Wednesday, May 2, 2018

Reconciling burden of proof in antitrust cases involving multi-sided platforms with judicial precedent in Bazemore v. Friday & Sobel v. Yeshiva


Ted Tatos 
Empirical Analytics, LLC

The treatment of multi-sidedness in antitrust cases has been the subject of recent attention in the Ohio v. American Express matter.  The point of this article is neither to reiterate nor to summarize the arguments made by both sides and detailed extensively in the parties’ legal briefs as well as those of amici, but rather to address a specific argument regarding burden of proof.  Briefly, multi-sided platforms serve as intermediaries between two or more agents where indirect network externalities exist between them.  In other words, such platforms bring together groups of agents that can mutually benefit each other, but high transaction costs make their interaction prohibitively expensive.  The intermediary platform allows them to interact without incurring such costs.  A typical example offered is that of video game platforms, which bring together players and video game developers.  Players benefit from more developers on a platform, and video game developers benefit by creating games for platforms that have many players.  On this point, Rochet and Tirole (and subsequently, Evans and Schmalensee) have drawn the distinction between membership and usage externalities.

The debate surrounding the definition of multi-sided markets in antitrust focuses on two issues:

1) should the effects of a restraint on one side of the market be measured against potential pro-competitive effects on the other side, or is a showing of antitrust injury on one side sufficiently probative of anticompetitive conduct, and

2) should the Plaintiffs bear the burden of proof of showing that the restraint’s net effect across all sides is anti-competitive, or should the burden shift to the Defendant after the Plaintiff shows antitrust injury on one side? 

This commentary deals with the latter.  I argue that levying the entire burden of proof on the Plaintiff would be inconsistent with judicial reasoning and precedent set forth both in the Supreme Court’s seminal Bazemore v. Friday (478 US 385 – 1986) decision and the Second Circuit’s decision in Sobel v. Yeshiva (Sobel v. Yeshiva University, 839 F. 2d 18 – 1988).  Both cases deal with issues involving regression analysis, but the reasoning therein applies equally to the burden of proof in multi-sided markets.

In Ohio v. Amex, the Second Circuit found that:

Under the direct method of proving by the rule of reason that Amex violated § 1, Plaintiffs bore the initial burden to show that Amex’s NDPs have ‘an actual adverse effect on competition as a whole in the relevant market.’ K.M.B., 61 F.3d at 127…the Plaintiffs’ initial burden was to show that the NDPs made all Amex consumers on both sides of the platform—i.e., both merchants and cardholders— worse off overall…Without evidence of the net price affecting consumers on both sides of the platform, the District Court could not have properly concluded that a reduction in the merchant‐discount fee would benefit the two‐sided platform overall.

The 2nd Circuit’s decision effectively placed the evidentiary burden for demonstrating anticompetitive conduct in the entire market on the Plaintiff.  To understand why this represents such an enormous burden, consider the NCAA’s economics expert, Prof. Kenneth Elzinga’s view of the college or university market in the Grant-in-Aid Cap Litigation matter, as demonstrated by the figure in his expert report.


Source: Report of Kenneth Elzinga, March 21, 2017, In Re NCAA Grant-in-Aid Cap Antitrust Litigation.

Though in this matter, the parties stipulated to the definition of the market as one-sided, and the Court excluded the testimony of Prof. Elzinga as irrelevant to the remaining issues in the case, his report nevertheless informs the burden Plaintiffs would face under logic imposed by the Second Circuit in Ohio v. Amex.  Plaintiffs would have to investigate potential anticompetitive conduct across the entire market covered by all the various agents involved in the multi-sided platform. As Prof. Elzinga’s figure demonstrates, this burden could be considerable, if not insurmountable, particularly where platforms with more than two sides are involved. Moreover, this astronomic burden is inconsistent with judicial precedent in cases involving similar logical reasoning.

The Chicago School view of consumer welfare as the overarching aim of antitrust jurisprudence holds sway under current judicial orthodoxy.  Consumer welfare, however, is a latent variable. Since it cannot be observed directly, we measure it through inferences drawn from other observable variables that bear a theoretical nexus to consumer welfare. Chief among these variables is output.  The total consumer welfare on a multi-sided platform may be expressed as a function of the effects on individual participating agents (sides). Of course, the measurement involves some degree of error, as the standard is reasonable, not absolute, certainty.  

In this regard, the estimation of the platform-wide consumer welfare effects from a restraint imposed on one or more agent(s) in the platform can be expressed in the same fashion as one would express a regression equation.  There is nothing novel about this approach, as econometric evidence of consumer welfare effects on one side is commonly presented in litigation as indicative of anticompetitive conduct.  But, suppose we were to extend this to the whole market, as the Second Circuit indicated in Ohio v. Amex.  The 2nd Circuit's logic appears to imply that such an equation must include all measurable variables that may affect total consumer welfare.  This approach runs directly counter to the Supreme Court’s seminal decision in Bazemore v. Friday.

In reversing the 4th Circuit Court of Appeals, Justice Brennan delivered the Supreme Court’s unanimous opinion in Bazemore v. Friday, stating:

The Court of Appeals erred in stating that petitioners' regression analyses were "unacceptable as evidence of discrimination," because they did not include "all measurable variables thought to have an effect on salary level." The court's view of the evidentiary value of the regression analyses was plainly incorrect. While the omission of variables from a regression analysis may render the analysis less probative than it otherwise might be, it can hardly be said, absent some other infirmity, that an analysis which accounts for the major factors" must be considered unacceptable as evidence of discrimination." Importantly, it is clear that a regression analysis that includes less than "all measurable variables" may serve to prove a plaintiff's case.

The Bazemore Court also emphasized Defendants’ burdens in responding to Plaintiffs’ evidence:

Respondents' strategy at trial was to declare simply that many factors go into making up an individual employee's salary; they made no attempt that we are aware of - statistical or otherwise - to demonstrate that when these factors were properly organized and [478 U.S. 385, 404] accounted for there was no significant disparity between the salaries of blacks and whites.

The Second Circuit, in its Sobel v. Yeshiva decision, leaned on the Supreme Court’s finding in Bazemore v. Friday, opining that,

We read Bazemore to require a defendant challenging the validity of a multiple regression analysis to make a showing that the factors it contends ought to have been included would weaken the showing of a salary disparity made by the analysis.

Likewise, the District of Columbia Circuit Court of Appeals, in Palmer v. Schultz, opined that

Implicit in the Bazemore holding is the principle that a mere conjecture or assertion on the defendant's part that some missing factor would explain the existing disparities between men and women generally cannot defeat the inference of discrimination created by plaintiffs' statistics…The logic of Bazemore, however, dictates that in most cases a defendant cannot rebut statistical evidence by mere conjectures or assertions, without introducing evidence to support the contention that the missing factor can explain the disparities as a product of a legitimate nondiscriminatory selection criterion.

The Appeals Court in Palmer also referenced Baldus and Cole’s treatise, Statistical Proof of Discrimination, which noted:

when otherwise relevant evidence is challenged on methodological grounds, the burden should normally be on the challenger (a) to present credible evidence that the statistical proof is defective and (b) to present a plausible explanation of how the asserted flaw is likely to bias the results against his or her position.

These arguments are consistent with the District of Columbia Court of Appeals’ opinion in Seger vs. Smith (738 F.2d 1249 (1984)), finding:

Of course, when a defendant claims that a specific factor was sufficiently objective to permit quantification, the defendant's failure to present alternative statistics incorporating the factor will severely undermine its rebuttal.

Together, these decisions consistently point to a burden-shifting approach to evidentiary showing.  

Contrary to the 2nd Circuit in Ohio v. Amex, which placed the entire burden squarely on the Plaintiffs, courts in Title VII cases dealing with statistical evidence have shifted the burden of providing empirical evidence to rebut Plaintiffs’ allegations once an initial threshold has been met.  Once econometric evidence is presented that demonstrates Defendant liability, the burden then shifts to the Defendant to provide data it contends were missing and demonstrate that the inclusion of such data undermines the Plaintiff's case.  The Defendant's burden goes well beyond simply throwing stones at the model. 

That is not to say all Plaintiff models meet the threshold of admissibility.  The Bazemore Court noted that "There may, of course, be some regressions so incomplete as to be inadmissible as irrelevant;" Just as with any other expert testimony, evidence presented in markets involving multi-sided platform would be subject to Rule 702, which governs testimony by expert witnesses. But, by the logic applied in Bazemore, Sobel v. Yeshiva, and other cases cited above, that rule should be applied on the validity of the Plaintiff's model in one market.  Once Plaintiffs' model showing Defendant liability in one market is ruled admissible, or its admissibility goes unchallenged by Defendants, then the burden should shift to Defendants to provide expert evidence to demonstrate both the existence and the effects of multi-sided effects that Plaintiffs may have omitted. 

The 2nd Circuit in Amex would elevate the threshold Plaintiffs must meet to the ceiling, thus standing in direct opposition to the burden-shifting logic adopted by the Courts with respect to statistical evidence.  Indeed, this same burden-shifting approach represents the most common method courts employ to adjudicate antitrust cases under the rule of reason. (see e.g., Michael Carrier, The Real Rule of Reason: Bridging the Disconnect, Brigham Young University Law Review, Vol. 1999, p. 1265)

Of course, one may respond by pointing out that the decisions cited herein deal with Title VII cases involving statistical evidence, not with relevant markets in an antitrust setting.  I find such a counter-argument unpersuasive. Courts are concerned with substance, not form (see, e.g., In Re Urethane Antitrust Litigation, 152 F.Supp.3d 357, in which the Court rejected a motion to exclude expert testimony, finding that “the Court is concerned with substance, not form;”). Whether in Title VII or antitrust cases, Courts attempt to balance the costs of committing errors of Type 1 (finding defendant liability where none exists) vs. Type 2 (rejecting defendant liability when it exists).  Lowering the risk of committing one type of error comes at the cost of increased risk of committing the other. The burden-shifting approaches advocated by the Supreme Court and appeals courts cited herein reflect this concern.

Daniel Rubinfeld offered this same burden-shifting alternative in his Columbia Law Review article Econometrics in the Courtroom.  Rubinfeld, who also discussed the nature of Type 1 and Type 2 errors in the context of the social disutility each may cause, noted, 


After some evidence of a relationship between sex and hiring is presented, the burden of production could then be shifted to the defendant to attack the plaintiff's hypothesis.

The shifting of the burden of production becomes more important when the number and forms of alternative hypotheses are more complex. This occurs naturally when defendants argue that  differential treatment of men and women is valid. One difficulty relates to the availability of information concerning possible discrimination. If the defendant has the data or other information that are necessary for the alternative hypotheses to be well specified, then it may be appropriate to make it easy for the plaintiff to shift the burden of production to  the defendant." (emphasis added)

Unfortunately, the Second Circuit’s decision in Ohio v. Amex ignores the role of such errors in determining burden of proof. Rather, it reflects a singular focus on errors of the first type, exposing future judicial decisions to a significant risk of errors of the second and society to considerable social costs.  In doing so, the Second Circuit's finding in Ohio v. Amex is clearly at odds with its own precedent in Sobel v. Yeshiva and Supreme Court precedent in Bazemore v. Friday. Should the Second Circuit’s approach be affirmed by the Supreme Court, it would levy a Sisyphean burden on Plaintiffs, allowing Defendants to successfully stave off antitrust challenges to multi-sided platforms by pointing to myriad factors that Plaintiffs may not have considered, regardless of whether Defendants have measured them or can even do so.