Duran v. U.S. Bank National Assn.

A trial plan that uses statistical sampling to establish classwide liability must comport with due process: the sample must be developed with expert input, genuinely random and representative, large enough to produce manageable margins of error — and the defendant must be permitted to impeach the model and to present its affirmative defenses, including evidence that individual class members are outside liability entirely. Justice Corrigan wrote for a unanimous court, reversing a $15 million judgment built on a 21-witness sample from which the defense was forbidden to depart.

The court did not outlaw sampling. Statistical methods remain available, particularly on damages, and the opinion is explicit that defendants have no right to litigate defenses against every individual class member. What due process forbids is a trial plan that manufactures uniformity by excluding the evidence of variation — the difference between proving a generalized fact and assuming one.

The opinion opens by calling the case 'an exceedingly rare beast: a wage and hour class action that proceeded through trial to verdict.' U.S. Bank's business banking officers — outside salespeople on paper, classified exempt under Labor Code § 1171Lab. Code § 1171Operative text of § 11711171. The provisions of this chapter shall apply to and include men, women and minors employed in any occupation, trade, or industry, whether compensation is measured by time, piece, or otherwise, but shall not include any individual employed as an outside salesman or any individual participating in a national service program carried out using assistance provided under Section 12571 of Title 42 of the United States Code. …Fetched from leginfo 2026-08-02's more-than-half-time-outside test — sued for misclassification. A class of 260 was certified on the theory that the position was standardized and the bank had classified everyone exempt without examining anyone's actual work. The certification record itself foreshadowed the problem: 34 declarations saying BBOs worked mostly inside, 75 saying the opposite, and four successive named plaintiffs replaced after testifying themselves out of the case.

When trial planning arrived, the court rejected both sides' proposals and devised its own: the clerk drew 20 class members' names 'from the proverbial hat,' added the two named plaintiffs, and decreed that liability and damages for all 260 would be extrapolated from this representative witness group — a group that, as finally constituted after opt-outs and substitutions, numbered 21. Nothing in the record explained why 20, or why that method. The bank's statistician warned the sample was far too small; the objection was overruled.

Every failure mode the defense predicted then occurred. When class members were given a second chance to opt out, four of the twenty randomly drawn RWG members left — a 20 percent opt-out rate against 2 percent for the rest of the class, a divergence the bank's expert called very unlikely to be random, and one that predictably skewed the remaining sample toward claimants. The court refused to readmit them. It then granted the order that decided the case: the bank could not introduce testimony, evidence, or argument about any BBO outside the sample — not the 75 favorable declarations, not the former named plaintiffs, not the opt-outs — where the purpose was to affect the liability or damages analysis.

Phase one ran 40 court days; the court found the entire class misclassified. Phase two extrapolated the sample's average overtime to all 260 members — with a margin of error the record put at 43 percent — producing roughly $15 million, over $57,000 per person, including recovery for class members who, on the excluded evidence, were properly exempt and owed nothing. The Court of Appeal reversed and ordered decertification; the Supreme Court affirmed in full.

The constitutional defect was not sampling; it was the exclusion. The exemption was an affirmative defense on which the bank bore the burden as to each class member, and the trial plan barred the bank from proving it as to 239 of them. Class actions are procedural devices — they aggregate claims, but they 'may not be used to abridge a party's substantive rights,' and a defendant 'must be given a chance to impeach [the] model or otherwise show that its liability is reduced.' A verdict that awards damages to people the excluded evidence would have shown were exempt is not efficient adjudication; it is liability by assumption.

The court was equally careful about the holding's limits, and Estrada v. Royalty Carpet Mills later enforced them: there is no due process right to individually litigate a defense against every class member, no 'unfettered right to present individualized evidence,' and courts retain discretion over how defenses are tried 'within whatever method the court and the parties fashion.' Duran's floor is an opportunity — a genuine one — to contest the plaintiff's generalized proof and to present the defense in some adequate form. The floor is low; the trial court here tunneled beneath it.

Duran leaves a functioning methodology corridor. Sampling and surveys remain available — most safely for damages once liability is established (the Bell v. Farmers line), and even for liability-adjacent findings where variability is genuinely low and the model is built properly: designed by experts rather than judges, drawn from a complete and unbiased frame, sized to the population's variability, reported with confidence intervals, and subjected to adversarial testing. Justice Liu's concurrence — later quoted in Estrada — adds that statistical methods 'designed to reveal generalized characteristics of a population' can be particularly apt where aggregate liability rather than individual compensation is at stake.

The federal contrast sharpens the rule. Tyson Foods, Inc. v. Bouaphakeo (2016) 577 U.S. 442 permitted representative time-study proof where each employee could have relied on the same study in an individual action — a records-gap inference, not a substitute for one. Duran's own sequel shows the other boundary: on remand, recertification was denied, and Duran v. U.S. Bank National Assn. (2018) 19 Cal.App.5th 630 affirmed, because no workable trial plan could bridge the class's variability. Sampling is a bridge for proving common patterns, not a solvent for dissolving individual differences.

Estrada made Duran more important, not less. Manageability can no longer kill a PAGA claim, so the fight moved inside the trial plan — where Duran supplies the constitutional constraint on how aggregate proof may work, and § 2699(p)Lab. Code § 2699Operative text of § 2699(p)(p) The superior court may limit the evidence to be presented at trial or otherwise limit the scope of any claim filed pursuant to this part to ensure that the claim can be effectively tried.Fetched from leginfo 2026-08-02 now supplies the statutory vehicle for demanding a plan that respects it. A PAGA plaintiff proposing to prove thousands of violations through a handful of declarants and an extrapolation model faces exactly the objections Duran sustained: sampling frame, selection bias, size, margin of error, and the employer's right to contest violation status for employees the model never examined.

In practice the decision functions as a deposition outline. Every element of the plaintiff's methodology is a cross-examination module — who designed the sample and with what expertise; whether the frame includes departed employees and non-claimants; how nonresponse was handled; what the confidence intervals actually are; whether the model even permits individualized rebuttal. The 174-question expert deposition framework used in my practice is organized around precisely these Duran vulnerabilities, because a sampling model that cannot survive them cannot support a judgment.

Duran is the constitutional floor under every aggregation fight in California wage and hour law — class trials, PAGA trial plans, and the settlement-table shadow of both. Its practical teaching runs to leverage: a plaintiff whose proof model cannot be executed within Duran's limits holds a claim that cannot actually be tried at full asserted value, and demonstrating that early — through expert work, not adjectives — reprices the case. The bank spent 40 trial days losing before winning; the modern defense goal is to front-load the same showing into the trial-plan and expert phase.

The decision also disciplines the defense side. Duran rewards employers who can produce the evidence of variation — real records of how work actually varied across employees, sites, and periods. An employer with no records invites the very inferences and approximations it will later protest; Donohue makes that explicit for break records, and Tyson makes it explicit federally. The right to challenge the plaintiff's model is worth most to the party holding better data.