Olean Wholesale Grocery Co-Op v. Bumble Bee Foods LLC

Opinion

FOR PUBLICATION

UNITED STATES COURT OF APPEALS FOR THE NINTH CIRCUIT

OLEAN WHOLESALE GROCERY No. 19-56514 COOPERATIVE, INC.; BEVERLY YOUNGBLOOD; PACIFIC GROSERVICE, D.C. No. INC., DBA Pitco Foods; CAPITOL 3:15-md-02670- HILL SUPERMARKET; LOUISE ANN JLS-MDD DAVIS MATTHEWS; JAMES WALNUM; COLIN MOORE; JENNIFER A. NELSON; ELIZABETH DAVIS-BERG; LAURA OPINION CHILDS; NANCY STILLER; BONNIE VANDERLAAN; KRISTIN MILLICAN; TREPCO IMPORTS AND DISTRIBUTION, LTD.; JINKYOUNG MOON; COREY NORRIS; CLARISSA SIMON; AMBER SARTORI; NIGEL WARREN; AMY JOSEPH; MICHAEL JUETTEN; CARLA LOWN; TRUYEN TON-VUONG, AKA David Ton; A-1 DINER; DWAYNE KENNEDY; RICK MUSGRAVE; DUTCH VILLAGE RESTAURANT; LISA BURR; LARRY DEMONACO; MICHAEL BUFF; ELLEN PINTO; ROBBY REED; BLAIR HYSNI; DENNIS YELVINGTON; KATHY DURAND GORE; THOMAS E. WILLOUGHBY III; ROBERT FRAGOSO; SAMUEL SEIDENBURG; JANELLE ALBARELLO; MICHAEL COFFEY; JASON WILSON; JADE CANTERBURY; NAY ALIDAD; GALYNA

ANDRUSYSHYN; ROBERT BENJAMIN; BARBARA BUENNING; DANIELLE GREENBERG; SHERYL HALEY; LISA HALL; TYA HUGHES; MARISSA JACOBUS; GABRIELLE KURDT; ERICA PRUESS; SETH SALENGER; HAROLD STAFFORD; CARL LESHER; SARAH METIVIER SCHADT; GREG STEARNS; KARREN FABIAN; MELISSA BOWMAN; VIVEK DRAVID; JODY COOPER; DANIELLE JOHNSON; HERBERT H. KLIEGERMAN; BETH MILLINER; LIZA MILLINER; JEFFREY POTVIN; STEPHANIE GIPSON; BARBARA LYBARGER; SCOTT A. CALDWELL; RAMON RUIZ; THYME CAFE & MARKET, INC.; HARVESTERS ENTERPRISES, LLC; AFFILIATED FOODS, INC.; PIGGLY WIGGLY ALABAMA DISTRIBUTING CO., INC.; ELIZABETH TWITCHELL; TINA GRANT; JOHN TRENT; BRIAN LEVY; LOUISE ADAMS; MARC BLUMSTEIN; JESSICA BREITBACH; SALLY CRNKOVICH; PAUL BERGER; STERLING KING; EVELYN OLIVE; BARBARA BLUMSTEIN; MARY HUDSON; DIANA MEY; ASSOCIATED GROCERS OF NEW ENGLAND, INC.; NORTH CENTRAL DISTRIBUTORS, LLC; CASHWA DISTRIBUTING CO. OF KEARNEY, INC.; URM STORES, INC.; WESTERN FAMILY FOODS, INC.; ASSOCIATED FOOD STORES, INC.; OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 3

GIANT EAGLE, INC.; MCLANE COMPANY, INC.; MEADOWBROOK MEAT COMPANY, INC.; ASSOCIATED GROCERS, INC.; BILO HOLDING, LLC; WINNDIXIE STORES, INC.; JANEY MACHIN; DEBRA L. DAMSKE; KEN DUNLAP; BARBARA E. OLSON; JOHN PEYCHAL; VIRGINIA RAKIPI; ADAM BUEHRENS; CASEY CHRISTENSEN; SCOTT DENNIS; BRIAN DEPPERSCHMIDT; AMY E. WATERMAN; CENTRAL GROCERS, INC.; ASSOCIATED GROCERS OF FLORIDA, INC.; BENJAMIN FOODS LLC; ALBERTSONS COMPANIES LLC; H.E. BUTT GROCERY COMPANY; HYVEE, INC.; THE KROGER CO.; LESGO PERSONAL CHEF LLC; KATHY VANGEMERT; EDY YEE; SUNDE DANIELS; CHRISTOPHER TODD; PUBLIX SUPER MARKETS, INC.; WAKEFERN FOOD CORP.; ROBERT SKAFF; WEGMANS FOOD MARKETS, INC.; JULIE WIESE; MEIJER DISTRIBUTION, INC.; DANIEL ZWIRLEIN; MEIJER, INC.; SUPERVALU INC.; JOHN GROSS & COMPANY; SUPER STORE INDUSTRIES; W. LEE FLOWERS & CO. INC.; FAMILY DOLLAR SERVICES, LLC; AMY JACKSON; FAMILY DOLLAR STORES, INC.; KATHERINE MCMAHON; DOLLAR TREE DISTRIBUTION, INC.; JONATHAN RIZZO; GREENBRIER

INTERNATIONAL, INC.; JOELYNA A. SAN AGUSTIN; ALEX LEE, INC.; REBECCA LEE SIMOENS; BIG Y FOODS, INC.; DAVID TON; KVAT FOOD STORES, INC., DBA FOOD CITY; AFFILIATED FOODS MIDWEST COOPERATIVE, INC.; MERCHANTS DISTRIBUTORS, LLC; BROOKSHIRE BROTHERS, INC.; SCHNUCK MARKETS, INC.; BROOKSHIRE GROCERY COMPANY; KMART CORPORATION; CERTCO, INC.; RUSHIN GOLD, LLC, DBA The Gold Rush; UNIFIED GROCERS, INC.; TARGET CORPORATION; SIMON- HINDI, LLC; FAREWAY STORES, INC.; MORAN FOODS, LLC, DBA Save-A- Lot; WOODMAN’S FOOD MARKET, INC.; DOLLAR GENERAL CORPORATION; SAM’S EAST, INC.; DOLGENCORP, LLC; SAM’S WEST, INC.; KRASDALE FOODS, INC.; WALMART STORES EAST, LLC; CVS PHARMACY, INC.; WALMART STORES EAST, LP; BASHAS’ INC.; WAL-MART STORES TEXAS, LLC; MARC GLASSMAN, INC.; WAL-MART STORES, INC.; 99 CENTS ONLY STORES; JESSICA BARTLING; AHOLD U.S.A., INC.; GAY BIRNBAUM; DELHAIZE AMERICA, LLC; SALLY BREDBERG; ASSOCIATED WHOLESALE GROCERS, INC.; KIM CRAIG; MAQUOKETA CARE CENTER; OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 5

GLORIA EMERY; ERBERT & GERBERT’S, INC.; ANA GABRIELA FELIX GARCIA; JANET MACHEN; JOHN FRICK; PAINTED PLATE CATERING; KATHLEEN GARNER; ROBERT ETTEN; ANDREW GORMAN; GROUCHO’S DELI OF FIVE POINTS, LLC; EDGARDO GUTIERREZ; GROUCHO’S DELI OF RALEIGH; ZENDA JOHNSTON; SANDEE’S CATERING; STEVEN KRATKY; CONFETTI’S ICE CREAM SHOPPE; KATHY LINGNOFSKI; END PAYER PLAINTIFFS; LAURA MONTOYA; KIRSTEN PECK; JOHN PELS; VALERIE PETERS; ELIZABETH PERRON; AUDRA RICKMAN; ERICA C. RODRIGUEZ, Plaintiffs-Appellees,

and

JESSICA DECKER; JOSEPH A. LANGSTON; SANDRA POWERS; GRAND SUPERCENTER, INC.; THE CHEROKEE NATION; US FOODS, INC.; SYSCO CORPORATION; GLADYS, LLC; SPARTANNASH COMPANY; BRYAN ANTHONY REO, Plaintiffs,

v.

BUMBLE BEE FOODS LLC; TRI- UNION SEAFOODS, LLC, DBA

Chicken of the Sea International, DBA Thai Union Group PCL, DBA Thai Union North America, Inc.; STARKIST CO.; DONGWON INDUSTRIES CO., LTD.; THAI UNION GROUP PCL, Defendants-Appellants,

and

KING OSCAR, INC.; THAI UNION FROZEN PRODUCTS PCL; DEL MONTE FOODS COMPANY; TRI MARINE INTERNATIONAL, INC.; DONGWON ENTERPRISES; DEL MONTE CORP.; CHRISTOPHER D. LISCHEWSKI; LION CAPITAL (AMERICAS), INC.; BIG CATCH CAYMAN LP, AKA Lion/Big Catch Cayman LP; FRANCIS T. ENTERPRISES; GLOWFISCH HOSPITALITY; THAI UNION NORTH AMERICA, INC., Defendants.

Appeal from the United States District Court for the Southern District of California Janis L. Sammartino, District Judge, Presiding

Argued and Submitted October 9, 2020 Pasadena, California

Filed April 6, 2021 OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 7

Before: Andrew J. Kleinfeld, Andrew D. Hurwitz, and Patrick J. Bumatay, Circuit Judges.

Opinion by Judge Bumatay; Partial Concurrence and Partial Dissent by Judge Hurwitz

SUMMARY *

Class Certification

The panel vacated the district court’s order certifying three classes in a multi-district antitrust case alleging a price- fixing conspiracy by producers of packaged tuna.

The panel held that statistical or “representative” evidence, finding classwide impact based on averaging assumptions and pooled transaction data, can be used to establish the “predominance” requirement of Fed. R. Civ. P. 23(b)(3), under which a putative class must establish that “the questions of law or fact common to class members predominate over any questions affecting only individual members.” Agreeing with other circuits, the panel held that a district court must find by a preponderance of the evidence that the plaintiff has established predominance under Rule 23(b)(3). The panel concluded that plaintiffs’ representative evidence could be used to establish predominance because plaintiffs’ evidence could have been used to establish liability in a class member’s individual suit by demonstrating the antitrust impact of their price-fixing claims; the

* This summary constitutes no part of the opinion of the court. It has been prepared by court staff for the convenience of the reader.

representative evidence sufficiently linked plaintiffs’ injuries to their theory of antitrust violation; and plaintiffs’ use of averaging assumptions in their regression models did not defeat predominance.

The panel nonetheless concluded that the district court abused its discretion by not resolving the factual disputes necessary to decide the predominance requirement before certifying the classes. Accordingly, the panel vacated the district court’s order and remanded for the court to determine the number of uninjured parties in the proposed class based on the dueling statistical evidence, and only then to rule on whether predominance has been established.

Concurring in part and dissenting in part, Judge Hurwitz agreed with the majority’s conclusions that the district court, not the jury, must resolve factual disputes bearing on predominance; that a district court’s “rigorous analysis” of whether a putative class has satisfied Rule 23’s requirements should proceed by a preponderance of the evidence standard; and that the district court must conclude not that common issues could predominate at trial, but that they do predominate before certifying the class. Judge Hurwitz disagreed with the majority’s conclusion that, before certifying a class, the district court must find that only a de minimis number of class members are uninjured.

COUNSEL

Gregory G. Garre (argued) and Samir Deger-Sen, Latham & Watkins LLP, Washington, D.C.; Christopher S. Yates, Belinda S. Lee, and Ashley M. Bauer, Latham & Watkins LLP, San Francisco, California; John Roberti, Allen & Overy LLP, Washington, D.C.; Kenneth A. Gallo, Paul OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 9

Weiss Rifkind Wharton & Garrison LLP, Washington, D.C., for Defendants-Appellants.

Christopher L. Lebsock (argued), Michael P. Lehmann, Bonny E. Sweeney, and Samantha J. Stein, Hausfeld LLP, San Francisco, California; Jonathan W. Cuneo (argued), Joel Davidow, and Blaine Finley, Cuneo Gilbert & Laduca LLP, Washington, D.C.; Thomas H. Burt (argued), Wolf Haldenstein Adler Freeman & Herz LLP, New York, New York; Betsy C. Manifold, Rachele R. Byrd, Marisa C. Livesay, and Brittany N. Dejong, Wolf Haldenstein Adler Freeman & Herz LLP, San Diego, California; for Plaintiffs- Appellees.

Robert S. Kitchenoff, President, Committee to Support the Antitrust Laws, Washington, D.C.; Warren T. Burns and Kyle K. Oxford, Burns Charest LLP, Dallas, Texas; for Amicus Curiae Committee to Support the Antitrust Laws.

Ashley C. Parrish and Joshua N. Mitchell, King & Spalding LLP, Washington, D.C.; Anne M. Voigts and Quyen L. Ta, King & Spalding LLP, San Francisco, California; Steven P. Lehotsky and Jonathan D. Urick, U.S. Chamber Litigation Center, Washington, D.C.; for Amicus Curiae Chamber of Commerce of the United States.

Randy M. Stutz, American Antitrust Institute, Washington, D.C.; Ellen Meriwether, Cafferty Clobes Meriwether & Sprengal, Media, Pennsylvania; for Amicus Curiae American Antitrust Institute.

Scott L. Nelson and Allison M. Zieve, Public Citizen Litigation Group, Washington, D.C., for Amicus Curiae Public Citizen.

Corbin K. Barthold and Cory L. Andrews, Washington Legal Foundation, Washington, D.C., for Amicus Curiae Washington Legal Foundation.

OPINION

BUMATAY, Circuit Judge:

StarKist Company and Tri-Union Seafoods d/b/a Chicken of the Sea (collectively, “Defendants”), 1 producers of packaged tuna, appeal an order certifying three classes in a multidistrict antitrust case alleging a price-fixing conspiracy. Defendants challenge the district court’s determination that Rule 23(b)(3)’s “predominance” requirement was satisfied by expert statistical evidence finding classwide impact based on averaging assumptions and pooled transaction data.

We ultimately conclude that this form of statistical or “representative” evidence can be used to establish predominance, but the district court abused its discretion by not resolving the factual disputes necessary to decide the requirement before certifying these classes. We thus vacate the district court’s order certifying the classes and remand for the court to determine the number of uninjured parties in the proposed class based on the dueling statistical evidence. Only then should the district court rule on whether predominance has been established.

1 As a result of Appellant Bumble Bee Foods LLC’s bankruptcy proceeding, appellate proceedings against Bumble Bee Foods have been held in abeyance due to the automatic stay imposed by 11 U.S.C. § 362. Dkt. No. 51. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 11

I. FACTUAL AND PROCEDURAL BACKGROUND

A. The Price-Fixing Conspiracy

Various purchasers of tuna products (“Plaintiffs”) brought this class action alleging a price-fixing conspiracy by Defendants, the three largest domestic producers of packaged tuna. Together, Defendants account for over 80% of all branded packaged tuna sales in the country. Plaintiffs allege that Defendants colluded to artificially inflate the prices of their tuna products by engaging in various forms of anti-competitive conduct, including agreeing to (1) fix the net and list prices for packaged tuna, (2) limit promotional activity for packaged tuna, and (3) exchange sensitive or confidential business information in furtherance of the conspiracy. There is little dispute over the existence of a price-fixing scheme. Soon after this action was commenced, the Department of Justice initiated criminal charges against Defendants for their price-fixing conspiracy. Bumble Bee and StarKist have since pleaded guilty to federal, criminal price-fixing charges, as have several of their current and former executives. Chicken of the Sea has also admitted to price fixing and agreed to cooperate with the federal investigation.

B. Certifying the Classes

Plaintiffs proposed three classes of purchasers who bought packaged tuna products between November 2010 and December 2016.

The first proposed class, called the Direct Purchaser Plaintiff (“DPP”) Class, consists of retailers who directly purchased packaged tuna products during the relevant period. In support of certification, the Plaintiffs submitted the expert testimony and report of econometrician

Dr. Russell Mangum III. Dr. Mangum “primarily” relied on statistical evidence “in the form of a regression model which purports to prove that the price-fixing conspiracy harmed all, or nearly all, of the Class members.” First, Dr. Mangum calculated what the price for wholesale tuna would have been “but for” the alleged price fixing. To do so, he compared the prices during the period of the alleged price- fixing scheme to prices either before or after the alleged impacted period, while controlling for other factors that affect price differences. Comparing that but-for price to a “clean” benchmark period with no anticompetitive activity, Dr. Mangum concluded that the DPP Class was overcharged by an average of 10.28% because of the price fixing. Finally, assuming each class member experienced the same 10.28% average overcharge, Dr. Mangum ran a regression analysis and concluded that 1,111 out of 1,176 direct purchasers (or 94.5%) were injured by Defendants’ actions.

The Defendants’ expert econometrician, Dr. John Johnson, posed several objections to Dr. Mangum’s methodology. First, Dr. Johnson contended that because Dr. Mangum used an average estimated overcharge, his model incorrectly assumed “every direct purchaser was injured—and necessarily in the same way.” Dr. Johnson instead calculated a unique overcharge coefficient for 604 individual class members and concluded that only 72% paid an inflated price, meaning 28% of the class members suffered no injury at all. Second, Dr. Johnson argued that Dr. Mangum found “false positives” because his equation identified overcharges during the “clean” benchmark period by both Defendants and by packaged tuna sellers who are not Defendants. Additionally, Dr. Johnson claimed that Dr. Mangum relied on faulty economic assumptions. For example, Dr. Mangum’s report purportedly assumed that all Defendants would respond identically to changes in supply OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 13

and demand factors, and therefore costs would rise or fall identically across all producers. Dr. Johnson also commented that Dr. Mangum’s model failed a “Chow Test,” which examines the stability of coefficients among separate subgroups of a data set to determine if pooling them together to create an average is appropriate.

In rebuttal, Dr. Mangum noted that Dr. Johnson did not keep the average overcharge coefficient constant but rather allowed that coefficient to vary by customer. According to Dr. Mangum, this created too small sample sizes of customers with each coefficient, and this explained why Dr. Johnson was unable to create any results for some members of the DPP Class. Dr. Mangum claimed that, even under Dr. Johnson’s analysis, 98% of DPP customers were overcharged if those customers who showed no result whatsoever were excluded. 2

The district court certified the class, concluding that the Defendants’ challenges to Dr. Mangum’s methods were “ripe for use at trial” but “not fatal to a finding of classwide impact.” In re Packaged Seafood Prod. Antitrust Litig., 332 F.R.D. 308, 325 (S.D. Cal. 2019). The district court stressed that although Dr. Johnson’s “criticisms are serious and could be persuasive to a finder of fact . . . determining which expert is correct is beyond the scope” of a class certification motion. Id. at 328. The court instead thought the critical issue was to determine whether Dr. Mangum’s method is “capable of showing” impact on all or nearly all class members. Id. Because it was not persuaded that “Dr. Mangum’s model is unreliable or incapable of proving

2 This is compared to Dr. Mangum’s view that 94% of DPP customers were overcharged if only statistically significant results were considered.

impact on a class-wide basis,” the court found predominance established for the DPP Class. Id.

For the next two proposed classes, Plaintiffs offered expert reports and testimony that proceeded similarly to Dr. Mangum’s statistical analysis. The Commercial Food Service Product (“CFP”) Class consists of those who purchased packaged tuna products of 40 ounces or more from six major retailers (Dot Foods, Sysco, US Foods, Sam’s Club, Wal-Mart, and Costco). The End Payer Plaintiffs (“EPP”) Class is defined as consumers who bought Defendants’ packaged tuna products in cans or pouches smaller than 40 ounces for end consumption from any of the six major retailers. Defendants’ expert, Dr. Laila Haider, objected to Plaintiffs’ experts’ methodology largely for the same reasons raised in opposition to the DPPs’ methodology, focusing on benchmark selection, averaging, and false positives. Finding only “subtle differences” between the methodologies of Plaintiffs’ experts’ and Defendants’ objections in these two classes and the DPP Class, the district court certified the CFP and the EPP Classes. Despite finding “potential flaws” in Plaintiffs’ experts’ methodology, the court nonetheless concluded it was “reliable and capable of proving impact” and that the jury could determine whether liability and damages were proven.

A motions panel granted Defendants’ petition for permission to appeal the class certification order under Federal Rule of Civil Procedure Rule 23(f) and 28 U.S.C. § 1292(e). OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 15

II. LEGAL STANDARDS

A. Standard of Review

We review a district court’s decision to certify a class under Rule 23 for abuse of discretion and review the factual findings for clear error. Torres v. Mercer Canyons Inc., 835 F.3d 1125, 1132 (9th Cir. 2016).

B. The Predominance Requirement

Class actions are “an exception to the usual rule that litigation is conducted by and on behalf of the individual named parties only.” Comcast Corp. v. Behrend, 569 U.S. 27, 33 (2013) (simplified). To police this exception, Rule 23 imposes “stringent requirements” for class certification. Am. Express Co. v. Italian Colors Rest., 570 U.S. 228, 234 (2013). A party seeking class certification must first meet Rule 23(a)’s four requirements: numerosity, commonality, typicality, and adequacy of representation. Leyva v. Medline Indus., 716 F.3d 510, 512 (9th Cir. 2013); see Fed. R. Civ. P. 23(a). “To obtain certification of a class action for money damages under Rule 23(b)(3),” a putative class must also establish that “the questions of law or fact common to class members predominate over any questions affecting only individual members.” Amgen Inc. v. Conn. Ret. Plans & Tr. Funds, 568 U.S. 455, 460 (2013); see Fed. R. Civ. P. 23(b)(3).

When considering whether to certify a class, it is imperative that district courts “take a close look at whether common questions predominate over individual ones.” Comcast, 569 U.S. at 34. The Supreme Court has made clear that district courts must perform a “rigorous analysis” to determine whether this exacting burden has been met before certifying a class. Id. at 35; Wal-Mart Stores, Inc. v. Dukes,

564 U.S. 338, 350–51 (2011). This “rigorous analysis” requires “judging the persuasiveness of the evidence presented” for and against certification. Ellis v. Costco Wholesale Corp., 657 F.3d 970, 982 (9th Cir. 2011). Courts must resolve all factual and legal disputes relevant to class certification, even if doing so overlaps with the merits. Wal- Mart, 564 U.S. at 351. A district court abuses its discretion when it fails to adequately determine predominance was met before certifying the class. See Valentino v. Carter-Wallace, Inc., 97 F.3d 1227, 1234 (9th Cir. 1996).

C. The Burden of Proof for Predominance

Although we have not previously addressed the proper burden of proof at the class certification stage, we hold that a district court must find by a preponderance of the evidence that the plaintiff has established predominance under Rule 23(b)(3). See In re Lamictal Direct Purchaser Antitrust Litig., 957 F.3d 184, 191 (3d Cir. 2020) (holding that district courts must find by a “preponderance of the evidence that the plaintiffs’ claims are capable of common proof at trial”); In re Nexium Antitrust Litig., 777 F.3d 9, 27 (1st Cir. 2015) (holding that plaintiffs must show “each disputed requirement has been proven by a preponderance of evidence”); Messner v. Northshore Univ. HealthSystem, 669 F.3d 802, 811 (7th Cir. 2012) (“Plaintiffs bear the burden of showing that a proposed class satisfies the Rule 23 requirements, but they need not make that showing to a degree of absolute certainty. It is sufficient if each disputed requirement has been proven by a preponderance of evidence.”); Alaska Elec. Pension Fund v. Flowserve Corp., 572 F.3d 221, 228 (5th Cir. 2009) (“[A]n issue of predominance must be established at the class certification stage by a preponderance of all admissible evidence.”) (simplified); Teamsters Loc. 445 Freight Div. Pension Fund OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 17

v. Bombardier Inc., 546 F.3d 196, 202 (2d Cir. 2008) (“[We] hold that the preponderance of the evidence standard applies to evidence proffered to establish Rule 23’s requirements.”); see also Newberg on Class Actions, § 7:21 (5th ed.) (“The trend in recent cases has been a move . . . towards adoption of a preponderance of the evidence standard to facts necessary to establish the existence of a class.”). 3

Aside from joining our sister circuits, employing a preponderance of the evidence standard supports the district court’s role as the gatekeeper of Rule 23’s requirements. See Wal-Mart, 564 U.S. at 351; Crutchfield v. Sewerage & Water Bd. of New Orleans, 829 F.3d 370, 375 (5th Cir. 2016) (holding the predominance inquiry envisions “what a class

3 A number of district courts in our circuit have likewise applied a preponderance of the evidence standard to establish a class. See, e.g., Gomez v. J. Jacobo Farm Labor Contractor, Inc., 334 F.R.D. 234, 248 (E.D. Cal. 2019) (“Federal courts throughout the country require the movant to demonstrate by a preponderance of the evidence that class certification is appropriate.”); Martin v. Sysco Corporation, 325 F.R.D. 343, 354 (E.D. Cal. 2018) (“While Rule 23 does not specifically address the burden of proof to be applied, courts routinely employ the preponderance of the evidence standard.”); Valenzuela v. Ducey, 2017 WL 6033737, at *3 (D. Ariz. Dec. 6, 2017) (“[The preponderance of the evidence] standard appears to be the trend in federal courts and will be applied in this case.”) (simplified); Southwell v. Mortg. Inv’rs Corp. of Ohio, 2014 WL 3956699, at *1 (W.D. Wash. Aug. 12, 2014) (“[T]his Court finds itself in need of such a standard and chooses to align itself with the emerging trend in other districts towards the adoption of a preponderance of the evidence standard[.]”); Smilovits v. First Solar, Inc., 295 F.R.D. 423, 427 (D. Ariz. 2013) (“[The preponderance] standard appears to be the trend in federal courts[.]”); Keegan v. American Honda Motor Co., Inc., 284 F.R.D. 504, 521 n.83 (C.D. Cal. 2012) (“[D]efendants cite no Ninth Circuit authority that directs use of a preponderance standard in deciding class certification motions. Because that is the general standard of proof used in civil cases, however, the court applies it here.”).

trial would look like”). It best accords with the Supreme Court’s warning that class certification is “proper only if the trial court is satisfied, after a rigorous analysis, that the prerequisites of Rule 23(a) have been satisfied.” Wal-Mart Stores, 564 U.S. at 349–51 (emphasis added). And a preponderance standard is more faithful to Rule 23(b)(3)’s text, which provides that courts can certify a class “only if . . . the court finds that the questions of law or fact common to class members predominate” over individual ones. Fed. R. Civ. P. 23(b)(3) (emphasis added).

The preponderance standard also flows from the Supreme Court’s emphasis that the evidence used to satisfy predominance be “sufficient to sustain a jury finding as to [liability] if it were introduced in each [plaintiff’s] individual action.” Tyson Foods, Inc. v. Bouaphakeo, 136 S. Ct. 1036, 1048 (2016) (emphasis added). Establishing predominance, therefore, goes beyond determining whether the evidence would be admissible in an individual action. Instead, a “rigorous analysis” of predominance requires “judging the persuasiveness of the evidence presented” for and against certification. Ellis, 657 F.3d at 982 (vacating class certification because the district court “confused the Daubert standard” for admissibility of expert evidence “with the ‘rigorous analysis’ standard to be applied when analyzing” the Rule 23 factors). 4

4 We acknowledge that Tyson Foods stated that once a district court finds representative evidence “admissible, its persuasiveness is, in general, a matter for the jury,” and class certification should only be denied if “no reasonable juror” could have found the plaintiffs’ representative evidence persuasive. 136 S. Ct. at 1049. But that discussion was in the context of a wage-and-hour class action where representative evidence is explicitly permitted to establish liability in individual cases. Id. (citing Anderson v. Mt. Clemens Pottery Co., OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 19

D. The Use of Representative Evidence

The acceptance of representative evidence at the class certification stage is nothing new. The Supreme Court has held that representative evidence can be relied on to establish a class, but it has also declined to adopt “broad and categorical rules governing” its use. Tyson Foods, 136 S. Ct. at 1049. Instead, whether a representative sample can “establish classwide liability” at the certification stage “will depend on the purpose for which the sample is being introduced and on the underlying causes of action.” Id. While consideration of representative evidence may be flexible, it must be scrutinized with care and vigor. See Comcast, 569 U.S. at 35 (rejecting the use of representative evidence to establish predominance); Wal-Mart, 564 U.S.

328 U.S. 680, 687 (1946)). Such an evidentiary rule exists because defendants often fail to keep proper records of hours worked by employees. Id.; see also Lamictal, 957 F.3d at 191–92 (discussing how representative evidence is particularly appropriate in wage-and-hour suits since “a representative sample of employees may be the only feasible way to establish liability” in a wage-and-hour case due to the defendant’s own “inadequate record keeping”).

Given that representative evidence can be used to infer harm in individual wage-and-hour suits, Tyson Foods reasoned that representative evidence was presumptively usable at the class certification stage as well. See Tyson Foods, 136 S. Ct. at 1049; see also id. at 1046 (stating that representative evidence can be used to establish predominance if “each class member could have relied on that sample to establish liability if he or she had brought an individual action.”). But the “no reasonable jury” standard is cabined to wage-and-hour suits and doesn’t apply here. See Senne v. Kan. City Royals Baseball Corp., 934 F.3d 918, 923, 947 & n.27 (9th Cir. 2019) (“Tyson expressly cautioned that this rule should be read narrowly and not assumed to apply outside of the wage and hour context.”).

at 350–51 (rejecting the use of representative evidence to establish commonality).

There is reason to be wary of overreliance on statistical evidence to establish classwide liability. Academic literature abounds observing that “judges and jurors, because they lack knowledge of statistical theory, are both overawed and easily deceived by statistical evidence.” United States v. Veysey, 334 F.3d 600, 604 (7th Cir. 2003). 5 If “highly consequential evidence emerges from what looks like an indecipherable” statistical model to most “non-statisticians,” it is “imperative that qualified individuals explain how the [model] works,” and courts must “ensure that it produces reliable information.” United States v. Gissantaner, — F.3d —, 2021 WL 834005, at *3 (6th Cir. 2021). 6

Moreover, the use of representative evidence cannot “abridge, enlarge or modify [a plaintiff’s] substantive right[s].” See 28 U.S.C. § 2072(b). Otherwise, its use would

5 See, e.g., Douglas H. Ginsburg & Eric M. Fraser, The Role of Economic Analysis in Competition Law (May 16, 2010) (“[Courts] almost certainly will not have the assistance of even one staff economist, nor will the judges likely be familiar with the economic concepts about the application of which [the parties] are debating.”); Laurence H. Tribe, Trial by Mathematics: Precision and Ritual in the Legal Process, 84 Harv. L. Rev. 1329, 1342 n.40 (1971) (discussing how courts misunderstand and misapply statistical evidence); G. Alexander Nunn, The Incompatibility of Due Process and Naked Statistical Evidence, 68 Vand. L. Rev. 1407 (2015) (discussing how the use of statistical evidence in certain circumstances can constitute a due process violation). 6 As Mark Twain famously popularized, “[t]here are three kinds of lies: lies, damn lies, and statistics.” See Mark Twain, Chapters from My Autobiography—XX, 186 N. Am. Rev. 465, 471 (1907). Although we welcome the use of statistical evidence when appropriate, it would be injudicious to swallow it uncritically. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 21

contravene the Rules Enabling Act. Id. Class actions are merely a procedural tool aggregating claims, Sprint Commc’ns Co. v. APPCC Servs., Inc., 554 U.S. 269, 291 (2008), and Rule 23 “leaves the parties’ legal rights and duties intact and the rules of decision unchanged,” Shady Grove Orthopedic Assocs., P.A. v. Allstate Ins. Co., 559 U.S. 393, 408 (2010) (plurality opinion). The use of representative evidence at the class certification stage must therefore be closely and carefully scrutinized, and “[a]ctual, not presumed, conformance” with Rule 23’s requirements is “indispensable.” Wal-Mart, 564 U.S. at 351 (simplified).

With these background legal principles in mind, we turn to Defendants’ contentions on appeal.

III. DEFENDANTS’ CLAIMS

Defendants raise two challenges to the district court’s reliance on Plaintiffs’ representative evidence. First, Defendants argue that this type of representative evidence— especially the use of averaging assumptions—cannot be used to establish predominance. Second, Defendants claim that, even if this type of evidence can show predominance, Plaintiffs’ econometric analysis does not in fact establish predominance because a significant percentage of the class may have suffered no injury at all under Plaintiffs’ experts’ statistical modeling. We consider each argument in turn.

A. Whether Plaintiffs’ Representative Evidence Can Establish Predominance

The threshold consideration is whether Plaintiffs’ representative evidence can be used to establish predominance. We believe this question raises several considerations.

First, we address whether the representative evidence could be used to establish liability in an individual suit. Tyson Foods, 136. S. Ct. at 1048. Second, we ensure that classwide liability is “capable of proof” through the representative analysis. Comcast, 569 U.S. at 30. Finally, we assess whether the use of averaging assumptions masks the predominance question itself “by assuming away the very differences that make the case inappropriate for classwide resolution.” Tyson Foods, 136 S. Ct. at 1046.

We conclude that Plaintiffs’ representative evidence can prove the classwide impact element of Plaintiffs’ price- fixing theory of liability and, thus, may be used to establish predominance.

1. Plaintiffs’ Evidence Could Have Been Used to Establish Liability in a Class Member’s Individual Suit

To establish predominance, the representative evidence must be capable of use at trial in individual—not just class action—antitrust cases. See Tyson Foods, 136 S. Ct. at 1046 (Representative evidence is permissible to establish predominance if “each class member could have relied on that sample to establish liability if he or she had brought an individual action.”). This is because plaintiffs and defendants cannot have “different rights in a class proceeding than they could have asserted in an individual action.” Id. at 1048. If the representative evidence could not be “relied on . . . to establish liability” in an “individual action,” id. at 1046, then it cannot establish predominance at the class certification stage.

The District Court held that to meet the predominance requirement on their antitrust claims, Plaintiffs had to establish: (1) the existence of an antitrust conspiracy; (2) the OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 23

existence of individual injury, also referred to as “antitrust impact,” as a result of the conspiracy; and (3) resultant damages. Packaged Seafood, 332 F.R.D. at 320; see 1 McLaughlin on Class Actions § 5:33 (17th ed. 2020); see also In re New Motor Vehicles Canadian Exp. Antitrust Litig., 522 F.3d 6, 19 n. 18 (1st Cir. 2008).

Plaintiffs rely on their representative evidence to establish the “antitrust impact” of their price-fixing claims against the Defendants. Statistical evidence has long been used to prove antitrust impact in individual suits. To establish impact in any antitrust action, plaintiffs must “delineate a relevant market and show that the defendant plays enough of a role in that market to impair competition significantly.” Metro Indus., Inc. v. Sammi Corp., 82 F.3d 839, 847–48 (9th Cir. 1996). Even in individual suits, doing so often requires comparing the actual world with a “hypothetical” world that would have existed “‘but for’ the defendant’s unlawful activities.” See LePage’s Inc. v. 3M, 324 F.3d 141, 165 (3d Cir. 2003); see, e.g., MM Steel, L.P. v. JSW Steel (USA) Inc., 806 F.3d 835, 851–52 (5th Cir. 2015) (holding that the district court didn’t abuse its discretion by using a “yardstick” calculation of damages in an antitrust suit where the individual plaintiffs did a but-for analysis by comparing their profits with “a study of the profits of business operations that are closely comparable to the plaintiff’s”).

In individual cases, constructing these “but-for” comparisons usually requires the use of statistical evidence. See Manual of Complex Litigation (Fourth) § 23.1, at pp. 470–71 (“[S]tatistical evidence is routinely introduced . . . in antitrust litigation.”). And injury may be inferred from statistical evidence. See Zenith Radio Corp. v. Hazeltine Research, Inc., 395 U.S. 100, 125 (1969) (stating that

antitrust impact can be inferred from “circumstantial evidence”); see also ABA Section of Antitrust Law, Econometrics: Legal, Practical, and Technical Issues § 13.B.1.c. (2d ed. 2014) (discussing the use of regression models in antitrust actions).

Here, each class member could have relied on Dr. Mangum’s models to show classwide impact in each of their individual suits. By constructing a clean, “benchmark” period and comparing it to market price before and after the benchmark, Dr. Mangum created a “yardstick” comparison to isolate the “but-for” effect of the price-fixing conspiracy, similar to the type of evidence relied upon in individual antitrust actions. See, e.g., LePage’s Inc., 324 F.3d at 165; MM Steel, 806 F.3d at 851–52. And the regression analysis Dr. Mangum ran to calculate that 94% of the DPP Class suffered an injury is consistent with the use of regression models to prove price-fixing impact in other cases. See, e.g., In re Linerboard Antitrust Litig., 305 F.3d 145, 153 (3d Cir. 2002) (affirming use of plaintiffs’ “multiple regression analysis” to prove “impact on a class-wide basis” in price- fixing suit). In short, Plaintiffs’ statistical evidence is not materially different than the type of evidence that would be used against Defendants in individual cases brought by each class member.

2. Plaintiffs’ Representative Evidence Sufficiently Links Their Injuries to Their Theory of Antitrust Violation

Plaintiffs’ representative evidence must also be consistent with their underlying theory of liability. Comcast, 569 U.S. at 35 (“[A]ny model supporting a plaintiff’s damages case must be consistent with its liability case, particularly with respect to the alleged anticompetitive effect of the violation.”). We have interpreted Comcast to require OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 25

that plaintiffs “show that their damages stemmed from the defendant’s actions.” Pulaski & Middleman, LLC v. Google, Inc., 802 F.3d 979, 987–88 (9th Cir. 2015) (simplified). Put another way, the evidence must be capable of linking the harm from the defendant’s conduct to the class members.

In this case, there is a sufficient nexus between Plaintiffs’ representative evidence and their price-fixing theory of liability. See Allied Orthopedic, 592 F.3d at 996. Dr. Mangum’s regression model can show antitrust impact by isolating the but-for effect of the price inflation attributable to Defendants’ alleged anticompetitive price list (the 10.28% average overcharge), and by using a regression model to calculate how much of the class would have been impacted by that overcharge. Plaintiffs thus present a “theory of injury and damages” that is “provable and measurable by an aggregate model relying on class-wide data.” In re Suboxone Antitrust Litig., 967 F.3d 264, 272 (3d Cir. 2020) (affirming representative evidence in an antitrust class action).

Accordingly, this is unlike cases where courts have disapproved of representative evidence. In Comcast, for example, the Court rejected representative evidence because the posited regression analysis showed common injury that did not track the plaintiffs’ underlying theory of liability. 569 U.S. at 35–38. There, the plaintiffs’ regression model accounted for four different antitrust theories of harm, even though the district court had only allowed the plaintiffs to proceed on one of these theories. Id. at 31–32, 35. Such a model “failed to measure damages resulting from the particular antitrust injury on which” the class premised its claim and “identifie[d] damages that are not the result of the wrong” suffered by the certified class. Id. at 36–37. By contrast, here Plaintiffs’ regression models test only one

theory of liability: the but-for impact of Defendants’ price- fixing conspiracy.

3. Plaintiffs’ Use of Averaging Assumptions Does Not Defeat Predominance

Defendants also argue that the representative evidence at issue here is categorically impermissible because Plaintiffs’ experts used averaging assumptions in their regression models. But the Supreme Court rejected “categorical exclusion” of representative evidence. Tyson Foods, 136 S. Ct. at 1046. Instead, Tyson approved the use of averaging assumptions so long as the statistical evidence was “reliable in proving or disproving the elements of the relevant cause of action.” Id.

The use of averaging assumptions in a regression analysis may be inappropriate “where [a] small sample size may distort the statistical analysis and may render any findings not statistically probative.” Paige v. California, 291 F.3d 1141, 1148 (9th Cir. 2002). Indeed, Dr. Mangum’s rebuttal to Dr. Johnson’s testimony was that varying the overcharge value in his regression analysis resulted in too small sample sizes that were not statistically robust.

Here, we see no issue with Plaintiffs’ use of averaging assumptions in its regression models. Dr. Mangum averaged the overcharge calculation using Defendants’ own data, and then used that average in a regression model to calculate what percentage of the class was impacted. Presuming the reliability of Plaintiffs’ statistical methodology (which we discuss later), the representative evidence can show that virtually all class members suffered an injury due to Defendants’ alleged wrongdoing. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 27

According to Defendants, Plaintiffs’ averaging assumptions papered over the very individualized differences that make classwide resolution of this case inappropriate. Defendants stress that “innumerable individualized differences” among the class members make it impossible to show class-wide impact through “common proof.” For instance, direct purchasers often individually negotiate prices, and the prices retailers actually pay may vary based on purchasing power, retail price strategy, and other factors. Some retailers may have even sold Defendants’ tuna products as a loss leader to drive customers to their stores. Defendants also contend that these averaging assumptions are even more inappropriate when applied to the indirect-purchaser class, which contains “even more disparate” class members, including millions of individuals who bought billions of tuna products from “countless stores across the country over a four-year period.”

But even assuming the existence of these individualized differences, a higher initial list price as a result of Defendants’ price-fixing scheme could have raised the baseline price at the start of negotiations and could have affected the range of prices that resulted from negotiation. Even Walmart, which as the largest retailer in the country would have had the strongest bargaining power of any class member, was shown to have suffered overcharges as a result of Defendants’ conduct. This relieves concerns that the class members were not “similarly situated,” and would allow the “reasonable inference of class-wide liability.” See Tyson Foods, 136 S. Ct. at 1045 (citations omitted).

Moreover, even if class members suffered individualized damages that diverged from the average overcharge calculated by Plaintiffs’ expert, “the presence of individualized damages cannot, by itself, defeat class

certification under Rule 23(b)(3).” Leyva, 716 F.3d at 514. Indeed, we have consistently distinguished the existence of injury from the calculation of damages. See Vaquero, 824 F.3d at 1155; Senne, 934 F.3d at 943. Consequently, individualized damages calculations do not, alone, defeat predominance—although, as we discuss below, the presence of class members who suffered no injury at all may defeat predominance.

*****

Because this type of representative evidence can be used to prove injury in individual antitrust suits, is consistent with Plaintiffs’ underlying cause of action, and doesn’t necessarily mask a lack of predominance, we hold it is permissible to rely on Plaintiffs’ representative evidence at the class certification stage.

B. Whether the District Court Must Rule on the Presence of Uninjured Class Members

Even if Plaintiffs’ representative evidence could be used to satisfy predominance, we cannot embrace their conclusions and averaging assumptions uncritically. Statistical evidence is not a talisman. Courts must still rigorously analyze the use of such evidence to test its reliability and to see if the statistical modeling does in fact mask individualized differences.

As stated earlier, reliability is the touchstone for establishing predominance through representative sampling. See Tyson Foods, 136 S. Ct. at 1046. It is thus necessary for courts to consider “the degree to which the evidence is reliable in proving or disproving” whether a common question of law or fact predominates over the class members. Id. (emphasis added); see also Vaquero, 824 F.3d at 1155. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 29

To do so, courts must “resolve any factual disputes necessary to determine whether” predominance has in fact been met. Ellis, 657 F.3d at 982–84. In other words, the threshold predominance determination cannot be outsourced to a jury. Lamictal, 957 F.3d at 191 (“[T]he court must resolve all factual or legal disputes relevant to class certification[.]”) (simplified).

When considering if predominance has been met, a key factual determination courts must make is whether the plaintiffs’ statistical evidence sweeps in uninjured class members. As the district court recognized, Plaintiffs “must establish, predominantly with generalized evidence, that all (or nearly all) members of the class suffered damage as a result of Defendants’ alleged anti-competitive conduct.” Packaged Seafood, 332 F.R.D. at 320 (simplified). If a substantial number of class members “in fact suffered no injury,” the “need to identify those individuals will predominate.” In re Asacol Antitrust Litig., 907 F.3d 42, 53 (1st Cir. 2018); see Halvorson v. Auto-Owners Ins. Co., 718 F.3d 773, 779 (8th Cir. 2013). If injury cannot be proved or disproved through common evidence, then “individual trials are necessary to establish whether a particular [class member] suffered harm from the [alleged misconduct],” and class treatment under Rule 23 is accordingly inappropriate. In re Rail Freight Fuel Surcharge Antitrust Litig.-MDL No. 1869, 725 F.3d 244, 252 (D.C. Cir. 2013); see also Tyson Foods, 136 S. Ct. at 1045. 7

7 The presence of uninjured parties in a certified class also raises serious standing implications under Article III. The federal court system is reserved only for those that have suffered an injury. See Lujan v. Defs. of Wildlife, 504 U.S. 555, 566 (1992). To that end, standing requires each plaintiff provide “a factual showing of perceptible harm.” Id. A

In this case, the district court abused its discretion in declining to resolve the competing expert claims on the reliability of Plaintiffs’ statistical model. Defendants’ expert provided testimony and alternative statistical modeling that suggested Plaintiffs’ data was methodologically flawed and was unable to show impact for up to 28% of the class—not 5.5%, as Plaintiffs’ expert insists. Rather than resolving the dispute, however, the district court merely considered whether Plaintiffs’ statistical evidence was “plausibly reliable” and otherwise left determination of this question to the jury. It concluded that “determining which expert is correct is beyond the scope” of class certification and was “ultimately a merits decision” for the jury to decide. 8

But resolving this dispute is of paramount importance to certification of the class. If Plaintiffs’ model indeed shows that more than one-fourth of the class may have suffered no

class action should be no different. See Tyson Foods, 136 S. Ct. at 1053 (Roberts, C.J., concurring) (“Article III does not give federal courts the power to order relief to any uninjured plaintiff, class action or not.”). Accordingly, as the Fifth Circuit recently expressed, we are skeptical that Article III permits certification of a class where “[c]ountless unnamed class members lack standing.” Flecha v. Medicredit, Inc., 946 F.3d 762, 768 (5th Cir. 2020). But we do not reach this issue because, as we lay out, class certification fails under Rule 23(b)(3), which is dispositive of the matter. In re Hyundai & Kia Fuel Economy Litig., 926 F.3d 539, 565 n.12 (9th Cir. 2019). 8 Courts cannot relocate the predominance inquiry to the merits stage of the trial. Rule 23 requires this determination be made at the pre- trial stage. And for good reason. Suppose the jury ultimately decides Defendants’ expert is right and Plaintiffs’ model sweeps in 28% uninjured class members. Too late: the damage has been done. By then, Defendants would have possibly weathered years of litigation at untold costs, only to discover that the case never should have reached the merits at all. Rule 23’s objective—that only cases suitable for class adjudication be certified—would have been effectively undermined. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 31

injury at all, the district court cannot find by a preponderance of the evidence that “questions of law or fact common to class members predominate over any questions affecting only individual members.” Fed. R. Civ. P. 23(b)(3).

Rule 23(b)(3)’s choice of wording matters. The word “common” means “belonging to or shared . . . by all members of a group.” 9 Meanwhile, “predominate” means “to hold advantage in numbers or quantity.” 10 Similarly, when used as a noun, the word “predominance” means “the state of . . . being most frequent or common.” 11 Thus, Rule 23(b)(3) requires that questions of law or fact be shared by substantially all the class members, and these common questions must be superior in strength or pervasiveness to individual questions within the class.

If 28% of the class were uninjured, common questions of law or fact would not be shared by substantially all the class members, nor would they prevail in strength or pervasiveness over individual questions. This would raise concerns that Plaintiffs’ experts’ use of average assumptions did mask individual differences among the class members,

9 Common, Merriam-Webster’s Collegiate Dictionary (11th ed. 2007). 10 Predominate, Merriam-Webster’s Collegiate Dictionary; see also Predominate, Oxford English Dictionary, https://www.oed.com/view/Entry/149893 (defining “predominate” as “[t]o have or exert controlling power; to be of greater authority or influence, to be superior”). 11 Predominance, Merriam-Webster’s Collegiate Dictionary; see also Predominance, Oxford Online English Dictionary, https://www.oed.com/view/Entry/149888 (defining “predominance” as “preponderance, prevalence; prevailing or superior influence, power, or authority”).

such as bargaining power, negotiation positions, and marketing strategies.

Although we have not established a threshold for how great a percentage of uninjured class members would be enough to defeat predominance, it must be de minimis. Even though “a well-defined class may inevitably contain some individuals who have suffered no harm,” Torres, 835 F.3d at 1136, the few reported decisions involving uninjured class members “suggest that 5% to 6% constitutes the outer limits of a de minimis number,” In re Rail Freight Fuel Surcharge Antitrust Litig., 934 F.3d 619, 624–25 (D.C. Cir. 2019) (simplified) (finding no predominance where 12.7% of class members were conceded to be uninjured by plaintiffs’ own expert). The First Circuit reversed certification where the district court had concluded that “around 10%” of the proposed class was uninjured. See In re Asacol, 907 F.3d at 47, 51–58. And even the district court recognized that the inclusion of 28% uninjured class members would “unquestionably” defeat predominance. Packaged Seafood, 332 F.R.D. at 325. Contrary to the dissent’s suggestion, we do not adopt a numerical or bright-line rule today. 12 But under any rubric, if Plaintiffs’ model is unable to show impact for more than one-fourth of the class members, predominance has not been met. 13 While we do not set the

12 The dissent also claims that we ignore Ninth Circuit case law. Dissent at 36. Not so. We agree with Torres v. Mercer Canyons Inc., 835 F.3d 1125, 1136 (9th Cir. 2016) that the mere presence of some non- injured class members does not defeat predominance, but we hold that the number of uninjured class members must be de minimis. As Torres stated, the “existence of large numbers of class members” who were never exposed to injurious conduct may defeat predominance. Id. 13 This is over double the percentage of uninjured class members considered sufficient to defeat predominance in In re Rail Freight OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 33

upper bound of what is de minimis, it’s easy enough to tell that 28% would be out-of-bounds.

The district court’s gloss over the number of uninjured class members was an abuse of discretion. Rule 23(b)(3) requires courts “to make findings about predominance and superiority before allowing the class.” Wal-Mart, 564 U.S. at 363 (emphasis added). Deferring determination of classwide impact effectively “amounts to a delegation of judicial power to the plaintiffs, who can obtain class certification just by hiring a competent expert.” West v. Prudential Sec., Inc., 282 F.3d 935, 938 (7th Cir. 2002). If “savvy crafting of the evidence” were enough to guarantee predominance, there would be little limit to class certification in our modern world of increasingly sophisticated aggregate proof.” See Richard A. Nagareda, Class Certification in the Age of Aggregate Proof, 84 N.Y.U. L. Rev. 97, 103 (2009). In Ellis, we vacated the district court’s certification of a class for the failure to resolve “critical factual disputes” in a “battle of the experts” regarding commonality. 657 F.3d at 982, 984. So too here, the district court failed to resolve the factual disputes as to how many uninjured class members are included in Plaintiffs’ proposed class—an essential component of predominance.

Plaintiffs emphasize that the district court stated its inquiry went beyond a Daubert analysis and that the court recognized it was required to determine whether the expert evidence was “in fact persuasive.” The district court even walked through the strengths and weaknesses of the experts’

(12.7%), almost triple the percentage disapproved of in In re Asacol (10%), and around five times greater than the percentages at issue in the district courts cited (5–6%).

competing testimony. Yet despite acknowledging there were “potential flaws” in the Plaintiffs’ expert’s methodology, the district court made no finding. A district court that “has doubts about whether the requirements of Rule 23 have been met should refuse certification until they have been met.” Brown v. Electrolux Home Prods., Inc., 817 F.3d 1225, 1233–34 (11th Cir. 2016) (simplified). 14

Despite admirably and thoroughly marshaling the evidence in this difficult case, the district court needed to go further by resolving the parties’ dispute over whether the representative evidence swept in only 5.5% or as much as 28% uninjured DPP Class members. The district court also needed to make a similar determination for the other putative classes. Deciding this preliminary question is necessary to determine whether Plaintiffs have established predominance.

IV. CONCLUSION

Accordingly, we vacate the district court’s order certifying the classes and remand with instructions to resolve the factual disputes concerning the number of uninjured

14 Compounding these concerns, the burden of persuasion may have been improperly shifted to Defendants to affirmatively disprove the claims made by Plaintiffs’ expert. In certifying the classes, the district court reasoned that “Defendants have not persuaded the Court that Dr. Mangum’s model is unreliable.” Packaged Seafood, 332 F.R.D. at 326. Additionally, the district court concluded that the predominance requirement was met because Defendants had not shown that Plaintiffs’ models were “glaringly erroneous.” Id. But the “party seeking class certification has the burden of affirmatively demonstrating that the class meets the requirements of [Rule] 23.” Mazza v. Am. Honda Motor Co., 666 F.3d 581, 588 (9th Cir. 2012). OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 35

parties in each proposed class before determining predominance. 15

VACATED and REMANDED.

HURWITZ, Circuit Judge, concurring in part and dissenting in part:

The majority is faithful to the plain text of Rule 23 in concluding that the district court, not a jury, must resolve factual disputes bearing on predominance. See Fed. R. Civ. P. 23(b)(3) (permitting a class action to be maintained if “the court finds that the questions of law or fact common to class members predominate over any questions affecting only individual members”) (emphasis added). I also agree with the majority that a district court’s “rigorous analysis” of whether a putative class has satisfied Rule 23’s requirements should proceed by a preponderance of the evidence standard. See Ellis v. Costco Wholesale Corp., 657 F.3d 970, 982 (9th Cir. 2011). And, the majority correctly holds that the question for the district court is not whether common issues could predominate at trial; the court must determine that they do predominate before certifying the class. See Comcast Corp. v. Behrend, 569 U.S. 27, 34 (2013). I therefore agree that remand is required.

I part company, however, with the majority’s conclusion that, before certifying a class, the district court must find that only a “de minimis” number of class members are uninjured.

15 Pursuant to Federal Rule of Appellate Procedure 39(a) and Ninth Circuit General Order 4.5(e), each party shall bear its own costs on appeal.

The text of Rule 23 contains no such requirement, nor do our precedents. The majority’s effective amendment of Rule 23 not only ignores our case law but also circumvents the established process for modifying a Rule of Civil Procedure—study and advice from the relevant committees, followed by the consent of the Supreme Court and Congress’s tacit approval. See Rules Enabling Act, 28 U.S.C. § 2072; Shady Grove Orthopedic Assocs., P.A. v. Allstate Ins. Co., 559 U.S 393, 407 (2010) (describing the Supreme Court’s rulemaking power). I therefore respectfully dissent from Part III.B of the majority opinion. 1

I

As an initial matter, our caselaw squarely forecloses the majority’s approach. The critical question is not what percentage of class members is injured, but rather whether the district court can economically “winnow out” uninjured plaintiffs to ensure they cannot recover for injuries they did not suffer. See Torres v. Mercer Canyons, Inc., 835 F.3d 1125, 1137 (9th Cir. 2016). If the district court can ensure that uninjured plaintiffs will not recover, their mere presence in the putative class does not mean that common issues will

1 The majority also notes that “[a]cademic literature abounds observing that ‘judges and jurors, because they lack knowledge of statistical theory, are both overawed and easily deceived by statistical evidence.’” Op. at 20 (quoting United States v. Veysey, 334 F.3d 600, 604 (7th Cir. 2003)). But even assuming that academic literature does so “abound,” see Op. at 20, n.5, that doesn’t establish that Article III judges in general, or the distinguished district judge in this case, are so easily fooled. The cited literature is, for better or worse, based on the observations of the authors, not on a rigorous scientific survey of the lack of knowledge of statistical theory by district judges (or even federal appellate judges). OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 37

not predominate. See Yokoyama v. Midland Nat’l Life Ins. Co., 594 F.3d 1087, 1089 (9th Cir. 2010).

The plain text of Rule 23 requires only that “questions of law or fact common to the class predominate over any questions affecting only individual members.” Fed. R. Civ. P. 23(b)(3) (emphasis added). The noun “predominant” means “[m]ore powerful, more common, or more noticeable.” Predominant, Black’s Law Dictionary (11th ed. 2019). In Rule 23(b)(3), the subject of the verb “predominate” is “common questions of law or fact.” The Rule therefore simply instructs the district court to determine whether common questions exceed others. See Pavelic & LeFlore v. Marvel Ent. Grp., 493 U.S. 120, 123 (1989) (applying statutory interpretation maxims to a Federal Rule of Civil Procedure); see also Weyerhauser Co. v. U.S. Fish and Wildlife Serv., 139 S. Ct. 361, 368 (2018) (reading statutory text “[a]ccording to the ordinary understanding of how adjectives work” to determine how the statute “modif[ies] nouns”).

We have therefore stressed that “[t]he potential existence of individualized damage assessments . . . does not detract from the action’s suitability for class certification.” Yokoyama, 594 F.3d at 1089; see also Advisory Comm. Note to 1966 Amendment, Rule 23 (“It is only where this predominance exists that economies can be achieved by means of the class-action device. In this view, a fraud perpetrated on numerous persons by the use of similar misrepresentations may be an appealing situation for a class action, and it may remain so despite the need, if liability is found, for separate determination of the damages suffered by individuals within the class.”). In Levya, for example, we stated that although “plaintiffs must be able to show that their damages stemmed from the defendant’s actions that

created the legal liability,” the presence of putative class members “allegedly entitled to different damage awards” did not defeat predominance. Levya v. Medline Indus., 716 F.3d 510, 513–14 (9th Cir. 2013); see also Vaquero v. Ashley Furniture Indus., Inc., 824 F.3d 1150, 1155 (9th Cir. 2016), Pulaski & Middleman, LLC v. Google, Inc., 802 F.3d 979, 988 (9th Cir. 2015) (holding that Comcast did not disturb Yokoyama). Even in a properly certified class, “[d]amages may well vary, and may require individualized calculations.” Senne v. Kansas City Royals Baseball Corp., 934 F.3d 918, 943 (9th Cir. 2019).

Most importantly, we have held that because “even a well-defined class may inevitably contain some individuals who have suffered no harm,” the same approach governs even if there are uninjured plaintiffs. Torres, 835 F.3d at 1136–37. Rather, the presence of some plaintiffs not harmed by the defendants’ conduct merely highlights the “possibility that an injurious course of conduct may sometimes fail to cause injury.” Id. at 1136. And, no Ninth Circuit case imposes a cap on the number of uninjured plaintiffs as a prerequisite to class certification.

Our settled law is consistent with the basic principles underlying Rule 23. A class plainly may be certified solely on discrete issues. See Fed. R. Civ. P. 23(c)(4). So, in the case before us, the district court could well certify a class on liability, followed by a more narrowly defined class (or even individual trials, if necessary) on damages. As the majority recognizes, there is little dispute the defendants engaged in an antitrust conspiracy. I perceive no bar in Rule 23 to certifying a liability class, while leaving open which members of the class suffered damage from the defendants’ illegal conduct. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 39

As the Fifth Circuit has recognized, the predominance inquiry focuses on “what a class trial would look like.” Crutchfield v. Sewerage & Water Bd. of New Orleans, 829 F.3d 370, 375 (5th Cir. 2016). The crucial question, left to the district court’s sound discretion, is whether “common questions present a significant aspect of the case and they can be resolved for all members of the class in a single adjudication.” True Health Chiropractic, Inc. v. McKesson Corp., 896 F.3d 923, 931 (9th Cir. 2018). Certification should fail only when the individual questions “threaten to become the focus of the litigation.” Torres, 835 F.3d at 1142.

II

A numerical cap on uninjured class members is not very helpful to district courts analyzing predominance. To be sure, a large percentage of uninjured plaintiffs may raise predominance concerns. See In re Asacol Antitrust Litig., 907 F.3d 42, 53–54 (1st Cir. 2018). Our cases plainly recognize that concern. See Torres, 835 F.3d at 1142.

But, as written, the Rule is not categorical with respect to the number of uninjured plaintiffs. If the questions of law or fact about whether a defendant breached a legal duty to a class are common, and identifying the uninjured members would be relatively simple, there is likely no reason to deny Rule 23 certification on liability. For example, if a telecommunications company were alleged to have erroneously charged many California customers double rates for certain interstate calls, a district court could certify a class of all the company’s California customers even if an expert testified that only 80 percent of them were likely to have made the calls in question. Determining who did, which likely could be done from available records, could be left to a damages stage.

This variation among cases is why we review decisions on class certification for abuse of discretion. Torres, 835 F.3d at 1132. We give the district court “noticeably more deference” when it certifies a class than when it denies certification. Abdulla v. U.S. Sec. Assocs., Inc., 731 F.3d 952, 956 (9th Cir. 2013) (cleaned up). That deference is appropriate because Rule 23 certification is at bottom a trial management decision; it simply allows the class litigation to continue under the district court’s ongoing supervision. The district court retains the power to alter or amend a class certification order at any time before final judgment. Fed. R. Civ. P. 23(c)(1)(C).

I recognize that one of our sister Circuits has suggested that “5% to 6%” is the “outer limit[]” of an acceptable number of uninjured class members. In re Rail Freight Fuel Surcharge Antitrust Litig., 934 F.3d 619, 625 (D.C. Cir. 2019). 2 While disclaiming any particular numerical cap, the majority suggests that something between 5 and 10 percent approaches the outer limit. Op. at 32. But this effectively rewrites Rule 23. If the Supreme Court finds that approach wise, after the usual input and recommendations from the advisory committees, and Congress does not see fit to act to the contrary, then so be it. But we should not legislate from the appellate bench based on our personal concerns with the class action device. Under the Rule as currently written, we should instead leave fact-based decisions on predominance

2 Although the First Circuit has adopted a “de minimis” rule, it has defined it in “functional terms,” asking whether there is a “mechanism that can manageably remove uninjured persons.” Asacol Antitrust Litig., 907 F.3d at 53–54 (cleaned up). That rule corresponds in practical application to Ninth Circuit precedent. See Torres, 835 F.3d at 1137. OLEAN WHOLESALE GROCERY COOP. V. BUMBLE BEE FOODS 41

and case management to the sound discretion of the district courts.

Nor is a “de minimis” rule necessary to address Article III concerns. “[O]nly the representative plaintiff need allege standing at the motion to dismiss and class certification stages.” Ramirez v. TransUnion LLC, 951 F.3d 1008, 1023 (9th Cir. 2020). Class members “must satisfy the requirements of Article III standing at the final stage of a money damages suit when class members are to be awarded individual monetary damages.” Id. at 1017 (emphasis added). To be sure, Torres instructs the district court to “winnow out” uninjured class members, 835 F.3d at 1137, but their presence at the certification stage is not a barrier to standing. Put simply, the de minimis rule is a solution in search of a problem.

III

Defendants may well be correct that Plaintiffs’ data was “methodologically flawed and was unable to show impact for up to 28% of the class.” Op. at 30. And, in the exercise of its discretion, the district court might find that such a large percentage of uninjured class members means that common issues of law or fact do not predominate in this case. But, by the same measure, the district court could find that Plaintiffs’ aggregated proof could establish liability to a predominant portion of the class, and that uninjured members could be identified in future (perhaps non-class) proceedings. Because the majority removes from the district court the broad discretion Rule 23 provides and instead replaces it with a “de miminis” requirement found nowhere in the Rule or our precedents, I respectfully dissent from Part III.B.3 of the majority opinion.