The Ninth Circuit approved use of statistical analysis that relies on averaging but reversed class certification because the district court failed to resolve whether more than a de minimis number of putative class members were injured.

On April 6, 2021, in Olean Wholesale Grocery Coop. v. Bumble Bee Foods LLC, the U.S. Court of Appeals for the Ninth Circuit resolved several important class certification issues that had long been unsettled in the circuit. The case involved antitrust claims and thus is of particular interest to the antitrust bar and companies facing antitrust class actions. But the rulings will affect other class actions as well.

Olean Wholesale and other plaintiffs alleged that producers of packaged tuna conspired to fix prices. The conspiracy's existence was largely undisputed, as several defendants had pleaded guilty to criminal conspiracy charges. As in most antitrust class actions, the primary issue for class certification was whether the plaintiffs could prove classwide impact through common proof. And, as in most class actions, the plaintiffs sought to do so using statistical regression models that purported to estimate an average overcharge across all purchasers—in this case, finding a 10.28% overcharge paid by 94.5% of the direct-purchaser class members. The defense expert, by contrast, concluded that at best only 72% of the purchasers paid any overcharge. The district court granted class certification without resolving this dispute, finding it enough that the plaintiffs' expert's method was "plausibly reliable" and concluding that the defense's challenge to that method was a "merits decision" for the jury to make.

The Ninth Circuit reversed. Joining every other circuit to have addressed the issue, the court first held that a plaintiff seeking class certification must prove by a preponderance of the evidence that Rule 23's requirements are met, including the predominance requirement. The court ruled that this standard flows from Rule 23's language and is consistent with the Supreme Court's mandate that district courts engage in "rigorous analysis" in deciding whether class certification is proper.

The court next broadly endorsed the use of representative sampling and statistical methods of proving injury at the class certification stage. Such proof is proper in individual antitrust damages actions and thus may be relied on in class actions as well. The court concluded that such proof is not improper merely because it relies on averaging, so long as the sample size is large enough to yield "statistically robust" results. While averaging may mask some individual differences in the amount of damages, the court, relying on general principles applicable to all class actions, ruled that individual issues as to the amount of damages do not defeat class certification.

Finally, the court ruled that, while individual amount-of-damages issues do not preclude certification, the district court erred in not resolving the dispute among the experts over what portion of the class was not injured at all. Again joining other circuits that have addressed the issue, the court held that plaintiffs must establish with common proof that "all (or nearly all)" class members were injured. Accordingly, class certification is not proper if more than a de minimis number of class members were uninjured. And the court suggested that anything above 5% to 6% of the class would exceed that threshold. The district court was thus required to resolve the factual dispute between the experts over whether the only 5.5% were uninjured, as the plaintiffs contended, or 28%, as the defense contended.

In a concurring and dissenting opinion, Judge Hurwitz disagreed with the last part of the court's ruling. In his view, more than a de minimis number of uninjured class members is not necessarily fatal to class certification. Certification might still be proper if a common method exists for separating the injured from the uninjured or if "uninjured members could be identified in future (perhaps non-class) proceedings."

Each of the court's rulings—the preponderance of evidence standard, the general propriety of using statistical models that rely on averaging, and the need to resolve disputes over the number of uninjured class members—broke new ground in the Ninth Circuit and resolved issues on which district courts in the circuit had disagreed. The court's decision is an important precedent for all class actions, and particularly important for antitrust and other cases in which the predominance question turns on individual-injury issues.