Statistical sampling allows experts to draw inferences confidently about the characteristics of a large group from a sample of the group that is sufficiently representative. In class actions, given the impracticality of proving a defendant’s liability with respect to each and every member of the class, plaintiffs have increasingly sought to use statistical sampling to establish class-wide liability. However, the use of statistics for that purpose has proven highly controversial, as illustrated in a California Supreme Court decision issued on May 29, 2014. Duran v. U.S. Bank N.A., 325 P.2d 916 (2014).

In Duran, loan officers of the defendant, a national bank, sued for unpaid overtime on the ground that they had been misclassified as exempt employees under the “outside salesperson” exemption in the California Labor Code. That exemption covers employees who spend more than 50% of their workday engaged in sales activities outside the office.

After certifying a class of 260 individuals, the trial court proposed selecting a random sample of 20 class members to testify at trial concerning their daily work activities inside and outside the office in order to determine liability. Defendant’s experts objected that such a small sample was not representative and thus would present a high risk of error. Instead, defendant’s experts advised the court to first conduct a pilot study to determine the optimal sample size. While plaintiff’s experts did not oppose this step, the trial court nevertheless rejected this and other objections lodged by defendant to the trial court’s statistical methodology. The trial court also later denied defendant’s request to introduce declarations and testimony at trial from class members outside the sample. Separately, despite a margin of error of plus or minus 43.3%, the trial court admitted statistical evidence from one of plaintiff’s experts who purported to project damages for the entire class based reliably on a random survey of class members concerning their overtime hours.

After trial, the lower court ruled that based on the testimony from the plaintiffs in the sample, all of the loan officers in the class had been misclassified by defendant and that all class members were owed overtime. To calculate damages, the trial court used the random survey conducted by plaintiff’s experts. Defendant appealed.

The California Supreme Court declined to hold that statistical sampling could never be used to prove liability in a class action. However, it criticized the trial court’s implementation of statistical sampling in the case before it as profoundly flawed.

First, the California Supreme Court credited the testimony of defendant’s experts that the sample size was far too small. It observed that the court selected a sample size of 20 without input from either side’s statistical experts. The Supreme Court ruled that going forward, lower courts that wish to use statistical sampling to establish liability must consult with the parties’ experts to “determine that a chosen sample size is statistically appropriate and capable of producing valid results within a reasonable margin of error.”

Second, the California Supreme Court explained that the trial court’s purportedly “random” group of class members testifying at trial was not at all random because of multiple instances of “selection bias,” i.e., inclusion and exclusion of class members from the sample based on nonrandom criteria that favored the plaintiffs. For example, one plaintiff initially included in the sample was later excluded because his work activities appeared to differ substantially from those of other loan officers. Additionally, the court allowed the named plaintiffs to testify even though they were not selected at random (to the contrary, they were selected by class counsel). Another plaintiff in the sample group failed to appear at trial. Finally, after the members of the sample had been selected, class members were given another opportunity to opt out, resulting in certain class members being excluded from, and new class members being added to, the sample (rather than selecting a new random sample from scratch).

Third, the trial court tolerated a large margin of error when it calculated damages. Specifically, as noted above, plaintiff’s experts calculated an estimate of average overtime hours with a margin of error of nearly 50%. This meant that the judgment entered against the defendant might be close to double the true amount of damages.

In the end, the California Supreme Court ruled that the trial court’s biased sampling plan infected its findings on both liability and damages and reversed the judgment. The upshot ofDuran is that courts implementing statistical sampling must utilize experts to ensure that the methodology used is sound and reliable.

Have you had occasion to use statistical sampling in class action litigation? If so, please share your experience.