Challenging Plaintiff's Expert: A Deposition Framework for Statistical Sampling

The deposition is won or lost in what was obtained beforehand. The materials that matter are the expert’s underlying data set, the code or workbook that produced the analysis, the sampling frame from which the sample was drawn, the coding rules applied to classify records as violations, and any prior versions of the analysis.

Prior versions are the most frequently overlooked and the most productive. An analysis that moved substantially between drafts moved for a reason, and the reason is either a corrected error or a changed assumption — both of which are worth understanding before the expert is asked to explain them under oath.

The second preparatory step is running the analysis independently. An expert’s methodology described in a report is a summary; the methodology as implemented is in the workbook, and the two are not always the same. Discrepancies found in advance are the strongest material available in the room.

Population. Establish precisely who is in the analyzed population and who is not, then test the boundaries against the claim being adjudicated. Excluded categories and silently included ones both distort the result, and the expert frequently did not choose the boundary — counsel did.

Selection. Establish how the sample was actually drawn, in operational terms rather than in the language of the report. Ask what list was used, when it was generated, and whether it was complete. Sampling only from currently employed workers is the recurring defect; departed employees frequently have different violation patterns, and their exclusion has a direction.

Violation definition. Establish the coding rule, then test it against the governing legal standard. A rule that counts every meal under thirty minutes as a violation treats a twenty-nine minute meal taken voluntarily as equivalent to a meal never provided, which is not what the law requires and materially inflates the rate.

Premiums paid. Establish whether the analysis netted out violations for which a premium was already paid. Frequently it did not, which means the rate counts occurrences the employer already remedied.

Uncertainty. Establish the confidence interval, then have the expert state the range in plain terms. An interval wide enough that the true rate could be half the point estimate is a fact the report often does not foreground.

Defenses. Establish whether the methodology permits individualized defenses to be raised at all. This is the Duran point, and it is the answer least likely to have been prepared.

The first five domains produce criticism of an analysis. The sixth produces a constitutional objection to a trial plan, and the difference matters when the material is deployed. Duran v. U.S. Bank National Assn. (2014) 59 Cal.4th 1Duran v. U.S. Bank National Assn.(2014) 59 Cal.4th 1 · Cal. Supreme Court · May 29, 2014A trial plan using statistical sampling to establish classwide liability must comport with due process: the sample must be developed with expert input, be genuinely random and representative, and produce a manageable margin of error, and the defendant must be permitted to impeach the model.Does not holdDuran contains zero references to Sargon — confirmed against a control query, so the null is real rather than an indexing artifact. Neither does the other's work.Verified against the opinion sets the floor: a sampling plan used to establish classwide liability must be built with expert input, must be genuinely random and representative, must carry a manageable margin of error, and — the part that does the work here — must leave the defendant able to impeach the model and to present its affirmative defenses.

What failed in Duran was not the use of statistics. It was exclusion. The exemption was an affirmative defense on which the employer bore the burden as to each class member, and the trial plan foreclosed proof of it as to most of them. That is the shape to look for in the expert’s methodology: not imprecision, which is ordinary, but a design that resolves variation by leaving it out.

The limit is what keeps the objection credible. There is no due process right to litigate a defense against every class member, and no unfettered right to present individualized evidence — Estrada v. Royalty Carpet Mills (2024) 15 Cal.5th 582Estrada v. Royalty Carpet Mills, Inc.(2024) 15 Cal.5th 582 · Cal. Supreme Court · Jan. 18, 2024Trial courts lack inherent authority to strike a PAGA claim on manageability grounds based on judicial economy — the Chief Justice's own section heading.Does not holdNot a due process holding. The court expressly reserved: “we express no opinion as to the hypothetical questions of whether, and under what circumstances, a defendant's right to due process might ever support striking a PAGA …Verified against the opinion enforced that boundary directly. An objection phrased as a demand to examine all four hundred employees invites the answer Estrada supplies. The sustainable framing is narrower and harder to answer: this plan admits no mechanism by which contrary evidence, of any volume, could change the result.

In the room, that reduces to a single line of questioning. Ask what the model does when a class member’s facts differ from the sampled pattern. If the answer is that the model does not accommodate the difference, the Duran point has been made in the expert’s own words, which is worth considerably more than making it in a brief.

A deposition transcript in this area supports two different motions, decided under two different standards, and the questions that build one are not the questions that build the other.

The first is the Duran objection, and it is aimed at the trial plan rather than at the opinion. It says the plan leaves no route by which the employer’s evidence could affect the outcome. It is a due process argument, and its remedy is structure — a defined universe, an agreed protocol, a mechanism for individualized defenses — or a narrower claim.

The second is an admissibility objection under Sargon Enterprises, Inc. v. University of Southern California (2012) 55 Cal.4th 747Sargon Enterprises, Inc. v. University of Southern California(2012) 55 Cal.4th 747 · Cal. Supreme Court · Nov. 26, 2012Trial courts have a substantial gatekeeping responsibility under Evidence Code sections 801(b) and 802 to exclude expert opinion that is speculative, rests on matter an expert may not reasonably rely on, or is unsupported by the material relied on.Does not holdA lost-profits case, not wage and hour. It states the general California gatekeeping standard and nothing specific to this practice area.Verified against the opinion. Sargon is not a wage and hour case — it concerned lost profits — but it states the general California standard, and it is the authority under which a statistical opinion is excluded rather than merely criticized. Trial courts, it holds, have “a substantial ‘gatekeeping’ responsibility,” and under Evidence Code sections 801, subdivision (b), and 802, the court “acts as a gatekeeper to exclude expert opinion testimony that is (1) based on matter of a type on which an expert may not reasonably rely, (2) based on reasons unsupported by the material on which the expert relies, or (3) speculative.”

Section 802 is the provision that does the work on a sampling opinion, because it lets the court inquire into “not only the type of material on which an expert relies, but also whether that material actually supports the expert’s reasoning.” A court may conclude, Sargon says, “that there is simply too great an analytical gap between the data and the opinion proffered.”

That maps onto the six domains directly. A frame drawn from the wrong population, or a sample whose randomness was compromised, is an analytical gap between the data and the classwide inference. A coding rule that departs from the governing legal standard is a reason unsupported by the material relied on. A point estimate carrying an interval wide enough to admit half the asserted rate is the speculation ground. The same answers serve both motions; only the framing changes.

The limits are what keep the motion credible, and they are as important as the grounds. Sargon is explicit that the gatekeeping role “does not involve choosing between competing expert opinions,” that the court “must not weigh an opinion’s probative value or substitute its own opinion for the expert’s opinion,” and that it “does not resolve scientific controversies.” The goal is “simply to exclude ‘clearly invalid and unreliable’ expert opinion,” and the ruling is reviewed for abuse of discretion. A motion that invites the court to prefer the defense expert asks for something Sargon forbids. A motion that identifies a specific gap between the data relied on and the conclusion drawn asks for exactly what it authorizes.

One practical consequence is worth carrying into the room. Duran does not cite Sargon, and Sargon predates the sampling line entirely; neither opinion does the other’s work. A deposition that establishes only that the model cannot accommodate contrary facts has made the Duran point and left the Sargon point untouched. The Sargon point requires something more specific — the expert conceding that a particular inference is not supported by the material actually relied on. That is a different question, and it has to be asked.

Sometimes it does. The frame was drawn correctly, the sample is genuinely random, the coding rule tracks the legal standard, premiums were netted, and the interval is stated honestly. A deposition run on the assumption that defects must exist will produce nothing in that case, and will spend credibility doing it.

The purpose then changes from attacking the rate to bounding it. Pin the confidence interval and have the witness state it in plain terms. Establish which single assumption moves the result most, and by how much. Establish what the rate becomes under the defensible alternative to that assumption. The output is not a discredited expert but a range with known drivers — which is what an exposure model actually needs.

That has its own value at mediation. A rate that survived examination is a rate both sides can price against, and the argument moves from whether the number is real to what it implies. Counsel who can say precisely which assumptions were tested, and that the number held, is in a stronger position than counsel asserting the number is wrong without having shown where.

The purpose of the deposition is not to defeat the expert in the room. It is to produce a transcript in which the expert has conceded, in their own words, the specific assumptions on which the rate depends — because those concessions are what a motion, a mediation brief, or a cross-examination is built from.

That means asking for the arithmetic consequence of each concession while the witness is still under oath. If the coding rule were changed to exclude voluntarily shortened meals, what happens to the rate. If departed employees had been included, what is the direction of the effect. If premiums paid were netted out, what is the resulting figure. Answers to those questions convert methodological criticism into numbers.

The resulting analysis then feeds directly into the exposure model, where the difference between the expert’s rate and the defensible rate is the single largest line item in the case.