Expert Evidence at the Certification Stage

A certification motion is decided on a record, and in most cases the decisive part of that record is an expert report. Whether members can be identified from available data, whether a common impact can be shown across a group, and whether damages are capable of measurement on a classwide basis are all questions answered by methodology rather than by testimony about what happened.

Why experts arrive so early

The prerequisites are matters to be proved, so a proponent who asserts that a common question can be answered classwide has to show how. In a pricing case that means an analysis capable of separating the effect of the challenged conduct from everything else moving the price. In an employment case it may mean statistical evidence about the operation of a policy. In a consumer case it may mean an analysis of transaction records showing that members can be identified from the defendant's own data.

Defendants respond in kind, and the certification record fills with opposing reports before any merits discovery is complete. That sequencing creates its own difficulty: an expert is asked to opine on the feasibility of a method using data that has been only partly produced, and the opposing expert attacks the opinion for resting on an incomplete record. Courts manage this by scheduling certification discovery separately and by defining what data must be produced to support the analysis.

The admissibility standard applied

The evidentiary rule permits opinion testimony from a qualified witness where the proponent demonstrates that it is more likely than not that the testimony will help the trier of fact, that it is based on sufficient facts or data, that it is the product of reliable principles and methods, and that the expert's opinion reflects a reliable application of those principles and methods to the facts. Each condition is the proponent's to establish, and the last of them addresses application rather than method in the abstract.

Two features of that standard carry particular weight in this setting. The proponent bears the burden on each condition, so an unrebutted opinion is not admissible merely because nobody contradicted it. And the reliable-application condition means that a method accepted in a field can still be excluded where the expert applied it to data that cannot support it, which is the usual objection where a model is run on records the defendant produced for a different purpose.

Applying that standard at certification raises a sequencing question the rules do not resolve. Some courts perform the full analysis on any opinion the decision will rest on, reasoning that a finding cannot be supported by evidence the court would exclude. Others screen more narrowly, addressing whether the opinion is adequate for the limited purpose of the certification question. The divergence is real and it affects how motions are briefed in different circuits.

The fit between the model and the theory

The most consequential failure is not unreliability but mismatch. A damages model offered to show that injury can be measured classwide must measure only the injury attributable to the theory on which liability would be established. Where a class is certified on one theory and the model aggregates the effects of several kinds of conduct, including conduct that was never certified or was dismissed, the model cannot support the finding it is offered for.

The point is easily obscured because the defect is invisible from inside the methodology. A regression can be specified correctly, run on complete data and produce a defensible number while still answering a question nobody is litigating. Courts therefore start with the certified theory, ask what the model measures, and compare. Where the two diverge, the failure enters the analysis as a predominance problem, as described in predominance and superiority in a damages class.

Managing competing reports

Courts have several tools and use them unevenly. An evidentiary hearing with live testimony is the most common where the dispute is genuinely methodological, because cross-examination exposes assumptions that reports state without defending. Some courts appoint a master or a court-appointed expert to assist with technical questions, particularly where both sides' analyses rest on the same underlying data and differ only in specification.

Scheduling is the quieter tool and often the most effective. Courts that require simultaneous exchange of reports, followed by rebuttals and then depositions confined to methodology, produce records in which the actual disagreement is visible. Courts that allow sequential filing without a defined scope receive reports that talk past each other, and the resulting hearing is spent establishing what is in dispute rather than resolving it.

Another approach is to narrow the question. Where the disagreement concerns whether a model can be built rather than what it will show, a court may certify on the strength of a described methodology and revisit the issue if the completed analysis fails. That is consistent with the power to alter or amend a certification order before final judgment, and it explains why decertification motions are a normal feature of the case rather than an extraordinary one.

What the certification record does afterward

The reports produced for certification do not disappear when the order issues. They frame the merits, they are the starting point for the trial experts, and they are the material an appellate court reads when reviewing the order. A permissive appeal, where granted, is decided on this record rather than on a later one, which is examined in appealing a certification decision immediately.

The record also shapes settlement, since a model that survived scrutiny establishes the outer boundary of exposure and one that did not removes it. That is why the sufficiency of the certification evidence is often the last contested question before negotiation begins, and why the prerequisites it addresses are set out in the four threshold requirements.