There seem to be two prevailing conceptions of class actions.  In one view, a class action is a way of determining many similar claims at once by evaluating common evidence that reliably establishes liability (and lays a ground work for efficiently calculating damages) for each class member.  That is, the class device produces the same results as individual actions would, but more efficiently.  In the other view—one we consider misguided—a “class” of plaintiffs complaining about similar conduct can have their claims determined through statistical sampling even if no common evidence will provide a common answer to common factual or legal questions. Instead, this theory holds, the results of mini-trials can simply be extrapolated to the entire class, even if individual results would vary widely.

Last week, the Ninth Circuit took a step deeper into the second camp in Jimenez v. Allstate Insurance Co. (pdf), delivering a ringing endorsement of statistical sampling as a way to establish liability as well as damages.

In Jimenez, a wage-and-hour class action, the plaintiffs alleged that Allstate had converted its claims adjusters from salaried employees who were exempt from the overtime laws to hourly employees who were entitled to overtime.  The plaintiffs contended that the reclassified adjusters were still paid a “salary” based on a 40-hour work week and that overtime, while still routinely worked, was treated as an “exception” and rarely compensated. The district court certified a class based on purportedly common questions about whether Allstate had acted unlawfully in treating overtime as an “exception” and whether its failure to reduce routine work hours amounted to an “unofficial” policy of requiring overtime work without overtime pay.  According to the district court, it was this unofficial policy that held the class together.

The Ninth Circuit agreed that the potential existence of an unofficial policy to violate the company’s entirely lawful overtime policy presented a common question that could be answered by statistically sampling the records of selected plaintiffs.  Taken to the extreme, that holding might permit class certification whenever any plaintiff contended that a lawful policy was not followed more than once—no matter how individualized and anecdotal the evidence of departures from that policy might be.  The Ninth Circuit also approved as common the question whether Allstate knew or should have known that claims adjusters were working unpaid overtime, which plaintiffs contended could be shown both through testimony of individual managers and by analyzing log-on and log-off times on class members’ computers.  The third common question, according to the Ninth Circuit, was whether Allstate “stood idly by” in the face of this knowledge.  The court did not explain what proportion of class members would have to have engaged in known and uncompensated overtime work in order to resolve these issues in favor of the entire class, i.e., to prove the existence of an unofficial policy to violate the lawful official policy.

It’s hard to understand how any of these purportedly common questions could pass muster under Wal-Mart Stores, Inc. v. Dukes, which rejected similarly vague assertions of commonality.  Indeed, the Supreme Court emphasized in Dukes that a plaintiff must have “significant proof” of the existence of the alleged “general policy” (in that case, an alleged policy of discrimination against female employees) in order to establish commonality.

But while Dukes treated proof of the general policy as the “glue” needed to hold class members’ claims together, the Ninth Circuit appears to skip a step in Jimenez by allowing the disputed existence of the policy to serve as the common question.  That bootstrapping approach assumes the necessary “glue” rather than requiring “significant proof” as a prerequisite to class treatment, as Dukes requires.

The Ninth Circuit also appeared willing to allow even more bootstrapping when it came time to prove the existence of the policy that was necessary to hold the class together in the first place.  In rejecting Allstate’s due process challenge to the use of statistical sampling to prove the allegedly common questions, the Ninth Circuit held that “statistical sampling and representative testimony are acceptable ways to determine liability so long as the use of these techniques is not expanded into the realm of damages.”  The panel chiefly relied on Leyva v. Medline Industries, Inc., a decision that had nothing to do with statistical sampling or representative testimony, but held only that automated analysis of time and compensation records could suffice to generate each class member’s share of damages.  More broadly, Leyva reaffirmed Ninth Circuit precedent that presumptively exempts damages issues from the commonality and predominance analyses—a position in severe tension with the Supreme Court’s decision in Comcast Corp. v. Behrend, which overturned class certification because “respondents’ model falls far short of establishing that damages are capable of measurement on a classwide basis.”

In accepting “statistical sampling and representative testimony” as “acceptable ways to determine liability,” the Ninth Circuit did not explain how the plaintiff’s techniques were supposed to work, let alone how they could produce class-wide answers as to any aspect of liability.  Nor did the Ninth Circuit seriously address the Supreme Court’s rejection in Dukes of the use of sampling at the liability phase as a forbidden “Trial by Formula.”

In particular, the court did not explain how a district court could reliably resolve the claims of each class member based on sampled results.  What would the court do if the records of only 80% of sampled employees reflected overtime violations?  Would it hold that an “unofficial policy” existed as to all class members? If not, how would it tell who was injured and who was not?  What about 60%? Is that enough to prove a policy that can, or should, result in class-wide liability? What about 40%?  20%? In such circumstances would the class be redefined to exclude those whose sampled records did not indicated that had not worked unpaid overtime, or would it include all adjusters? These intractable questions underscore why the Supreme Court—in the unanimous portion of the decision in Dukes—was so skeptical of the use of statistical sampling to determine liability.

The Ninth Circuit’s answer to these concerns is unsatisfying.  The court held that any divergences from the sampled or extrapolated results could be dealt with through individualized trials of any affirmative defenses at the damages phase.  In practice, however, that could encompass hundreds of mini-trials on causation, injury, and damages.  That unmanageable result represents (to us) a conception of the class action at odds with Rule 23.