In Duran v. U.S. Bank N.A. (pdf), the California Supreme Court recently addressed an important question in the context of state-court class actions: Can plaintiffs invoke statistical sampling in an attempt to prove class-wide liability and overcome the presence of individual questions that ordinarily would defeat class certification?

The court’s answer to that question is a mixed bag for business. The court firmly rejected the haphazard approach to sampling used by the trial court in the lawsuit against U.S. Bank. But the court left open the troubling possibility that sampling might be used in support of class certification in the future.

Why do plaintiffs try to use statistical sampling in the first place? To certify a class, plaintiffs must show that the elements of their claims can be proved for each class member by using common evidence. Common abstract questions are relatively easy to identify, but in many cases common answers are harder to come by. Even when there have been similar legal violations (a question that itself is often subject to debate), those violations may affect different class members in very different ways—or not at all—making it impossible to use common evidence to prove each class member’s individual case.

But what if plaintiffs could avoid the search for dispositive common evidence by identifying a class, selecting a few members as a sample, trying their individual cases with their individual evidence, and then extrapolating the results to the entire class? That would make common evidence largely unnecessary, potentially opening the door to class actions of all sorts.

In our view, this possibility should have been nipped in the bud three years ago when, in Wal-Mart Stores, Inc v. Dukes, the U.S. Supreme Court “disapprove[d] th[e] novel project” of “Trial by Formula,” under which “[a] sample set of the class members would be selected” for liability determinations, and then “[t]he percentage of claims determined to be valid would then be applied to the entire remaining class . . . to arrive at the entire class recovery—without further individualized proceedings.”

As the Court in Dukes explained—in the unanimous portion of that opinion—the problem with assessing liability through sampling is that it disregards a defendant’s right to “litigate its . . . defenses to individual claims,” contravening the Rules Enabling Act, which forbids the use of the federal rules of procedure (like Rule 23) to “abridge, enlarge or modify any substantive right.”

While Dukes rested on an interpretation of Rule 23 rather than constitutional due process guarantees, the core of Dukes’ holding—a defendant’s right “to present every available defense”—is based on long-standing principles of due process. As we have explained elsewhere (pdf), these due-process principles are applicable in state court. Under those principles, the use of “Trial by Formula” should be equally impermissible in state-court class actions.

Unfortunately, the California Supreme Court did not adopt that categorical rule in Duran, instead concluding that the trial court had erred in devising its own deeply flawed sampling methodology.

The plaintiffs in Duran had sought to represent a class of 260 loan officers, contending that they were improperly classified as “outside salespersons” (and thus exempt from California’s overtime-pay requirements). The “outside salesperson” exemption applies to employees who spend more than half their work day in sales activities outside the office. It’s easy to see why this issue is often individualized: When different class members spend different amounts of time in outside sales—some more and some less than 50%—the application of the exemption, and therefore liability, varies from person to person, likely precluding class certification. By the time Duran went to trial, plaintiffs’ counsel were on their third set of class representatives because the first two sets had admitted that they spent more than half their time in outside sales. At class certification, U.S. Bank presented declarations from 75 class members attesting that they, too, spent more than half their time in outside sales—and therefore were properly classified as outside salespersons. More important, the evidence showed that liability determinations required scrutiny of individual class members’ circumstances.

Plaintiffs and the trial court found a way around that problem: simply select a sample of 21 class members, take evidence on their (and only their) work habits, and extrapolate liability (and the average award) to all 239 other class members. Rejecting both US Bank’s proposal to try all class members’ claims and plaintiffs’ proposal to select an allegedly representative sample of plaintiffs using methods devised by the parties’ experts, the trial court selected class members at random to go to trial along with the two class representatives. Several members of the trial group opted out of the class, two of them explicitly because they had worked primarily in outside sales and were asked to withdraw by plaintiffs’ counsel. The trial court precluded US Bank from presenting evidence that class members outside the group (including former class representatives) were exempt.

Based on the sample trial, the trial court found that the entire class had been misclassified—though even the plaintiffs’ expert admitted that the results from the sample had a 13% margin of error. The trial court also adopted the calculations of plaintiffs’ expert as to the average overtime worked, though the conceded margin of error for that figure was a whopping 43%.

Not so fast, said a unanimous California Supreme Court. But while its decision made clear that the trial court had exceeded permissible limits on the use of sampling and statistics to override individual variations, it remains unclear where the limits actually lie.

The court laid down some clear principles. Echoing the federal Rules Enabling Act, the court held that “the class action procedural device may not be used to abridge a party‘s substantive rights,” so that “[p]rocedural innovation must conform to the substantive rights of the parties.”

In another significant holding, the court rejected plaintiffs’ contention that the burden-shifting “pattern or practice” analysis used in some intentional discrimination cases under the federal civil rights laws should be extended to wage-and-hour claims. In wage-and-hour litigation, the court held, liability depends primarily on the individual plaintiff’s circumstances rather than the employer’s intent. By contrast with the intentional discrimination context, “[l]iability to one employee is in no way excused or established by the employer‘s classification of other employees.”

And the court disapproved another common tactic: reframing “decisions about the fact of liability … as questions about the extent of liability.” Plaintiffs frequently try to characterize questions of causation and injury (which fall within the first category) as questions of damages (which fall within the second). But the court declined to adopt a clear rule for liability questions: “individual issues will not necessarily overwhelm common issues when a case involves exemptions premised on how employees spend the workday.”

Unfortunately, the opinion provides little guidance on when individualized issues overwhelm common ones and preclude a finding of predominance. The court suggested that some individualized issues might be swept into a trial-management plan, so long as the plan assesses whether the degree of individual variability precludes class treatment and the individualized issues can be “managed fairly and efficiently.” The Duran trial court’s statistical sampling failed this test in spectacular fashion because it ignored individual issues rather than managing them and used a sample size that was too small and that was cherry-picked to exclude members who might undermine the plaintiffs’ case.

Yet the California Supreme Court stopped short of explaining whether or when sampling might be available to prove liability in a class action, other than to say that statistical methods were not permissible for that purpose when “incompatible with the nature of the plaintiffs’ claims or the defendant‘s defenses.” While holding that“[s]tatistical methods cannot entirely substitute for common proof,” the court speculated that it “may be possible to manage individual issues through the use of surveys and statistical sampling.” The court noted that any permissible method would have to be developed with expert input and that the defendant would have to have an opportunity both to impeach the sampling model and to show through other evidence that its liability should be reduced.

While the California Supreme Court held that “a class action trial management plan must permit the litigation of relevant affirmative defenses, even when these defenses turn on individual questions,” the court asserted that “[n]o case, to our knowledge, holds that a defendant has a due process right to litigate an affirmative defense as to each individual class member.” That conclusion seems at odds with Dukes’ clear message that “a class cannot be certified on the premise that [the defendant] will not be entitled to litigate its statutory defenses to individual claims.”

In a separate concurring opinion, Justice Liu observed that “it is important that courts employ a proper understanding of the substantive governing law to inform the class certification decision.” At the same time, Justice Liu’s concurrence sought to underscore that the court’s opinion “leaves open ‘the appropriate use of representative testimony, sampling, or other procedures employing statistical methodology.’” And the concurrence “encourage[d] trial courts to be ‘procedurally innovative’ in managing class actions”—an approach that contrasts with the U.S. Supreme Court’s disapproval of “novel project[s]” like trial by formula.

In short, Duran raises as many questions as it answers. On the one hand, the decision is a major victory for U.S. Bank, as the court of appeal’s decision rejecting the trial court’s flawed trial plan was “affirmed in its entirety.” At the same time, the plaintiffs’ bar has characterized Duran as a victory for its side. In fact, the opinion provides grist for the mills of both sides: Every declaration of a guiding legal principle is accompanied by a countervailing qualification. Common issues must predominate, but the opinion does not make clear when individualized issues preclude certification and when they must merely be managed as part of a trial plan. The court suggests that statistical sampling may be better suited to determining damages rather than liability, but provides little guidance as to where to draw the line when the proffered statistical methods are less outlandish than those addressed in the opinion. The only sure thing is that the battle over class certification in the California state courts—including the use of statistical sampling—will rage on for the foreseeable future.