Imagine a legal system in which justice for thousands, or even millions, of litigants can be handed out with the push of a button. A system where a computer algorithm short circuits the need for plaintiffs to pursue claims in individual lawsuits. A system where both a defendant’s liability and the damages of an entire class of plaintiffs are determined statistically, rather than the old-fashioned way with human witnesses and documentary evidence. This system is the ultimate dream for plaintiffs’ class action lawyers and the worst nightmare for businesses, but is it a pipe dream or a near reality?
Statistics in Class Actions: Court Decisions
In 2011, the U.S. Supreme Court decided Wal-Mart Stores, Inc. v. Dukes, in which plaintiffs proposed to use a statistical regression analysis to tie adverse employment decisions to allegedly discriminatory decision-making, and to determine damages through a method of statistical sampling. Justice Scalia’s majority opinion rejected both suggestions, finding that the proposed method for establishing the existence of a discriminatory policy was not reliable and that the proposed method of calculating damages amounted to a “trial by formula,” which the Court would not endorse. 131 S. Ct. 2541, 2561 (2011). In 2013, the Court decided Comcast Corp. v. Behrend, in which the Court reversed a class certification order resting on expert opinion claiming class-wide impact using statistical methods. 133 S. Ct. 1426 (2013).
After Wal-Mart and Comcast, the prospects of ever successfully using statistical evidence as a shortcut for individual proof in class actions appeared grim. A case on the Court’s 2015 docket, Tyson Foods, Inc. v. Bouaphakeo, 136 S. Ct. 1036 (2016), looked like it might be the final blow, but, in fact, it was not. There, the Court accepted for review the question of
[w]hether differences among individual class members may be ignored and a class action certified under Federal Rule of Civil Procedure 23(b)(3), or a collective action certified under the Fair Labor Standards Act, where liability and damages will be determined with statistical techniques that presume all class members are identical to the average observed in a sample.
U.S. Supreme Court, Questions Presented: Tyson Foods (2015). In the end, the Court affirmed the class certification order in Tyson Foods, concluding in an opinion authored by Justice Kennedy that because the statistical evidence relied upon would have been admissible in any individual case brought by a class member, it was also admissible as class-wide proof. The Tyson Foods decision makes clear that there may still be a place for statistical evidence in class actions.
Questions of Causation, Methodology, and Representative Evidence
Statistics used to prove claims on a class-wide basis must correspond to questions at issue through statistically sound methodology. Figures that may seem persuasive and relevant are often discredited after closer examination, whether it be for a failure to show causation, improper methodology, or a misapplication of representative evidence.
Obstacles abound for establishing causation. Class-wide causation is the first area in which courts have heightened the burden on class action plaintiffs. Plaintiffs are required to prove causation on a class-wide basis and often rely on statistical evidence to do so.
In consumer class actions, plaintiffs have tried to use statistics to show causation on a class-wide basis through the “fraud on the market” presumption of reliance established in the securities context, where the market price of securities is viewed as an accurate measure of value by investors. But the courts have rejected the “fraud on the market” presumption in consumer class actions.
In UFCW Local 1776 v. Eli Lilly & Co., for example, the plaintiffs argued that Lilly had made misrepresentations about Zyprexa that caused its sales to rise and that the increase in sales was evidence of class-wide reliance on the alleged misrepresentations. 620 F.3d 121, 134–35 (2d Cir. 2010). The Second Circuit Court of Appeals rejected this argument, explaining that due to “the nature of prescriptions . . . this theory of causation is interrupted by the independent actions of prescribing physicians, which thwarts any attempt to show proximate cause through generalized proof.” Id. at 135. In other words, causation could not be proven on a class-wide basis because of individual issues.
The Second Circuit in McLaughlin v. Am. Tobacco Co., a RICO class brought against the tobacco industry related to alleged claims about “light” cigarettes being healthier, likewise rejected the “fraud on the market” theory. 522 F.3d 215 (2d Cir. 2008). There, the plaintiffs tried to establish class-wide reliance using a statistical model showing a market shift in brand preferences allegedly resulting from defendants’ marketing of light cigarettes. But the Court held that the plaintiffs’ statistical evidence did not prove that the shift was caused by health claims as opposed to some other factor (including that smoking light cigarettes was “cool”). Id. at 225. The plaintiffs’ “‘loss of value’ model purport[ing] to measure the difference between the price plaintiffs paid for light cigarettes as represented by defendants and the (presumably lower) price they would have paid (but for defendant’s misrepresentation)” was likewise rejected as “pure speculation” discounting all other variables. Id. at 228–29 (further noting that expert’s survey asking plaintiffs to compare what they would pay for a truly healthy light cigarette with one that was misrepresented “conceptualize[d] the impossible”).
But a 2007 decision, In re Neurontin Mktg. & Sales Practices Litig., 44 F.R.D. 89 (D. Mass. 2007), suggests that there may be room for presumptions of causation with strongly suggestive statistics in class actions. There, the plaintiffs alleged fraudulent marketing of off-label uses of the drug Neurontin but were denied class certification due to an inability to identify, on a class-wide basis, which consumers experienced injuries as a result of the off-label marketing rather than prescribing decisions. The plaintiffs presented expert opinions that assumed all Neurontin marketing during a certain period was both off-label and fraudulent and found that increased prescriptions were the result of fraudulent promotion, a methodology that failed to account for other variables contributing to prescription decisions. See In re Neurontin Mktg., Sales Practices & Prods. Liab. Litig., 257 F.R.D. 315, 329–30 (D. Mass. 2009). The defendant won summary judgment, but the First Circuit reversed and remanded, finding that just because “some physicians may have considered factors other than Pfizer’s detailing material [on off-label uses],” the causal chain was not so attenuated as to eliminate proximate cause and was a jury question. In re Neurontin Mktg. & Sales Practices Litig., 712 F.3d 60, 67–68 (1st Cir. 2013).
Courts are more strongly scrutinizing methodology. Statistics are only as reliable as their methodologies are sound, and class action plaintiffs cannot succeed without showing that their theories are verifiable. While plaintiffs in the past may have relied on weaker statistical evidence to get classes certified, courts have heightened the burden on plaintiffs to demonstrate the soundness of their statistical methodology.
One early example of this phenomenon is In re Facebook, Inc., PPC Advertising Litigation, where plaintiffs claimed that Facebook failed to properly charge advertisers per click on ads as agreed because Facebook failed to set up “‘click’ filters so that they were consistent with ‘reasonable industry standards,’” resulting in charges for invalid clicks. 282 F.R.D. 446, 458 (N.D. Cal. 2012). The plaintiffs claimed that their expert would use a “rule-based operational algorithmic methodology to distinguish between ‘valid’ and ‘invalid’ clicks.” Id. In his deposition, however, their expert admitted he had not yet developed and did not know how to develop such a methodology. The court held that the expert’s opinion “that he could design algorithms that would distinguish between valid and invalid clicks, and that would also determine clicks that are based on ‘fraud,’” for purposes of an unfair business practices claim did not prove that there was “a viable method for proving each class member’s individual recovery” and did not prove common questions predominated. Id. at 460.
Comcast has become the lead-impact decision in this area. In that case, the plaintiffs alleged that Comcast’s area coverage swapping arrangements with competitors resulted in supracompetitive prices that injured cable customers in one region as a class. 133 S. Ct. 1426, 1430 (2013). To get the class certified, plaintiffs needed to show that the antitrust impact could be proved through evidence common to the class and that “the damages were measurable ‘on a class-wide’ basis through use of a ‘common methodology.’” Id. at 1430. Plaintiffs presented four theories of antitrust impact that could be proved through evidence common to the class, but the district court had only accepted one theory—that “Comcast’s activities reduced the level of competition from ‘overbuilders,’ companies that build competing cable networks in areas where an incumbent cable company already operates.” Id. at 1430–31. An expert designed a regression model to compare actual cable prices with hypothetical prices that would have prevailed but for Comcast’s purported unfair practices. However, the model considered the joint effect of all four theories, not simply the one the court accepted. The Supreme Court held that the expert’s failure to isolate the alleged antitrust impact of the one allowed theory meant the model could not be used to show class-wide damages. The case was remanded with instructions to the lower courts to probe the merits of the model.
Courts distinguish trial by formula situations from valid representative evidence. In some cases, plaintiffs present statistical sampling in place of the individual experiences of each class member to prove one or more elements of a cause of action. For example, in Wal-Mart, the plaintiffs proposed to take a sample of class members and determine, through depositions, the percentage of their claims that were valid. This percentage would be applied to the rest of the class to determine the number of valid claims. This “number of presumptively valid claims” would then be multiplied by “the average backpay award in the sample set” to determine the aggregate damage award for the class. 131 S. Ct. 2541, 2561 (2011). The Court categorized this methodology as a “trial by formula” that “enlarge[d]” the class members’ “substantive right[s]” in violation of the Rules Enabling Act by, in essence, precluding defendants from raising individualized defenses. Id.
In Duran v. U.S. Bank National Ass’n, 325 P.3d 916 (Cal. 2014), the Supreme Court of California likewise rejected a “trial by formula” case in an action alleging that the bank had misclassified plaintiffs as falling under an exemption to overtime. The trial court certified a class of 260 plaintiffs and established a plan for determining liability through testimony on the work habits of 21 plaintiffs, excluding evidence from the bank about the work habits of plaintiffs outside the sample. Damages were then calculated by “extrapolat[ing] the average amount of overtime reported by the sample group to the class as a whole.” Id. at 920. Rejecting this approach, the Duran court found that “[s]tatistical methods cannot entirely substitute for common proof.” Id. at 932–33. While the trial court found a predominance of common questions in factors such as the uniform classifications of the class of employees as exempt without any inquiry into their work habits, the trial plan did not account for individual issues arising from the bank’s defenses. The method for establishing liability was another example of the trial by formula method discounted in Wal-Mart, and the trial court did not consult any experts in developing it. The plan “unreasonably prevented USB from supporting its affirmative defense” by refusing to admit evidence that plaintiffs outside the sample were properly exempt. Id. at 934, 935–36.
By contrast, the plaintiffs in Tyson Foods successfully used representative evidence to prove class-wide claims of unlawful denial of overtime compensation. Employees at a Tyson Foods pork processing plant were not being compensated for the time it took them to “don and doff” protective gear required in their jobs. Because Tyson did not keep records of time spent donning and doffing, the employees relied on “representative evidence”: the plaintiffs’ expert averaged the time taken to don and doff from a sample of 744 videotaped observations. 136 S. Ct. 1036, 1043 (2016). Tyson argued that because individuals spend varied amounts of time donning and doffing, individual issues predominated over questions common to the class, and, thus, the study could not be used to establish class-wide liability. Id. at 1045–46.
Tyson requested that the Court establish a broad rule against using representative evidence in class actions, but the Court declined, holding that “[w]hether and when statistical evidence can be used to establish classwide liability will depend on the purpose for which the evidence is being introduced.” Id. at 1046. The Court recognized that sometimes representative evidence is the only practicable means of presenting data and held that it may be used to establish class-wide liability if it can be shown that “each class member could have relied on that sample to establish liability if he or she had brought an individual action.” Id. Tyson was not deprived of an opportunity to litigate individual defenses because the defense that the study was unrepresentative or inaccurate was “itself common to the claims made by all class members.” Id. at 1047.
The Tyson Foods Court noted that in Wal-Mart, there was no evidence of a common policy of discrimination; instead, the plaintiffs tried “to use representative evidence as a means of overcoming this absence of a common policy.” Id. at 1048–49. The Wal-Mart defendant would have been deprived of the right to litigate individual defenses, unlike in Tyson, because the plaintiffs were not similarly situated and could not have individually prevailed by relying on depositions of how other employees were discriminated against. In other words, the Court determined that “trial by formula” deprives a defendant of the right to litigate individual defenses, but valid representative evidence that is truly common to all plaintiffs should not.
When used and formulated properly, statistics can help understand dynamics among large groups; in the class action context, statistics can be useful tools for proving or even disproving certain elements of a common claim or issue.
On the other hand, as the case law demonstrates, it is easy to abuse or misapply statistical methods. Plaintiffs often run into problems because they can’t establish causation, use improper methodology or methodology that does not correspond to a specific demand, or fail to differentiate between appropriate statistics and statistics that attempt to short circuit procedural safeguards.
Part II of this article (to be published separately on the Consumer Litigation Committee’s website) will address some of the practice considerations that should be considered based on the way that courts have dealt with these issues.
Copyright © 2017, American Bar Association. All rights reserved. This information or any portion thereof may not be copied or disseminated in any form or by any means or downloaded or stored in an electronic database or retrieval system without the express written consent of the American Bar Association. The views expressed in this article are those of the author(s) and do not necessarily reflect the positions or policies of the American Bar Association, the Section of Litigation, this committee, or the employer(s) of the author(s).