and international companies, regardless of size and diversity, based on nothing more than general and conclusory allegations, a handful of anecdotes, and statistical disparities that bear little relation to the alleged discriminatory decisions. The district court abused its discretion in certifying this class. The majority errs in concluding otherwise. While class actions have a vital role to play in the battle against discrimination, the majority’s decision to allow this case to go forward epitomizes the Supreme Court’s warning that “the rulemakers’ prescriptions for class actions may be endangered by those who embrace Rule 23 too enthusiastically just as they are by those who approach the Rule with distaste.” Amchem, 521 U.S. at 629, 117 S.Ct. 2231 (internal citations, quotation marks, and alterations omitted). I respectfully dissent.

1 . Sam’s Club is a separate corporate division within Wal-Mart, but operates in roughly the same manner and with a similar organizational structure as Wal-Mart.

2 . Edith Arana was terminated in October 2001 for falsifying her time sheets.

3 . The complaint was filed in June 2001. Patricia Surgeson left Wal-Mart in March 2001 to take another job, Cleo Page resigned in November 2000 after she received a written warning for performance issues, and Deborah Gunter resigned in August 1999 after WalMart reduced her hours.

4 . For example, proposed class representative Page alleges that she was passed over for promotion by her store manager, Monique Taylor. Proposed class representative Gunter complained about being passed over for a promotion by her store manager, Kathy Bishop, who Gunter also alleges brushed off allegations Gunter made about sexual harassment. Proposed class representative Kwapnoski discussed her interest in promotion with her assistant manager, Nancy Horn. Taylor, Bishop, and Horn are all members of the proposed class. This case features the unusual distinction of placing victims and their alleged victimizers on the same side of the counsel table.

5 . The majority criticizes the dissent for pointing out the large size of the class, reasoning that the class size may decrease by up to two-thirds if past employees are excluded. See Maj. Op. at 578 n. 3. In determining whether the district court abused its discretion, however, we must consider the class actually certified by the district court, which included past employees. Dukes I, 222 F.R.D. at 188. Moreover, as explained below, the reasons this class cannot be certified apply with equal force regardless of whether the class represents 1.5 million individuals or the class of 500,000 envisioned by the majority.

6 . Rule 23(a) of the Federal Rules of Civil Procedure provides that "[o]ne or more members of a class may sue or be sued as representative parties on behalf of all members only if” the class meets the prerequisites of numerosity, commonality, typicality, and adequacy of representation.

7 . The term "commonality” refers to the requirement in Rule 23(a) that "there are questions of law or fact common to the class”; "typicality” refers to the requirement that "the claims or defenses of the representative parties are typical of the claims or defenses of the class.” Fed.R.Civ.P. 23(a). Falcon noted that "[t]he commonality and typicality requirements of Rule 23(a) tend to merge. Both serve as guideposts for determining whether under the particular circumstances maintenance of a class action is economical and whether the named plaintiff's claim and the class claims are so interrelated that the interests of the class members will be fairly and adequately protected in their absence.” 457 U.S. at 158 n. 13, 102 S.Ct. 2364.

8 . To support this point, the majority cites to a dissent from the denial of rehearing en banc, Lopez-Rodriguez v. Holder, 560 F.3d 1098 (9th Cir.2009), which provides no support. See Maj. Op. 595 n. 15. Lopez-Rodriguez merely stated that the Supreme Court's language should not be applied "to create a new rule— one [the Court] never envisioned.” Maj. Op. at 595 n. 15 (quoting Lopez-Rodriguez, 560 F.3d at 1099). This truism does not support the majority's position that we have the option to ignore the Supreme Court's guidance on an issue directly before the Court, as in Falcon.

9 . As pointed out by the majority, the plaintiffs also produced "facts supporting the existence of company-wide policies and practices.” Maj. Op. at 600. Because it is undisputed that Wal-Mart maintains uniform company-wide policies, and because the mere existence of company-wide policies says nothing about whether such policies are discriminatory, there is no need to discuss such "facts” separately from plaintiffs’ other purported evidence.

10 . The number of Wal-Mart stores and managers set forth in this paragraph are based on information provided by plaintiffs' expert, Dr. Marc Bendick, and are current as of 1999.

11 . Until today, this circuit's largest stretch to uphold certification in an employment discrimination class action based on a small number of incidents of discrimination occurred in Staton, where 237 individuals sought to represent a class of 15,000 African American employees. 327 F.3d at 948 n. 4, 953. Cautioning that this was an "ambitious” claim for commonality that was "especially worthy of scrutiny,” id. at 954, we nevertheless upheld the district court's class certification. In doing so, we noted that plaintiffs' lawyers interviewed more than 1,300 employees from facilities across the country, filed detailed documentation of discrimination experienced by 200 of those employees, and produced a company-wide internal survey to support their claims of class-wide racial discrimination. Id. Accordingly, at a minimum, the plaintiffs presented anecdotal evidence from one out of every 75 class members. No court, however, has ever held that one anecdote could suffice to represent the experiences of as vast a number as 12,-500 class members.

12 . In fact, Haworth pointed out that the statistical disparities between men and women at the regional level could also be the result of the aggregation of the data itself. This problem is known as “Simpson's Paradox,” which refers to “illusory disparities in improperly aggregated data that disappear when the data are disaggregated.” Eng’g Contractors Ass'n of S. Fla. Inc. v. Met. Dade County, 122 F.3d 895, 919 n. 4 (11th Cir. 1997); see also P.J. Bickel, E.A. Hammel, & J.W. O’Connell, Sex Bias in Graduate Admissions: Data from Berkeley, 187 Science 398 (1975) (using Simpson's Paradox to explain how researchers found a statistically significant bias in favor of male applicants when analyzing admissions data from all 101 of Berkeley's graduate departments, despite the fact that only 4 departments had a statistically significant bias in favor of males and 6 departments had a statistically significant bias in favor of females). As the article cited by the majority shows, it is especially likely that Drogin's statistics would display illusory disparities, because Drogin based his study on pre-existing information that was collected from a large sample size. See Marios G. Pavlides & Michael D. Perlman, How Likely Is a Simpson’s Paradox, 63 Am. Statistician 226, 230 (2009).

13 . The district court also relied on a study prepared by Dr. Marc Bendick, Ph.D., an economist and consultant, which made a company-wide comparison of Wal-Mart with twenty purportedly similar (but much smaller) retail companies. The study concluded that on a company-wide basis WalMart promoted a smaller percentage of women than its competitors. Again, the district court applied the wrong legal standard. Instead of conducting a rigorous evaluation of the evidence as required by Falcon, it accepted Bendick’s study at face value because it held it was "sufficiently probative to assist the Court in evaluating the class certification requirements at issue in this case.” Dukes v. Wal-Mart, Inc. (Dukes II), 222 F.R.D. 189, 195 (N.D.Cal.2004). In fact, Bendick’s study did not provide support for plaintiffs' claim that the 1.5 million women in the class were subject to discriminatory treatment in promotions, because it too failed to provide information derived from the level at which promotion and pay decisions were actually made.

14 . Federal Rule of Evidence 702 states that:

15 . The majority cites to Melissa Hart & Paul M. Secunda, A Matter of Context: Social Framework Evidence in Employment Discrimination Class Actions, 78 Fordham L.Rev. 37 (2009), for a description of Bielby’s "social framework analysis.” Maj. Op. at 601 n. 21. In employment discrimination litigation, social framework experts purport to summarize social science research regarding the workplace to give the finder of fact a "context” in which to decide key legal issues. Hart & Secunda, supra, at 44. The article by Hart and Secunda discusses the dispute among social psychologists over whether it is appropriate for experts like Bielby to use social framework analysis to draw specific factual conclusions about a company's operations without the use of case-specific research. See id. at 51-55. The controversy over the reliability of this methodology highlights the need for a proper Daubert inquiry here.

16 . The majority writes that it is “not convinced by the dissent’s argument that Daubert has exactly the same application at the class certification stage as it does to expert testimony relevant at trial.” Maj. Op. at 602 n. 22. But the majority never accounts for why it is "not convinced,” nor does it explain why the district court can rely on an expert's testimony that is not reliable, at the class certification stage or any other. Rule 702 and the Supreme Court make clear that evidence that is not scientifically valid does not " 'assist the trier of fact to understand the evidence or to determine a fact in issue.’ ” Daubert, 509 U.S. at 589, 113 S.Ct. 2786 (quoting Fed. R.Evid. 702).

17 . The majority errs in suggesting that the district court performed a proper Daubert inquiry, see Maj. Op. at 602-03 n.22, because it stated that Bielby’s opinion "is based on valid principles.” Dukes II, 222 F.R.D. at 192. Whether Bielby based his testimony on valid principles says nothing about whether he applied those principles via a valid methodology to produce relevant and reliable evidence, which is the critical issue under Daubert,

18 . 28 U.S.C. § 2072(b) provides:

(b) Such rules [of practice and procedure prescribed by the Supreme Court] shall not abridge, enlarge or modify any substantive right. All laws in conflict with such rules shall be of no further force or effect after such rules have taken effect.

19 . Teamsters involved a suit by the Attorney General under 42 U.S.C. § 2000e-6, which allows the government to bring an action when it has cause to believe that a person "is engaged in a pattern or practice of resistance to the full enjoyment of any of the rights” secured by Title VII. 431 U.S. at 328 n. 1, 97 S.Ct. 1843. Teamsters's two-step framework is equally applicable to class actions brought by individuals alleging discrimination. See id. at 358-59, 97 S.Ct. 1843 (noting that it was adopting the two-step framework set forth in Franks, 424 U.S. 747, 96 S.Ct. 1251, 47 L.Ed.2d 444, a class action brought by individuals alleging discrimination); see also Cooper, 467 U.S. at 876 n. 9, 104 S.Ct. 2794 (applying the Teamsters pattern-or-practice framework to class actions brought by individuals).

20 . While Teamsters referred to the second stage of individualized hearings as the remedial” stage of trial, 431 U.S. at 361, 97 S.Ct. 1843, this is a misnomer. As explained by Justice O'Connor in her concurrence in Price Waterhouse v. Hopkins, 490 U.S. 228, 109 S.Ct. 1775, 104 L.Ed.2d 268 (1989), superceded in part by statute, Civil Rights Act of 1991, the second stage of individualized hearings is not remedial, but rather a second phase of liability, in which the employer is entitled to raise individual statutory defenses: ''It is misleading to speak of the additional proof required by an individual class member for relief as being a part of the damage phase, that evidence is actually an element of the liability portion of the case.” Id. at 266, 109 S.Ct. 1775 (O’Connor, J., concurring) (quoting Dillon v. Coles, 746 F.2d 998, 1004 (3d Cir. 1984)). We have explained the two-step procedure in a similar way. See EEOC v. Gen. Tel. Co. of Nw., Inc., 599 F.2d 322, 332 (9th Cir. 1979) (holding that the demonstration of a discriminatory pattern and practice in the first phase of the case establishes the individual class members’ prima facie case of discrimination, and shifts the burden to the employer "to prove that individuals were not in fact victims of previous discrimination”); see also Domingo v. New England Fish Co., 727 F.2d 1429, 1445 (9th Cir.1984) (holding that once class-wide discrimination has been shown, a claimant is presumptively eligible for back pay, subject to the employer "proving that the applicant was unqualified or showing some other valid reason why the claimant was not, or would not have been, acceptable”). Other circuits are in accord. See, e.g., Robinson v. Metro-North Commuter R.R. Co., 267 F.3d 147, 159 (2d Cir.2001); Allison v. Citgo Petroleum Corp., 151 F.3d 402, 417-18 (5th Cir. 1998); Dillon, 746 F.2d at 1004; Craik v. Minn. State Univ. Bd., 731 F.2d 465, 469-70 (8th Cir.1984); Mitchell v. Mid-Continent Spring Co. of Ky., 583 F.2d 275, 284 (6th Cir. 1978).

21 . The majority accuses the dissent of "ignor[ing] that the pattern and practice has to be proven on a group basis.” Maj. Op. at 625 n. 53 (emphasis in original). On the contrary, there is no doubt that a plaintiff must establish a pattern and practice of discrimination on a company-wide basis in the first phase of their case, as explained above. It is the majority that turns a blind eye to the second phase of Title VII litigation, where defendants have the right to raise affirmative defenses as to each class member. The majority never satisfactorily answers the question of how the district court would conduct up to 1.5 million individualized hearings during the second phase that Teamsters mandates.

22 . The Advisory Committee Notes cites the following cases, none of which involved monetary relief: Potts v. Flax, 313 F.2d 284 (5th Cir. 1963) (class action by two African-American parents against a “system-wide policy of racial segregation” in city schools); Bailey, 323 F.2d 201 (class action against laws segregating common carriers); Northcross v. Bd. of Ed., 302 F.2d 818 (6th Cir.1962) (class action by 18 African-American students and their parents seeking an order that the city end its biracial school system); Brunson v. Bd. of Trs. of Sch. Dist. No. 1, 311 F.2d 107 (4th Cir. 1962) (class action for school desegregation brought by 42 African-American students); Green v. Sch. Bd., 304 F.2d 118 (4th Cir.1962) (non-class action by 28 African American students seeking transfer to white schools and an injunction against school segregation); Mannings v. Bd. of Pub. Inst., 277 F.2d 370 (5th Cir. 1960) (class of African-American students seeking an injunction against school segregation); Orleans Parish Sch. Bd. v. Bush, 242 F.2d 156 (5th Cir. 1957) (class action on behalf of African-American school children to desegregate schools); Frasier v. Bd. of Trs. of Univ. of N.C., 134 F.Supp. 589 (M.D.N.C. 1955) (3-judge court) (class action by three African-Americans challenging a university’s discriminatory admissions policies), aff’d, 350 U.S. 979, 76 S.Ct. 467, 100 L.Ed. 848 (1956).

23 . For example, while Rule 23 provides that a court must give members of Rule 23(b)(3) classes notice and the right to opt out of the class, and may give notice to members of Rule 23(b)(2) classes, the Rule is silent on whether a court may give members of Rule 23(b)(2) classes the right to opt out. See Fed. R. Civ. Proc. 23(c)(2) and (d). The inference is that a court may not allow members of a Rule 23(b)(2) class to opt out of the class.

24 . To avoid the import of Ticor, the majority once again claims it is not bound by an order in which six Supreme Court justices have joined, characterizing Ticor as "not even dictum, let alone a holding.” Maj. Op. at 621 n.44. Yet, the majority cites no authority for this position, perhaps because the Ninth Circuit has relied upon such orders in prior opinions. See, e.g., Carroll v. Nakatani, 342 F.3d 934, 946-47 (9th Cir.2003) (relying on an order dismissing a writ as improvidently granted). Indeed, many of the cases relied upon by the majority cite to Ticor. E.g., Reeb, 435 F.3d at 646; Coleman v. Gen. Motors Acceptance Corp., 296 F.3d 443, 447 (6th Cir. 2002); Allison, 151 F.3d at 411 n. 3.

25 . The majority admits that it breaks with the Second Circuit in rejecting the subjective intent test, as well as with the Fifth, Sixth, Seventh, and Eleventh Circuits in rejecting the “incidental damages standard” test. Maj. Op. at 616. In creating this three-way circuit split, the majority not only ignores our own precedent, see Zimmerman v. State Dep't of Justice, 170 F.3d 1169, 1184 (9th Cir. 1999) (we create a circuit split “only after the most painstaking inquiry”), but aggravates the already-existing inconsistency between the circuits.

26 . As formulated by the majority, this test is essentially unusable. When is back pay “superior in strength” to an injunction? When do punitive damages have more "influence” than declaratory relief? It is unlikely that this test, particularly as applied by the majority, provides the district court with the guidance it needs on remand.

27 . Quite apart from the question whether Wal-Mart is objecting to the certification of the class (as it claims) or to the trial plan (as the majority states), the procedure set forth in Hilao cannot be used in a Title VII case and the majority errs in suggesting it can. We held in Hilao that a court could determine compensatory damages for over 10,000 claimants in a human rights action by means of a formula without violating the due process rights of the defendants. See Hilao, 103 F.3d at 782-87; but see id. at 788 (Rymer, J. dissenting) ("If due process in the form of a real prove-up of causation and damages cannot be accomplished because the class is too big or to do so would take too long, then (as the Estate contends) the class is unmanageable and should not have been certified in the first place.”); Cimino v. Raymark Indus., Inc., 151 F.3d 297, 319 (5th Cir.1998) (refusing to apply Hilao and noting that "we find ourselves in agreement with the thrust of the dissenting opinion there”). But Hilao has no bearing on this effort to certify a Title VII class action, because Title VII itself, and the Supreme Court’s interpretation of Title VII pattern-or-practice actions, see Teamsters, 431 U.S. at 360, 97 S.Ct. 1843, give defendants their right to a defense to each individual’s claim. See 42 U.S.C. § 2000e-5(g)(2).

28 . The majority claims it need not consider the problem posed by the class's pursuit of individual monetary relief because it is merely "joining] the consensus view that a request for back pay in a Title VII case is fully consistent with the certification of a Rule 23(b)(2) class action,” citing as support Thorn v. Jefferson-Pilot Life Insurance Co., 445 F.3d 311, 331-32 (4th Cir.2006) and Coleman, 296 F.3d 443. Maj. Op. at 619-20 & n. 41. Again, the majority misses the point. The issue is not that plaintiffs seek back pay; the issue is WalMart’s statutory right to raise individual defenses in response to the request for back pay, and whether such individualized treatment makes class-wide relief improper. Thom and Coleman reinforce this point. In Coleman, the Sixth Circuit held that the district court abused its discretion in certifying a class that included a claim for compensatory damages, in large part because such relief would require individual determinations as to each class member, thus obviating the efficiencies of pursuing the action on a class-wide basis. 296 F.3d at 449. Likewise, in Thom, the Fourth Circuit upheld the district court’s denial of class certification, in part because the availability of a statute of limitations defense presented individual issues that could not be determined on a class-wide basis. 445 F.3d at 32.7. Thom distinguished prior cases that certified classes requiring the calculation of back pay on the ground that such calculations "generally involve[d] ... fewer individualized issues.” Id. at 331 (alteration and internal quotation marks omitted). When considered as a whole (rather than taking a few phrases out of context), Thom and Coleman support the conclusion that class-wide relief is inappropriate where claims may need be litigated on an individual basis.

---

Dissenting Opinion

KOZINSKI, Chief Judge,

dissenting:

Maybe there’d be no difference between 500 employees and 500,000 employees if they all had similar jobs, worked at the same half-billion square foot store and were supervised by the same managers. But the half-million members of the majority’s approved class held a multitude of jobs, at different levels of Wal-Mart’s hierarchy, for variable lengths of time, in 3,400 stores, sprinkled across 50 states, with a kaleidoscope of supervisors (male and female), subject to a variety of regional policies that all differed depending on each class member’s job, location and period of employment. Some thrived while others did poorly. They have little in common but their sex and this lawsuit.

I therefore join fully Judge Ikuta’s dissent.

Case Details

Case Name: Dukes v. Wal-Mart Stores, Inc.

Court Name: Court of Appeals for the Ninth Circuit

Date Published: Apr 26, 2010

Citation: 603 F.3d 571

Docket Number: Nos. 04-16688, 04-16720

Court Abbreviation: 9th Cir.