National ATM Council, Inc. v. Visa Inc.

United States Court of Appeals
FOR THE DISTRICT OF COLUMBIA CIRCUIT

No. 21-7109 September Term, 2022
FILED ON: JULY 25, 2023
NATIONAL ATM COUNCIL, INC., ET AL.,
APPELLEES

v.

VISA INC., ET AL.,
APPELLANTS

Consolidated with 21-7110, 21-7111

Appeals from the United States District Court
for the District of Columbia
(No. 1:11-cv-01803)
(No. 1:11-cv-01831)
(No. 1:11-cv-01882)

Before: PILLARD, Circuit Judge, and EDWARDS and ROGERS, Senior Circuit Judges.

JUDGMENT

This case was considered on the record from the United States District Court for the District
of Columbia and on the briefs and oral arguments of the parties. The court has afforded the issues
full consideration and determined they do not warrant a published opinion. See D.C. CIR. R. 36(d).
For the reasons stated in the memorandum accompanying this judgment, it is hereby

ORDERED AND ADJUDGED that the judgment of the district court be AFFIRMED.

Pursuant to D.C. Circuit Rule 36, this disposition will not be published. The clerk is
directed to withhold issuance of the mandate herein until seven days after resolution of any timely
petition for rehearing or rehearing en banc. See Fed. R. App. P. 41(b); D.C. Cir. R. 41(a)(1).

1
Per Curiam

FOR THE COURT:
Mark J. Langer, Clerk

BY: /s/

Daniel J. Reidy
Deputy Clerk

2
National ATM Council, Inc., et al. v. Visa Inc., et al., No. 21-7109

MEMORANDUM

Mastercard and Visa operate networks that enable banks to communicate with automated
teller machines (ATMs) operated by third parties. The Mastercard and Visa networks (CIRRUS
and Plus) are enabled on nearly every debit card in circulation in the United States, and by some
estimates, are used to process over half of all ATM transactions. Mastercard and Visa require that
ATM operators seeking to use their networks agree to rules governing the “access fees” that ATMs
may charge cardholders (the Access Fee Rules). A group of ATM operators, along with debit
cardholders who were charged fees at bank and non-bank ATMs, filed class action lawsuits
challenging those Rules as a violation of antitrust law. Three groups of plaintiffs sought class
certification. The district court granted class certification, holding that each of the three proposed
classes satisfied the requirement under Rule 23(b)(3) that common questions of law or fact
predominate over individual ones. See Nat’l ATM Council, Inc. v. Visa Inc., No. 11-cv-1803, 2021
WL 4099451, at *5-7 (D.D.C. Aug. 4, 2021). We affirm.

I.

Cardholder access fees for ATM transactions were prohibited until the mid-1990s. As
governments rolled back those bans, banks began to charge for ATM use and nonbank operators
entered the ATM market. Banks developed ATM networks that allowed ATM operators—both
different banks and nonbank (independent) ATM operators—to serve debit cardholders more
interchangeably, without regard to whether the ATM is sponsored by the card-issuing bank.
Cardholders gained the convenience of access to their accounts at ATMs other than their bank’s—
for a price. And the businesses involved in providing that access vied for shares of the profits.

In 1996, shortly after legal restrictions on ATM access fees were lifted, Mastercard and
Visa developed the contract provision that is the focus of this antitrust case. Under that provision,
ATM operators that participate in Mastercard’s or Visa’s network must abide by Access Fee Rules
that prohibit them from charging cardholders lower access fees for transactions routed over other
ATM networks than they charge for transactions over a Mastercard or Visa network. In other
words, if an ATM operator is charged less by a non-Mastercard or Visa network, the contract
provision bars that operator from passing a portion of those savings on to cardholders in the form
of lower access fees. It is undisputed for current purposes that Mastercard and Visa exercise
market power such that all ATM transactions at issue in this case are subject to the Access Fee
Rules.

Plaintiffs claim the Access Fee Rules are an unlawful agreement “in restraint of trade” that
violates Section 1 of the Sherman Antitrust Act, 15 U.S.C. § 1. In their view, Mastercard and Visa
have engaged in anticompetitive conduct by preventing bank and nonbank ATM operators alike
from offering cardholders lower access fees for transactions routed over cheaper networks—and
by blocking independent ATM operators from competing with bank-owned ATMs by charging
cardholders less.

1
 To apprehend the claimed antitrust violation underlying this class action suit, one must
understand the standardized structure of a “foreign” ATM transaction—the only kind at issue
here—and the various fees involved. A “foreign transaction” does not refer to anything
international, but to a cardholder’s use of an ATM not operated by the financial institution that
issued the card. A foreign ATM transaction cannot proceed unless the card-issuing bank and the
ATM terminal share an ATM network. In exchange for enabling the transaction, the ATM network
charges the ATM operator a per-transaction usage fee, or “network fee” (also referred to in the
record as an “acquirer fee”). The ATM operator, for its part, relies on two revenue streams: an
“interchange fee” it charges the card-issuing bank for serving its card and a surcharge or “access
fee” it charges each cardholder. The parties also speak in terms of “net interchange”: the per-
transaction amount of interchange revenue left to the ATM operator after it pays the network fee.
As relevant here, high network fees depress ATM operators’ net interchange revenues.

Mastercard and Visa charge higher network fees than their competitors, commonly referred
to as “regional networks.” ATM operators accordingly receive less net-interchange revenue on a
transaction routed over a Mastercard or Visa network than if that same transaction used a regional
network. But Mastercard or Visa networks are enabled on nearly every debit card in the United
States, excluding government-issued electronic benefit transfer cards. More than half of all
domestic ATM withdrawals are routed over Mastercard or Visa networks. Some banks issue debit
cards that can complete transactions only over Mastercard or Visa ATM networks. Thus, an ATM
operator that forgoes access to Mastercard and Visa ATM networks risks being unable to process
transactions for, and earn revenue from, many banks and cardholders. It is unsurprising, then, that
Mastercard and Visa networks have achieved near-universal acceptance at domestic ATMs—
which is to say that the challenged contract provision affects virtually all ATM transactions,
whether or not they are completed over Visa’s or Mastercard’s networks.

Given that Mastercard’s and Visa’s network fees are comparatively high, all else being
equal, independent ATM operators would presumably favor ATM networks offering network
service for less. And one might expect that an ATM operator seeking to expand its customer base
would seek to attract customers whose transactions could be routed over lower-cost networks. An
ATM operator could attract customers by engaging in “differential surcharging,” that is, charging
higher access fees for transactions routed over higher-cost networks like Visa’s and
Mastercard’s—and sharing savings from use of regional networks with cardholders served via
those lower-cost networks. If lower access fees were not barred by the Access Fee Rules, demand
from customers seeking lower access fees, including those with cards currently limited to
Mastercard’s and Visa’s networks, would presumably drive card-issuing banks to offer them via
connections to networks other than Mastercard’s and Visa’s.

In sum, the ATM customer plaintiffs would benefit if independent ATM operators offered
them account access for less. ATM Operator plaintiffs would benefit if they could earn more per
transaction through network-fee savings, or if they could attract more customers and increase their
volume of transactions by differentially charging lower ATM access fees to process transactions
over less expensive ATM networks. But, Plaintiffs argue, the Access Fee Rules challenged here
foreclose those benefits, obstruct competition, and cause them class-wide harm.

2
 II.

Three plaintiff groups filed parallel class-action suits against Mastercard and Visa
(Defendants) claiming that the Access Fee Rules restrain competition in violation of Section 1 of
the Sherman Act. The ATM Operator Plaintiffs are entities who operate independent ATMs,
Burke Plaintiffs are cardholders who were charged unreimbursed access fees at independent
ATMs, and Mackmin Plaintiffs are cardholders who were charged unreimbursed access fees at
bank-operated ATMs. ATM Operators, Burke Plaintiffs, and Mackmin Plaintiffs (collectively,
Plaintiffs) allege that cardholders pay inflated access fees because of Defendants’ anticompetitive
restrictions on ATM operators, and that those same restrictions prevent independent operators from
using cardholder access-fee discounts to expand their own market share.

The district court consolidated the cases, and each plaintiff group moved for class
certification. As relevant here, Defendants opposed class certification on the ground that Plaintiffs
had not shown common antitrust injury and thus failed to meet the predominance requirement of
Rule 23(b)(3). As we observed in a prior appeal of these cases, the three plaintiff groups advance
distinct but complementary theories of antitrust harm. See Osborn v. Visa, 797 F.3d 1057, 1064
(D.C. Cir. 2015). ATM Operators claim that the Access Fee Rules’ ban on differential surcharging
enables Defendants to maintain artificially high network fees that eat into the ATM Operators’
revenue. They further claim that the Rules protect Visa’s and Mastercard’s high network fees from
downward pressure that might otherwise be exerted if independent ATMs could compete for
customers by charging lower access fees to use their machines to complete transactions over lower-
cost networks. According to the Burke and Mackmin plaintiffs, the artificially high network fees
lead ATM operators to preserve profits by charging higher access fees to the cardholder class
members than they would otherwise. Burke Plaintiffs aim to show that Visa’s and Mastercard’s
inflated network fees have price effects akin to those of a tax on the independent ATM operator
industry—that is, they place upward pressure on the final price of the service, raising the access
fee paid by the cardholder at the ATM terminal. And Mackmin Plaintiffs seek to illustrate that
effect with additional evidence that high network fees eat into the net revenue that bank ATM
operators receive from other banks, which leads ATM operators to charge higher access fees.

The district court granted the motions and certified three classes, finding that each satisfied
the requirements of Rule 23(a) and Rule 23(b)(3). Defendants timely sought interlocutory review
of the district court’s class certification ruling. We review that ruling for abuse of discretion. See
Garcia v. Johanns, 444 F.3d 625, 631 (D.C. Cir. 2006). Under this standard, “we review a
certification ruling ‘conservatively’” to guard against “erroneous application of legal criteria” or
unsupported assessments of evidence and will “affirm the district court even if we would have
ruled differently in the first instance.” Id. (quoting Wagner v. Taylor, 836 F.2d 578, 586 (D.C.
Cir. 1987)); see also Highmark Inc. v. Allcare Health Mgmt. Sys., Inc., 572 U.S. 559, 563 n.2
(2014) (“A district court would necessarily abuse its discretion if it based its ruling . . . on a clearly
erroneous assessment of the evidence.” (quoting Cooter & Gell v. Hartmarx Corp., 496 U.S. 384,
405 (1990))).

Mastercard and Visa argue that the district court failed to give a “hard” or “close look” at
Plaintiffs’ evidence or to perform a “rigorous analysis” as to whether Plaintiffs met the
predominance requirement. Appellants’ Br. 18, 20 (quoting In re Rail Freight Fuel Surcharge
3
Antitrust Litig.-MDL No. 1869 (Rail Freight I), 725 F.3d 244, 255 (D.C. Cir. 2013) and Comcast
Corp. v. Behrend, 569 U.S. 27, 34-35 (2013)). And they assert that the district court “erroneously
deferred the issue of classwide injury to the merits stage.” Id. at 19; see id. at 30-31. They contend
that the court thus abused its discretion in finding that Plaintiffs meet the Rule 23(b)(3) requirement
that “questions of law or fact common to the class predominate over any questions affecting only
individual members.” Id. at 19 (quoting Rail Freight I, 725 F.3d at 249 (citing Fed. R. Civ. P.
23(b)(3))). Defendants assert that the evidence indisputably shows each class to contain uninjured
members, id. at 36-41, and they fault Plaintiffs for not identifying a common “winnowing
mechanism” to exclude them, id. at 42; see id. at 42-45.

Put simply, Mastercard and Visa contend that the district court did not reach its class
certification ruling through rigorous analysis, and that Plaintiffs do not satisfy Rule 23(b)(3). We
disagree and remand to the district court for further proceedings consistent with the order.

III.

A.

As a threshold matter, we confirm our discretionary jurisdiction over this appeal. See Fed.
R. Civ. P. 23(f); 28 U.S.C. § 1292(e). In this circuit, we have jurisdiction over an interlocutory
appeal when we deem it an appropriate exercise of our discretion to take an early look at a class
certification decision that (1) is “questionable” and accompanied by a “death-knell situation for
either the plaintiff or defendant that is independent of the merits of the underlying claims,” (2)
“presents an unsettled and fundamental issue of law relating to class actions,” (3) is “manifestly
erroneous,” or (4) presents other “special circumstances” warranting immediate review. In re
Lorazepam & Clorazepate Antitrust Litig., 289 F.3d 98, 99-100, 105 (D.C. Cir. 2002).

The certification decision does not pose an important and unsettled, class action-related
legal question that we must resolve. Nor does it show manifest error by ignoring binding, on-point
precedent. Manifest error in class certification decisions is “rare[] . . . simply because class actions
typically involve complex facts that are unlikely to be on all fours with existing precedent.” In re
Johnson, 760 F.3d 66, 72 (D.C. Cir. 2014) (quoting Chamberlan v. Ford Motor Co., 402 F.3d 952,
962 (9th Cir. 2005)); see also In re Brewer, 863 F.3d 861, 875 (D.C. Cir. 2017) (noting that “[t]he
manifest error standard is extremely difficult to meet”).

We nonetheless exercise our jurisdiction on the ground that the district court’s certification
decision is at least “questionable” insofar as its statements of law were not entirely clear, its
citations were not current, and its record analysis was notably terse. See Nat’l ATM Council, Inc.,
2021 WL 4099451, at *1-7. That is surprising and unfortunate given the breadth of the asserted
unlawful practice, the amount of money asserted to be at stake, and the voluminous record
including dueling expert reports. The district court quoted older, nonbinding district court
decisions, and failed to cite the Supreme Court’s most recent case analyzing when common issues
predominate over individualized ones under the pertinent class action provision. See id. at *5-7
(citing In re Vitamins Antitrust Litig., 209 F.R.D. 251, 264 (D.D.C. 2002) and Meijer, Inc. v.
Warner Chilcott Holdings Co. III, 246 F.R.D. 293, 307 (D.D.C. 2007)). The district court’s
pronouncement that plaintiffs “need only demonstrate a colorable method by which they intend to

4
prove class-wide impact” and its citations to decades-old, nonbinding cases, see id. at *6 (citing
In re Vitamins, 209 F.R.D. at 264), arguably are in tension with our recent guidance that Rule 23
“commands” the court to take a “hard look at the soundness of statistical models that purport to
show predominance,” Rail Freight I, 725 F.3d at 255.

The certification decision also could be the “death knell” of the litigation by focusing
“irresistible pressure” on Defendants to settle rather than continue to litigate to a judgment on the
merits. Id. at 251 (quoting Prado-Steiman ex rel. Prado v. Bush, 221 F.3d 1266, 1274 (11th Cir.
2000)). Indeed, several of Mastercard’s and Visa’s codefendants have already settled. Order
Granting Motion for Final Approval of Settlements, ECF No. 261, Mackmin v. Visa Inc., No. 11-
cv-01831 (D.D.C. Aug. 8, 2022). If they were to lose on the merits, Defendants would face treble
damages under the Clayton Act, an amount that they assert would exceed their annual net income.
Such relatively large exposure “push[es] litigants inexorably toward settlement” and satisfies the
death knell requirement. Rail Freight I, 725 F.3d at 252. The questionable accuracy of unclear
language in the district court’s opinion combines with the “death knell” settlement pressure to
warrant our exercise of interlocutory appellate jurisdiction.

B.

Defendants’ sole challenge to the class certification order is that the district court abused
its discretion in holding that common questions predominate over individualized ones. A district
court may certify a class only if it is convinced “after a rigorous analysis” that the proposed class
can “satisfy through evidentiary proof” the requirements of Rule 23(a) and at least one provision
of Rule 23(b). Comcast, 569 U.S. at 33-34; see Gen. Tel. Co. v. Falcon, 457 U.S. 147, 160-61
(1982). The district court certified these classes under Rule 23(b)(3), which requires 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). That requirement is satisfied when questions
calling for individualized determination are outweighed by questions that are “capable of
classwide resolution—which means that determination of [their] truth or falsity will resolve an
issue that is central to the validity of each one of the claims in one stroke.” In re Rail Freight Fuel
Surcharge Antitrust Litig.-MDL No. 1869 (Rail Freight II), 934 F.3d 619, 622 (D.C. Cir. 2019)
(quoting Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338, 350 (2011)).

The district court’s legal analysis is brief but materially correct. The court’s certification
order does not rest on an incorrect legal standard. And we discern no substantive inadequacy or
other legal error in the district court’s evidentiary assessment. We accordingly affirm.

1.

Mastercard and Visa object that the district court failed to “take a hard look at whether
plaintiffs’ injury models can prove classwide injury through common proof,” as binding precedent
requires; they assert that the court instead pretermitted the requisite legal scrutiny by “improperly
deferring it to the ‘merits’ stage.” Appellants’ Br. 14; see id. at 19. The district court must state
and apply the correct legal standard, and failure to do so would itself constitute an abuse of
discretion requiring reversal and remand. See Kickapoo Tribe of Indians of Kickapoo Reservation
in Kan. v. Babbitt, 43 F.3d 1491, 1497 (D.C. Cir. 1995). We conclude that Defendants’ assertions

5
that the district court applied an incorrect legal standard are unfounded.

To support certification based on predominance, plaintiffs’ evidence must show their
claims are susceptible to common proof. The rigorous analysis a district court is required to
conduct in support of its certification order sometimes involves probing the merits of plaintiffs’
claims, which the Supreme Court has made clear is permissible insofar as necessary to ensure that
the Rule 23 requirements are met. Amgen Inc. v. Conn. Ret. Plans & Tr. Funds, 568 U.S. 455, 466
(2013). That said, plaintiffs need only genuinely contest, not definitively rule out, defendants’
alternative ways that a reasonable factfinder might view the evidence. Tyson Foods, Inc. v.
Bouaphakeo, 577 U.S. 442, 453 (2016). In Amgen Inc. v. Connecticut Retirement Plans and Trust
Funds, for example, the Supreme Court held that plaintiffs in a securities fraud class action seeking
to show predominance need not prove at the class-certification stage that an alleged
misrepresentation did in fact materially affect all class members’ stock price, even though
plaintiffs “certainly must prove materiality to prevail on the merits.” 568 U.S. at 459. As the Court
explained, “Rule 23(b)(3) requires a showing that questions common to the class predominate, not
that those questions will be answered, on the merits, in favor of the class.” Id.

The district court applied the correct legal standard here when it concluded that Plaintiffs
had carried that burden, explaining that “[u]pon careful review of the parties’ submissions,
including their expert reports,” it found that “there is sufficient common evidence of resulting
injuries and the amount of those injuries” to satisfy Rule 23(b)(3). Nat’l ATM Council, Inc., 2021
WL 4099451, at *5. The court expressly acknowledged its obligation to “give careful scrutiny to
the relation between common and individual questions” and correctly recognized that “a common
question is one where the same evidence will suffice for each member to make a prima facie
showing [or] the issue is susceptible to generalized class-wide proof.” Id. (quoting Tyson Foods,
577 U.S. at 453) (internal quotation marks omitted). The court aptly distinguished Plaintiffs’
burden of establishing predominance at the class certification stage from their burden of prevailing
on liability and damages at trial: The court recognized that common evidence sufficient for class
certification need not conclusively establish class-wide liability and damages, but that Plaintiffs
must present creditable evidence from which questions common to the class members’ claims
could be resolved at trial in one stroke. Id. at *5-6 (citing Amgen, 568 U.S. at 459, and Rail Freight
II, 934 F.3d at 622-23).

The Supreme Court has rejected the idea that “any method of measur[ing injury] is acceptable
so long as it can be applied classwide, no matter how arbitrary the measurements may be.”
Comcast, 569 U.S. at 36; accord Rail Freight I, 725 F.3d at 253-54. Since Comcast, this court has
explained that “[c]ommon questions of fact cannot predominate where there exists no reliable
means of proving classwide injury in fact.” Rail Freight I, 725 F.3d at 252-53 (emphasis added).
In keeping with these precedents, the district court here approved Plaintiffs’ evidentiary case as
“reasonable” and based on “well established,” “well-accepted methodolog[ies]” showing class-
wide injury. Nat’l ATM Council, Inc., 2021 WL 4099451, at *6.

Read in context, the district court’s comment that “plaintiffs, at this stage in the proceedings,
need only demonstrate a colorable method by which they intend to prove class-wide impact,” id.
(citing In re Vitamins, 209 F.R.D. at 264), does not undercut the soundness of the court’s reasoning.
Defendants highlight the term “colorable” as contrary to more recent rulings demanding that
6
district courts take a “hard look” at models purporting to show class-wide injury. They read the
district court’s use of the word “colorable” as adopting a “lenient standard” that runs counter to
Rule 23 and binding precedent. See Reply Br. 22. But in context, the district court appears to
have used “colorable” to denote not merely non-frivolousness, but evidence “appearing to be true,
valid, or right.” See Colorable, BLACK’S LAW DICTIONARY (11th ed. 2019). We do not read the
district court to indulge the kind of “arbitrary” or frivolous method of showing class injury that
Comcast warns against. 569 U.S. at 35; see id. at 35-36. As described below, the court’s review
of the evidence comports with Supreme Court holdings that require district courts to closely review
the record at the class certification stage—and thereby to ensure that plaintiffs provide a reliable
method for establishing class-wide injury, Rail Freight I, 725 F.3d at 252-53, that is tied to
plaintiffs’ theory of liability, Comcast, 569 U.S. at 35-36, and complies with our precedent
requiring that statistical models offered by plaintiffs “show all class members suffered some
injury,” Rail Freight I, 725 F.3d at 252; see also Rail Freight II, 934 F.3d at 623-25. In its analysis,
the district court confirmed that each group of plaintiffs cleared those hurdles. That does not
constitute reversible error.

2.

The district court did not cite Comcast, but it did adhere to its demands. The court’s
certification decision explains the basis on which it concluded that plaintiffs here “satisfy through
evidentiary proof” that common issues predominate. Comcast, 569 U.S. at 33-34. And the record
supports the district court’s finding that “all three plaintiff groups have demonstrated that common
evidence will predominate in proving each element of their claims.” Nat’l ATM Council, Inc.,
2021 WL 4099451, at *5. The court confirmed not only that Plaintiffs offered common proof of
injury, but also that their methods of establishing injury were reasonable, well accepted, and
reliable. We consider the district court’s approval of the showings of the Burke, Mackmin, and
ATM Operator Plaintiffs in turn.

a.

The district court acted within its informed discretion in concluding that the Burke
Plaintiffs put forth a “reasonable,” “well established” methodology that “provides for class-wide
resolution” of injury and damages. Id. at *6. Its analysis of the evidence comports with Comcast.
The district court based its predominance determination on the Burke expert’s opinion, drawn from
“significant academic literature,” that a supra-competitive network fee economically affects the
price of the service as would an industry-wide “tax.” Id. According to the Burke Plaintiffs’ theory,
that “tax” is “fully incorporated into industry-wide prices,” meaning that independent ATM
operators pass on a portion of that overcharge to direct consumers—the cardholders who use
independent ATMs. Id. The district court also appropriately concluded that the Burke expert’s
“overcharge damages model,” which tracks Burke Plaintiffs’ industry-wide tax theory, “us[es]
well-accepted methodology” to provide evidence of class-wide injury. Id.

In antitrust litigation, when a direct consumer argues that an intermediary business has
passed on an anticompetitive overcharge, “the typical way in which passed-on damages are
computed” is “by either the ‘yardstick’ method or the ‘before-and-after’ method.” PHILIP E.
AREEDA & HERBERT HOVENKAMP, ANTITRUST LAW ¶ 346k1 (5th ed. 2022) (footnotes omitted).
7
Yardstick analysis computes antitrust damages by comparing the prices paid by plaintiffs with
those paid by a consumer in a comparable market unaffected by the antitrust violation. Id. This
generally accepted method of calculating passed-on overcharges quantifies consumer harm
without reference to the precise amount of overcharge an intermediary passes on to a direct
purchaser. See id.

The Burke Plaintiffs’ expert, Dr. William Lehr, selected Puerto Rico as a yardstick. The
Burke Plaintiffs explained that he was unable to find other real-world examples because the ATM
Access Fee Rules have been in place in the United States for nearly as long as state and federal
governments have permitted ATM operators to charge access fees. The expert report identified
Puerto Rico as providing a useful real-world approximation of an otherwise comparable market
free from Mastercard’s and Visa’s alleged anticompetitive conduct, because in Puerto Rico
differential surcharging is permitted as between domestic and cross-border transactions. After
conducting an empirical analysis of the Puerto Rican ATM market, the expert concluded that in
the but-for world that Puerto Rico approximates, ATM operators charge lower access fees to
cardholders who initiate transactions that cost the ATM operator less to process. But because the
ATM Access Fee Rules forbid that practice in the United States, the expert reasoned, cardholders
in the Burke class pay more to withdraw money from independent ATMs than they otherwise
would. The expert also developed an overcharge damages model based on an industry-wide tax
theory. That model posited a one-to-one inverse relationship between cardholder access fees and
net interchange, which the expert’s analysis of the Puerto Rican ATM market corroborated.

Courts have deemed the kind of yardstick analysis performed by the Burke expert
appropriate in antitrust cases in which it is not possible to isolate the effect of assertedly
anticompetitive conduct by performing a before-and-after comparison. See, e.g., MM Steel, L.P.
v. JSW Steel (USA) Inc., 806 F.3d 835, 850-52 (5th Cir. 2015) (affirming district court’s decision
to admit as “reliable” expert testimony that used yardstick analysis to assess antitrust damages
when newly formed plaintiff “had no financial performance [data] to use as the ‘before’ in the
before and after test”); see also Zenith Radio Corp. v. Hazeltine Rsch., Inc., 395 U.S. 100, 114,
124-25 (1969) (holding it was reasonable for the district court to infer a causal relationship between
anticompetitive conduct and antitrust injury based on comparison of its performance in the
Canadian television market subject to anticompetitive practice with the U.S. market absent the
challenged practice). We have no basis to dispute the district court’s finding that this long-
accepted approach to establishing and quantifying injury in consumer overcharge cases was
reliable. See Conwood Co. v. U.S. Tobacco Co., 290 F.3d 768, 792-93 (6th Cir. 2002) (holding
district court did not abuse discretion by finding “sufficiently reliable” expert testimony that used,
inter alia, “generally accepted” “yardstick” method of proving antitrust damages).

Defendants assail the Burke Plaintiffs’ evidence of predominance by arguing that the
Defendants’ expert’s regression—developed as an alternative to the Mackmin Plaintiffs’
regression—identifies some independent ATM operators that did not raise access fees when their
net interchange fell. In other words, Defendants’ regression, which analyzed bank transactions
between 2010 and 2015, did not show all independent ATM Operators increasing cardholder
access fees in lockstep with rising network fees.

Defendants’ focus on evidence of pass-through misunderstands the Burke Plaintiffs’
8
theory. As the district court recognized, the Burke Plaintiffs’ theory of injury is that all class
members are deprived of the competitive opportunities and lower prices available to “consumers
in the but-for world,” in which anticompetitive obstacles are absent. See Nat’l ATM Council, Inc.,
2021 WL 4099451, at *4. In line with that theory, the Burke Plaintiffs’ evidentiary submissions
compare prices paid in the “real world” with the lower prices that would be paid in a world free
from the Access Fee Rules. This type of yardstick analysis does not depend on proof that an ATM
operator passes on the cost of the network fee to the ATM user in every transaction. See AREEDA
& HOVENKAMP ¶ 346c (explaining that yardstick analysis “do[es] not require computing the
overcharge passed on at each stage”).

Defendants’ real-world regression does not engage at all with Plaintiffs’ evidence that the
Burke class members paid more than they would have in a world with no ATM Access Fee Rules.
The Burke Plaintiffs’ evidence, which a reasonable factfinder could credit, shows that all class
members were injured. The Defendants’ expert analysis of real-world data, chosen without
accounting for how that data may be tainted by Defendants’ alleged anticompetitive conduct, may
raise some material issue for trial, but it cannot defeat predominance.

b.

The district court acted well within its discretion in also holding that the Mackmin Plaintiffs
had adduced class-wide evidence of antitrust injury. The Mackmin class did so through statistical
modeling, together with “defendants’ own documents support[ing] the relationship between net
interchange and surcharges,” “economic theory and empirical studies,” and “market structure
analysis.” Nat’l ATM Council, Inc., 2021 WL 4099451, at *6. That evidence showed that “all or
virtually all class members . . . pay higher [access fees]” than they would without the Access Fee
Rules. Id.

Mackmin Plaintiffs’ expert, Dr. Dennis W. Carlton, drew on academic literature to support
his conclusion that supra-competitive network fees made possible by the ATM Access Fee Rules
lead to increased cardholder access fees. Referencing peer-reviewed studies, Dr. Carlton observed
that “[c]omplete pass-through of an industry increase in costs [to consumers] is often expected and
found empirically.” J.A. 3098 n.96 (Carlton Expert Report). He also noted that complete pass-
through “is not necessary for my analysis,” since “[t]he key is that an increase in industry costs is
associated with an increase in retail prices, not whether that increase is precisely of the same
amount as the cost increase.” J.A. 3098 n.96 (Carlton Expert Report). In further support of
Mackmin Plaintiffs’ theory of injury, Dr. Carlton ran a regression analyzing changes in average
net interchange and cardholder access fees at Wells Fargo ATMs between 2010 and 2017 and
found a statistically significant relationship between net interchange and access fees. J.A. 3110
(Carlton Expert Report). Finally, Dr. Carlton highlighted Defendants’ statements corroborating
that relationship, such as Mastercard’s statement in a 2017 presentation that “[a]s interchange
decreases ATM operators increase their [access] fees.” J.A. 3105 (Carlton Expert Report) (quoting
J.A. 1930 (Berman Decl., Ex. 1, Mastercard Presentation 6)). Those common sources of proof all
evidence harm across the entire Mackmin class.

Defendants’ expert responded to the Mackmin Plaintiffs’ evidence with his own regression
analysis of data for the top 100 banks by volume of transactions but limited his analysis to one of
9
Visa’s ATM networks over a five-year period from 2010 to 2015. See J.A. 6675-76, 6764
(Hubbard Expert Report). Based on his alternative regression, Defendants’ expert concluded that
for 15 of the 100 banks the data did not show an inverse and statistically significant relationship
whereby lower average net interchange correlated with higher average cardholder access fees. See
J.A. 6677 (Hubbard Expert Report); Appellants’ Br. 34-35. This separate expert analysis does not
mean that the Mackmin Plaintiffs’ model is itself either necessarily flawed or incapable of
establishing class-wide injury.

The district court at the class certification stage did not focus on which of the conflicting
models to credit. Rather, the court correctly considered whether the Mackmin Plaintiffs offered
reliable, generalized proof of injury that a reasonable factfinder could credit and that, if credited,
would enable resolution of class claims without piecemeal proof. Nat’l ATM Council, Inc., 2021
WL 4099451, at *6; see Tyson Foods, 577 U.S. at 454-55 (affirming that “[a] representative or
statistical sample, like all evidence, is a means to establish or defend against liability” in class and
individual cases alike).

In sum, the district court saw that Defendants “attack primarily whether plaintiffs’ methods
will in fact prove injury and damages for each class-member,” not whether the elements of
Plaintiffs’ claims are generally susceptible to class-wide proof with the evidence Plaintiffs have
presented. Nat’l ATM Council, Inc., 2021 WL 4099451, at *6 (emphasis added). A defendant
does not defeat predominance simply by offering a contrary statistical model that the ultimate
factfinder might or might not embrace.

c.

The district court also acted within its sound discretion in holding that the third plaintiff
class, the ATM Operators, had shown predominance. The court explained that the ATM Operators
presented a logical method to arrive at a “reliable estimate” of class-wide harm, showing “that
individualized inquiries would not be necessary to ascertain the fees paid by each class member.”
Nat’l ATM Council, Inc., 2021 WL 4099451, at *6. This cuts to the heart of the predominance
requirement: avoiding individualized mini-trials. Rail Freight II, 934 F.3d at 625. Common
questions cannot “predominate over any questions affecting only individual members,” Fed. R.
Civ. P. 23(b)(3), if the district court must undertake individualized inquiries to sort injured from
uninjured class members, see Rail Freight I, 725 F.3d at 252. The district court reasonably
concluded that the ATM Operators presented common evidence of injury to all class members,
and thus need not identify any mechanism to weed out from the class uninjured members that
Defendants contend their evidence would prove if it were credited over the ATM Operators’
evidence.

Defendants argue that some members of the ATM Operator class did not pay fees for use
of Mastercard’s and Visa’s networks, whether directly or indirectly, and are therefore uninjured.
According to Defendants, three types of ATM Operator class members might not have paid
network fees: (1) ATM Operators that do not appear by name in Mastercard’s and Visa’s billing
records, Appellants’ Br. 42-43; (2) entities that do not “receive or pay any components of net
interchange,” because a third party, such as an independent sales agent, is entitled to receive the
interchange revenue from which network fees are subtracted, id. at 44; and (3) entities “involved
10
in operating and servicing independent ATMs,” such as a firm that “manages the initial [ATM]
setup and holds encryption keys,” or “a retail store [that] may own an ATM on its premises” but
“contract[s] with an independent sales organization to connect its ATM to banks and payment
networks,” id. at 43-44.

None of Defendants’ arguments defeats predominance. First, Mastercard and Visa do not
maintain billing accounts that separately list all ATM Operators, because independent ATM
operators generally access Mastercard and Visa networks through sponsoring institutions that pay
network fees on their behalf. Thus, one would not expect to find ATM Operator class members
identified by name in Defendants’ billing records. Mastercard does not maintain a dataset that
links ATM Operators to the sponsoring banks that pay network fees on ATM Operators’ behalf.
Moreover, Visa admits that some of the billing records it disclosed in discovery are “not entirely
comprehensive (and become[] less so further back in time).” J.A. 664 (Zona Expert Report).

Second, ATM Operator class members who assign a portion of their payment obligations
and revenues to other entities are no less injured by illegal practices that diminish the revenue. If,
for example, an ATM Operator assigns some (or all) of its net interchange revenue to a premises
owner, it is effectively making a rental payment. Whether an ATM Operator takes the intermediate
step of depositing any net interchange revenue into an account before drawing on it to pay bills is
not determinative of whether the ATM Operator suffers harm.

Third, entities that manage ATM setup, service providers that handle ATM encryption
keys, and retail stores that own on-site ATMs but do not originate transactions are excluded from
the Class Definition. The definition provides that “[p]ersons or entities that make space available
to ISOs or affiliates of ISOs to operate a Qualified ATM on property they own or control” (the
retail store) and “Encryption and Support Organizations that manage encryption keys or service
Qualified ATMs” (an entity that holds the encryption keys) “are not ATM Operators.” See J.A.
521 (Operators’ Mot. for Class Cert.) (defining Class Definition terms); J.A. 588 (Am. Order
Granting Class Cert.) (defining class subject to limitations and definitions set forth in Operators’
class certification motion). Any failure to show harm to them cannot detract from the
predominance of questions common to the members of the class as actually defined.

The district court acted within its discretion in determining that the record includes
common evidence of injury to all members of the ATM Operator class. Defendants’ contention
that predominance is defeated by the lack of a mechanism for weeding out uninjured class
members depends on a factfinder crediting their submission that such members exist. On this
record, however, nothing requires a conclusion that the ATM Operator class includes uninjured
members. The district court’s predominance determination is thus unaffected by lack of a common
method to identify uninjured members. The district court’s holding that plaintiffs’ evidence
showed the requisite predominance was thus legally correct and supported by the record.

3.

Finally, the district court’s order does not bear the hallmarks of class certification decisions
that we or the Supreme Court have invalidated for lack of rigor. Contrary to Defendants’
assertions, the district court did not rely in support of class certification on a statistical model that

11
is inconsistent with plaintiffs’ theory of injury, as in Comcast. See 569 U.S. at 35. It did not rely
on a statistical model offered by plaintiffs that detects injury where all agree none exists, as in Rail
Freight I. See 725 F.3d at 253-54. Nor did the court accept as common evidence of injury a
statistical model that, on its own terms, identifies a high percentage of uninjured class members,
as in Rail Freight II. See 934 F.3d at 623-24. Defendants fail to fit this case within the precedents
on which their opposition to class certification rests.

Mastercard’s and Visa’s attempts to show error depend instead on extension of our
precedents in unsupported and unworkable ways. Defendants challenge the Mackmin and Burke
Plaintiffs’ evidence by offering their own expert’s reconfiguration of the Mackmin statistical
model. That challenge does not point to internal flaws in either the Mackmin Plaintiffs’ statistical
model or the Burke Plaintiffs’ distinct industry-wide tax model. Defendants have simply offered
their own, contrasting model and argued that because their model treats some members in the
Mackmin and Burke classes as uninjured, predominance is just as lacking here as, for example, in
Rail Freight I or II. But the differences just noted between this case and the precedents are
decisive.

Defendants assert that the Wells Fargo dataset that the Mackmin Plaintiffs’ expert analyzes
is unrepresentative, and that Defendants’ expert analysis of a different dataset relating to 100 banks
is more trustworthy. But, by its own terms, Plaintiffs’ evidence shows that all class members were
injured. And Plaintiffs, like Defendants, have a theory that a factfinder could credit as to why their
data selection is superior. Defendants’ contention that their model showing unharmed members
is more accurate and credible than Plaintiffs’ different models showing that all members were
harmed is thus precisely the kind of material factual dispute that “is better suited for adjudication
of plaintiffs’ injury and damages on the merits.” Nat’l ATM Council, Inc., 2021 WL 4099451, at
*6; see also Tyson Foods, 577 U.S. at 457 (explaining that a defense that common proof is
unrepresentative “is itself common to the claims made by all class members” and so does not defeat
predominance).

The district court was not required at class certification to make the ultimate determination
which of two dueling experts to accept, and no party here argues that it would be either necessary
or appropriate to do so at this stage on this record. See Oral Argument Tr. 42:1-4; 58:14-16. We
find no error in the district court’s conclusion that the Burke, Mackmin, and ATM Operator
Plaintiffs satisfied Rule 23(b)(3) by providing reasonable, wholesale methodologies, tethered to
Plaintiffs’ respective theories of liability, showing that all class members suffered injury.

***

For those reasons, we affirm the district court’s class certification decision.

12