Measure a representative part, prove the whole. The statistics are standard. Whether a courtroom accepts them depends on design choices made at the start.
Start a conversation with Cournot, the Institute’s damages concierge, already scoped to sampling & extrapolation. Pick a starting point, or describe the dispute directly.
Some aggregate claims cannot be computed record-by-record: the records are too many, too incomplete, or the quantity at issue, what consumers would have paid, how employees actually spent their time, was never recorded at all. The statistical answer is to measure a part and infer the whole, and it arrives in three main forms. Sampling draws a subset of claims, employees or transactions, measures them carefully, and extrapolates, with the validity of the exercise living entirely in the design: random selection, adequate size, and honest confidence intervals around the result. Surveys measure what records never captured, what consumers understood, expected or would have done, under a distinct methodological discipline of question design, controls and pretesting that the field enforces and opposing experts police. Conjoint analysis, the most technical of the three, estimates the value of a specific product attribute by analyzing choices among systematically varied alternatives, and has become the standard proposal for valuing the misrepresented or omitted feature in consumer class actions, along with the standard target: critics press it on whether stated choices track real market behavior and on whether willingness-to-pay translates to a market price effect without modeling supply. All three tools are orthodox statistics with decades of use inside and outside litigation. What separates the admitted from the excluded is rarely the mathematics; it is design, executed and documented before anyone knew what answer the method would produce.
Each tool has a methodological rulebook, and the rebuttal expert has read it.
The sample’s claim to represent the whole. Judgment samples and convenience samples forfeit it, and everyone knows.
Enough observations for the inference claimed, with the uncertainty stated as intervals rather than hidden.
Neutral questions, proper universe, controls for guessing and noise, pretesting. The checklist the field enforces.
Attributes and levels chosen to isolate the feature at issue, with realistic alternatives and prices.
From measured part to claimed whole, valid only as far as the population matches the frame the sample was drawn from.
Design fixed and documented before results exist. The strongest available answer to the advocacy attack.
How the Institute approaches statistical proof.
Often whether an unmeasurable claim becomes a measurable one.
A sampling plan or survey designed after the litigation theory is set, by experts who know what answer helps, carries a suspicion that no later testimony fully removes. The protection is procedural: fix the design early, document it, and where possible pre-specify it before results exist. Rigor you can show beats rigor you can claim.
When the design is sound and the use fits the claim, and the two conditions do different work. Soundness is statistical: random selection from a well-defined frame, a size adequate to the precision claimed, measurement of the sampled units that would itself survive scrutiny, and results reported with their intervals. Fit is legal and contextual: representative proof has a long history in wage cases, government reimbursement disputes and claims administration, and a more contested one where individual circumstances dominate the question being extrapolated. Where the line sits for a given claim is counsel's research. What is consistent everywhere is the failure mode: a sample assembled for convenience, or measured with a thumb on the scale, converts a powerful tool into the exhibit for the other side.
The instrument before the results, almost always. The recurring targets are leading or loaded questions that telegraph the desired answer; the wrong universe, respondents who are not the relevant consumers; absence of a control condition, so the survey cannot separate the effect of the challenged statement from noise, guessing and preexisting beliefs; demand effects, respondents inferring what the sponsor wants; and unrealistic stimuli that present the product or claim in a way no market participant ever saw. Each has a recognized remedy in the survey literature, controls, pretesting, neutral wording, realistic presentation, which is exactly why omissions are treated as culpable rather than technical. A survey that ignores the field's own handbook invites exclusion in the field's own vocabulary.
It measures relative preference rigorously: how much a specific attribute, the certification, the ingredient claim, the security feature, contributes to consumers' choices, expressed as willingness to pay. That is genuinely useful where the legal question is the value of the thing misrepresented or omitted. Its boundary is the market: willingness to pay is a demand-side quantity, and a market price is set by demand and supply together, so a conjoint result does not automatically equal the price premium the class actually paid. The methodological debate over when the translation is defensible, and what supply-side information it requires, is live and jurisdiction-inflected. The practical guidance is to know which question the case needs answered, because conjoint answers the first question well and the second only with additional structure that must itself be defended.
It has to be carried honestly to the end, and doing so is a strength rather than a concession. A sampling-based aggregate arrives as an estimate with an interval, and the parties then argue about which point in the interval an award should reflect, a legal and strategic question layered on the statistics. What the expert controls is the integrity of the interval itself: stating it prominently rather than burying it, resisting the temptation to present the point estimate as exact, and showing how the estimate moves under reasonable design variations. Precision theater, an extrapolated figure quoted to the dollar from a sample with a wide interval, is a small dishonesty that rebuttal experts convert into a large credibility problem. The tribunal can weigh honest uncertainty; it cannot forgive concealed uncertainty.
Describe what has to be measured and for whom. The Institute will help you see which instrument fits and what its design must survive.