Document type: Checklist Practice area: Litigation — Class Actions Jurisdiction: United States Last reviewed: 5 September 2026

Section 1 — Selecting and engaging the expert

  • Subject-matter fit — industry experience, not only credentials
  • Capacity to handle the data volume; team and infrastructure confirmed
  • Deposition and hearing transcripts reviewed for performance under questioning
  • Prior exclusions checked — and the grounds understood
  • Prior inconsistent positions searched (opposing counsel will find them)
  • Conflicts cleared, including the expert's firm and other engagements
  • Engaged before the theory is fixed, so the model and the pleading are built together
  • Consulting expert considered for stress-testing without discoverable output
  • Engagement letter, scope, and budget agreed

Section 2 — The data foundation

  • What data exists, in what systems, for what periods, in what condition
  • Legacy systems, migrations, and acquisitions identified as sources of gaps
  • Data dictionary obtained: field names, definitions, permissible values, units, meaning of nulls
  • Data loaded and validated before anything is filed that depends on it
  • Totals reconciled against public financials or internal reports
  • Gaps, duplicates, impossible values, and mid-period unit changes identified
  • Anomalies raised quickly and specifically with the producing party
  • Production protocol agreed: format, delivery, anomaly resolution, named technical contact on each side
  • Protective order and expert acknowledgment process completed — allow weeks
  • Code, intermediate datasets, and specification runs preserved

Section 3 — Building the model

  • Fit to the certified theory — the model measures damages attributable to the pleaded theory and only that theory (Comcast)
  • Mechanism of class-wide injury stated in one sentence
  • Specification documented: every variable, and why it is included
  • Omitted variables considered and addressed
  • Functional form justified
  • Benchmark period tested for contamination — is the "clean" period actually clean?
  • Heterogeneity addressed directly rather than by aggregation
  • Sensitivity run across defensible alternative specifications
  • Model run on a period with no alleged conduct (false positive test) — before serving, not after
  • Member-level results produced, and the number showing no injury known
  • Results reproducible from the code by a third party

Section 4 — The uninjured member question

  • Number and proportion of class members showing no injury, quantified
  • Explanation prepared for negative or insignificant individual estimates
  • Alternative specification available that reduces the proportion, if defensible
  • Class definition narrowing considered to exclude categories where injury cannot be shown
  • Article III position addressed — after TransUnion, every member must have standing to recover damages
  • Circuit's approach to when standing must be established confirmed
  • Claims-process or trial-plan mechanism identified for screening uninjured members

Section 5 — Defence attack sequence

  • Obtain member-level fitted values from the plaintiff's model and count the uninjured
  • Attack specification: omitted variables, functional form, control selection, sensitivity
  • Test benchmark period integrity against the plaintiff's own documents
  • Disaggregate: run the model by product, customer type, channel, and period
  • Run the false positive test on a no-conduct period
  • Test Comcast fit — does the model measure the certified theory?
  • Identify individualized issues: negotiation, exposure, knowledge, reliance, mitigation
  • Replicate the model from the produced code — a model that cannot be replicated is attackable
  • Quantify rather than assert — "22%" beats "some"

Section 6 — Regression evidence

  • But-for price estimation method stated
  • Control variables listed with justification for each
  • Benchmark or before-and-after period defined and defended
  • Standard errors and clustering appropriate to the data structure
  • Statistical significance thresholds stated and applied consistently
  • Pass-through analysis where indirect purchasers are involved
  • Aggregation level justified against the heterogeneity in the data

Section 7 — Conjoint analysis

  • Supply side addressed — willingness to pay is not market price; this is the most effective attack
  • Market simulation incorporating competitor responses and actual prices
  • Attribute selection and level definitions justified
  • Focus effects considered — was the challenged attribute presented more prominently than in a real purchase?
  • "None" option included
  • Respondent population matched to actual purchasers
  • Survey instrument, data, and code produced
  • Fit to the specific misrepresentation certified
  • Real-world price comparison available where labelling varied

Section 8 — Event studies (securities)

  • Event windows stated and justified
  • Confounding information on each event date identified and addressed
  • Statistical significance threshold applied consistently
  • Market efficiency factors: volume, analyst coverage, bid-ask spread, market makers, float, and cause-and-effect evidence
  • Inflation maintenance theory articulated where no price increase occurred on the statement dates
  • Generic-statement mismatch analysis addressed (Goldman Sachs): is the alleged misstatement generic while the corrective disclosure is specific?
  • Burden allocation understood — the defendant bears persuasion on price impact by a preponderance
  • Analyst commentary tying the decline to the alleged subject collected

Section 9 — The Rule 702 motion

  • Filed — a defendant that does not move has conceded reliability and weakened the appeal
  • Timed with the certification opposition, not afterward
  • Framed as gatekeeping, not as a battle of experts
  • Grounds, in order of strength: model does not measure what the theory requires; methodology fails on its own terms; unreliable application to the facts; expert cannot replicate or explain
  • 2023 Rule 702 amendments invoked: proponent bears preponderance; opinion must reflect a reliable application of the methodology
  • Weight-versus-admissibility line addressed head-on
  • Plaintiff's opposition: methodology standard in the field, peer reviewed, criticisms go to weight, expert responds substantively to each

Section 10 — Expert depositions

  • Code, workpapers, intermediate datasets, and discarded specifications requested
  • All prior testimony and publications obtained
  • Questions prepared on: what the model measures; what it assumes; what happens if a control is removed; how many members show no injury; whether the model was run on other periods
  • Rule 26 protection boundaries understood — drafts and most counsel communications protected; facts and data provided by counsel and assumptions relied on are not
  • Preparation covers judicial questioning, not only cross-examination
  • Expert prepared to concede what is genuinely conceded

Section 11 — The hearing

  • Live testimony anticipated and prepared for
  • Concurrent expert evidence possibility considered — rewards genuine understanding, punishes scripting
  • Demonstratives: scatter with fitted line; member-level distribution; sensitivity table; false positive chart
  • Glossary of technical terms supplied to the court
  • Single-sentence statement of what the model does, prepared and rehearsed
  • Findings requested on each contested methodological dispute
  • Trial plan and draft verdict form available

Section 12 — Record and appeal

  • Expert reports, objections, Rule 702 motion, and rulings all in the record
  • Findings obtained on the contested expert disputes
  • Rule 23(f) deadline — fourteen days — calendared the day the order issues
  • Note: in most circuits a motion for reconsideration does not extend it
  • Petition framed as legal error rather than disagreement with the weighing
  • Decertification arguments preserved for renewal if merits discovery changes the picture

Related documents