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

Why certification is the case

In most class actions, the certification decision determines the outcome. A certified class of ten million consumers presents an aggregate exposure that no defendant will take to a jury, and the case settles. An uncertified case leaves each plaintiff with a claim worth less than the cost of pursuing it, and the case ends.

This is well understood, and it explains why certification proceedings look nothing like ordinary motion practice. They involve months of discovery, competing expert reports, depositions of those experts, evidentiary hearings that resemble trials, and briefing that runs to hundreds of pages. It also explains why the doctrine has developed the way it has: courts, aware that certification is effectively dispositive, have required increasingly rigorous scrutiny of what plaintiffs must actually prove.

The whole structure sits on Federal Rule of Civil Procedure 23, which requires numerosity, commonality, typicality, and adequacy under Rule 23(a), and then one of the three categories in Rule 23(b) — of which (b)(3), requiring predominance and superiority, is the battleground for damages classes.

The rigorous analysis

The governing standard is that a court must conduct a rigorous analysis to determine whether the Rule 23 prerequisites are satisfied, and that this analysis frequently overlaps with the merits.

Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), is the decision that made this operational. Addressing a proposed class of 1.5 million female employees alleging discrimination, the Court held that commonality under Rule 23(a)(2) requires more than a common question:

"What matters to class certification is not the raising of common 'questions' — even in droves — but rather the capacity of a class-wide proceeding to generate common answers apt to drive the resolution of the litigation."

The Court held that the plaintiffs needed "significant proof" of a general policy of discrimination, and found the sociological evidence offered insufficient. It also rejected "trial by formula" — the proposal to determine liability and damages for the class by extrapolating from a sample of individual claims — as inconsistent with the defendant's right to litigate its defences to individual claims.

Wal-Mart also confirmed that a court may not decline to consider the merits merely because they overlap with the certification question: "Frequently that 'rigorous analysis' will entail some overlap with the merits of the plaintiff's underlying claim. That cannot be helped."

The Third Circuit's decision in In re Hydrogen Peroxide Antitrust Litigation, 552 F.3d 305 (3d Cir. 2008), had already set out the framework that most courts now follow: the plaintiff must prove the Rule 23 requirements by a preponderance of the evidence; the court must resolve factual disputes bearing on certification, including disputes between experts; and a court may not simply accept a plaintiff's expert's assertion that common proof will be available.

Comcast and the damages model

Comcast Corp. v. Behrend, 569 U.S. 27 (2013), is the case defendants cite most and the case whose scope is most contested.

The plaintiffs alleged four antitrust theories. The district court certified on only one — the "overbuilder" theory. But the plaintiffs' damages model calculated damages resulting from all four theories combined, without isolating the overbuilder effect.

The Court reversed certification, holding that the model "failed to measure damages resulting from the particular antitrust injury on which petitioners' liability in this action is premised." A model that measures something other than the damages attributable to the certified theory cannot establish predominance:

"A model purporting to serve as evidence of damages in this class action must measure only those damages attributable to that theory. If the model does not even attempt to do that, it cannot possibly establish that damages are susceptible of measurement across the entire class for purposes of Rule 23(b)(3)."

What Comcast holds is narrow: the damages model must be tied to the liability theory that is actually being certified.

What defendants argue it holds is broader: that damages must be capable of measurement on a class-wide basis, and that individualized damages defeat predominance.

What most courts have concluded is closer to the narrow reading. The prevailing view is that individualized damages calculations do not, standing alone, defeat predominance, provided liability and injury can be established with common proof. Courts routinely certify classes where damages will require individual computation, using bifurcation, claims processes, or aggregate approaches. The dissent in Comcast said exactly this, and lower courts have largely followed it.

The practical drafting consequence for plaintiffs is disciplined: the damages model must be built for the theory actually pleaded and certified, and must be able to isolate that theory's effect. A model built on a broader theory, or that cannot separate a rejected theory's contribution, is a Comcast problem.

Predominance in practice

Predominance asks whether questions common to the class predominate over questions affecting only individual members. In practice the inquiry concentrates on injury — whether the plaintiff can show, with evidence common to the class, that each member suffered harm.

In antitrust cases, the question is whether an overcharge can be shown to have been passed through to all or nearly all class members. Plaintiffs typically offer a regression model estimating the but-for price and showing that class members paid more. Defendants attack the model's specification, its handling of heterogeneity across products, customers, and time, and — most effectively — its treatment of class members who show no overcharge.

In consumer cases, the question is whether exposure to and reliance on a misrepresentation can be shown class-wide, and whether the price premium attributable to the misrepresentation can be measured. Conjoint analysis has become the standard tool and the standard target.

In employment cases, Wal-Mart requires a common policy or practice capable of generating a common answer.

In securities cases, the Basic presumption of reliance does the work, subject to rebuttal — discussed below.

Uninjured class members

The most heavily litigated question in modern certification practice is what to do about class members who suffered no injury.

The problem is real. A class defined as all purchasers of a product will include people who would have bought at the higher price anyway, people whose contracts insulated them, and people whose transactions the challenged conduct never touched. If the number is small, courts generally certify and address it through claims administration. If the number is large, individualized inquiries may swamp the common ones.

Olean Wholesale Grocery Cooperative, Inc. v. Bumble Bee Foods, LLC, 31 F.4th 651 (9th Cir. 2022) (en banc), addressed this directly. The court held that Rule 23 does not impose a freestanding requirement that a class contain no uninjured members, that the question is whether the plaintiff can prove injury through common evidence, and that a district court must resolve the factual dispute between experts about how many members are uninjured. It declined to adopt a bright-line percentage threshold, while acknowledging that at some point the number of uninjured members would defeat predominance.

The related constitutional question comes from TransUnion LLC v. Ramirez, 594 U.S. 413 (2021), which held that every class member must have Article III standing to recover damages: "Every class member must have Article III standing in order to recover individual damages." The Court left open how and when that must be established, and the lower courts have divided on whether it must be shown at certification or can be addressed later.

The practical upshot for a defendant is that the uninjured-member argument is now the strongest available at certification, and it must be made with an expert who can quantify the number rather than merely assert that some exist.

Representative and statistical evidence

Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442 (2016), addressed whether statistical evidence may be used to establish class-wide liability. Employees sought compensation for donning and doffing time, and, because the employer had kept no records, offered a study of average times.

The Court affirmed certification and the use of the evidence, on a principle that is more limited than either side usually acknowledges:

"Whether and when statistical evidence can be used to establish classwide liability will depend on the purpose for which the evidence is being introduced and on the elements of the underlying cause of action."

The key was that each employee could have relied on the same study in an individual action to fill the evidentiary gap created by the employer's failure to keep records. Representative evidence admissible in an individual case does not become inadmissible because the case is a class action.

The Court distinguished Wal-Mart, where the sample could not have established liability in any individual case because the employees held different jobs under different supervisors at different stores. The distinction is between a sample that substitutes for evidence a plaintiff would otherwise have to produce, and a sample that papers over genuine dissimilarity.

Securities: the presumption and its rebuttal

Securities fraud class actions have their own certification architecture, built on Basic Inc. v. Levinson, 485 U.S. 224 (1988), which established a rebuttable presumption of reliance based on the fraud-on-the-market theory: in an efficient market, the price reflects public misstatements, and a purchaser who bought at the market price relied on them.

Erica P. John Fund, Inc. v. Halliburton Co., 563 U.S. 804 (2011), held that plaintiffs need not prove loss causation at certification — that is a merits question.

Halliburton Co. v. Erica P. John Fund, Inc., 134 S. Ct. 2398 (2014) ("Halliburton II"), declined to overrule Basic but held that a defendant may rebut the presumption at the certification stage by showing a lack of price impact — that the alleged misrepresentation did not in fact affect the stock price.

Goldman Sachs Group, Inc. v. Arkansas Teacher Retirement System, 594 U.S. 113 (2021), refined the inquiry. The Court held that: courts should consider all record evidence relevant to price impact, including evidence bearing on materiality, even if it overlaps with the merits; the generic nature of an alleged misrepresentation is important evidence of price impact, particularly in an inflation-maintenance case where the alleged corrective disclosure is far more specific than the alleged misstatement; and the defendant bears the burden of persuasion on price impact by a preponderance of the evidence.

Amgen Inc. v. Connecticut Retirement Plans & Trust Funds, 568 U.S. 455 (2013), held that materiality need not be proved at certification, because it is an objective question common to the class — if the statement was immaterial, the entire class fails together, which is exactly what predominance contemplates.

The result is a certification hearing that is, in substance, a mini-trial on price impact, conducted through event study evidence.

Daubert at the certification stage

Whether the admissibility standard of Federal Rule of Evidence 702 applies in full at certification was, for years, an open question. The answer that has emerged is essentially yes, at least as to the expert evidence offered to satisfy Rule 23.

The reasoning follows from Wal-Mart and the rigorous analysis requirement. If a court must resolve factual disputes bearing on certification, and if the plaintiff's showing depends on an expert model, the court cannot resolve those disputes while treating the model's reliability as a merits question for later. Wal-Mart itself contained a pointed observation about the district court's suggestion that Daubert did not apply at certification: "We doubt that is so."

Most circuits now hold that a court must conduct a full Daubert analysis of expert evidence that is critical to the certification decision. A minority approach applies a somewhat more limited or "tailored" inquiry. Either way, the practical position is the same: a defendant should move to exclude, and a plaintiff should assume the model will be tested.

The governing framework comes from Daubert v. Merrell Dow Pharmaceuticals, Inc., 509 U.S. 579 (1993), which assigned trial courts a gatekeeping role and identified factors bearing on reliability — testability, peer review, known error rate, standards controlling the technique's operation, and general acceptance. Kumho Tire Co. v. Carmichael, 526 U.S. 137 (1999), extended the gatekeeping function to all expert testimony, not only scientific, and emphasized the flexibility of the inquiry. General Electric Co. v. Joiner, 522 U.S. 136 (1997), established abuse-of-discretion review and — importantly for these disputes — held that a court may exclude an opinion where there is "simply too great an analytical gap between the data and the opinion proffered."

The 2023 amendments to Rule 702 clarified two points that had drifted in application: the proponent must establish admissibility by a preponderance of the evidence, and the court must find that the expert's opinion reflects a reliable application of the methodology to the facts. Both amendments strengthen the gatekeeping function and both are regularly invoked in certification practice.

The methods that get fought over

Regression and the antitrust overcharge

The standard antitrust class model is a multiple regression estimating the but-for price, with the difference between actual and but-for prices constituting the overcharge.

What defendants attack:

  • Specification. Which variables are included, which are omitted, and whether omitted variable bias drives the result. This is the most common and most productive line.
  • Benchmark period. Whether the "clean" period used to estimate the but-for relationship was in fact unaffected by the conduct.
  • Aggregation. Whether pooling across heterogeneous products, customers, or regions masks the fact that many class members show no overcharge. A model that produces an average positive overcharge is consistent with half the class being uninjured.
  • False positives. Running the model on data from a period when no conspiracy is alleged, and showing that it still generates an "overcharge," is devastating when it works.
  • Individual results. Requiring the expert to produce class-member-level results, and counting how many are negative or statistically indistinguishable from zero, is the core of the uninjured-member argument.
  • Pass-through. In indirect purchaser cases, whether the overcharge was passed down uniformly.

What plaintiffs must do: build the model to the certified theory; test its robustness across specifications; be able to produce member-level results and explain them; and address heterogeneity directly rather than by aggregation.

Conjoint analysis in consumer cases

Conjoint analysis estimates the value consumers place on individual product attributes by presenting choices among hypothetical products. In mislabeling cases it is used to estimate the price premium attributable to the challenged claim.

What defendants attack:

  • Demand only, not supply. A conjoint survey measures willingness to pay. Market price is set by supply and demand together. A model that ignores the supply side measures something other than the price premium the market would have produced — this is the single most effective attack and it has defeated many conjoint models.
  • Survey design. Attribute selection, level definitions, the number of choice tasks, focus effects that make respondents attend to attributes they would ignore in a store, and the absence of a "none" option.
  • Population. Whether respondents match actual purchasers.
  • Market simulation. Whether the model accounts for competitor responses and actual prices.
  • Fit to the liability theory. Comcast again: does the model measure the premium attributable to the specific misrepresentation certified?

Event studies in securities cases

An event study measures the abnormal return on days of alleged misstatements and corrective disclosures, controlling for market and industry movements.

What is fought over:

  • Event windows. One day, two days, intraday. Longer windows capture more noise.
  • Confounding information. Whether other news released the same day explains the movement. This is the defence's principal tool and it works when the corrective disclosure came bundled with an earnings miss.
  • Statistical significance. The threshold applied, and the treatment of results significant at 10% but not 5%.
  • Inflation maintenance. Where the alleged misstatement did not move the price up but allegedly maintained an existing inflation, the Goldman Sachs mismatch analysis applies: if the alleged misstatement is generic and the corrective disclosure is specific, that is important evidence that the disclosure's price impact did not result from the misstatement.
  • Market efficiency. Trading volume, analyst coverage, bid-ask spread, market makers, float, and — most importantly — evidence of a cause-and-effect relationship between unexpected news and price movements.

Rule 23(b)(1) and (b)(2), and why they matter less than they should

Most of the discussion above concerns (b)(3) damages classes. The other categories deserve a note, because they are sometimes the better route and are frequently overlooked.

Rule 23(b)(1)(B) covers limited fund classes — where individual adjudications would be dispositive of the interests of absent members or would substantially impair their ability to protect those interests. Ortiz v. Fibreboard Corp., 527 U.S. 815 (1999), constrained this sharply in the mass tort context, requiring that the fund be shown to be limited by something other than the parties' agreement, that the whole of it be devoted to the claims, and that class members be treated equitably among themselves.

Rule 23(b)(2) covers classes seeking injunctive or declaratory relief where the party opposing the class has acted on grounds generally applicable to the class. Wal-Mart held that (b)(2) does not authorize certification where each class member would be entitled to an individualized award of monetary damages, and that backpay claims could not be shoehorned in as "incidental" relief. Monetary relief is available under (b)(2) only where it is truly incidental to the injunctive relief — a standard that excludes most damages theories.

Issue certification under Rule 23(c)(4). A court may certify particular issues for class treatment even where the case as a whole would not qualify. This is a genuinely useful and underused tool: liability elements that turn entirely on the defendant's conduct — whether a product was defective, whether a policy existed, whether a statement was false — can be certified while causation and damages proceed individually. Circuits differ on whether predominance must be satisfied as to the case as a whole before issues may be certified, and this split matters for forum selection.

Amchem Products, Inc. v. Windsor, 521 U.S. 591 (1997), remains the foundational treatment of the tension between the settlement value of aggregation and the requirements of Rule 23. It held that settlement is a factor a court may consider — the manageability concerns of trial disappear — but that the other Rule 23 requirements, including predominance and adequacy, apply with full force. A settlement class must satisfy Rule 23 for reasons independent of whether the case could be tried.

The practical importance of Amchem and Ortiz is that a court reviewing a settlement class must actually apply Rule 23 rather than approve an aggregation the parties have agreed on. Intra-class conflicts — between present and future claimants, between those with different injuries, between those whose claims are strong and weak — require structural protections, usually separate representation and separate counsel.

The strategic architecture of a certification fight

Both sides make a small number of decisions early that shape everything.

For plaintiffs:

Define the class to the evidence. A class definition drawn broadly to maximize the recovery invites the uninjured-member attack. A narrower definition — excluding categories where injury cannot be shown class-wide, excluding periods where the evidence is weak — is often the difference between certification and denial. Plaintiffs' counsel resist narrowing because it reduces the settlement value, and then lose entirely.

Plead the theory the model can measure. Comcast means the pleading and the damages model must be built together. A complaint with four theories and a model that measures their combined effect is a certification denial waiting to happen.

Retain the expert early. The expert should be involved in framing the theory, not handed a complaint and asked to support it. The strongest models are built with the data in hand.

Take the data seriously. Certification-stage discovery of transaction data is the foundation of the model, and disputes about its completeness and usability consume months. Start early and be specific about what is needed.

Prepare for member-level results. Assume the court will require the expert to produce results for each class member and to say how many show no injury. An expert who cannot do this, or who resists doing it, has a problem.

For defendants:

Attack the model, not the merits. The temptation is to argue that the plaintiffs are wrong. The productive argument is that they cannot prove it with common evidence.

Find the uninjured. Quantify them. A defence expert who can say that 31% of the class shows no overcharge under the plaintiff's own model has made the strongest available argument.

Run the false positive test. Apply the plaintiff's model to a period when no conduct is alleged. If it generates an overcharge, say so.

Move to exclude. File a Rule 702 motion targeting the certification model specifically, and file it with the certification opposition rather than later.

Preserve everything. Certification denials and grants are reviewable under Rule 23(f) on a discretionary basis, and the record made at certification is the record on appeal.

Consider the alternative structures. Where certification is likely, the fight may be better spent on narrowing the class definition, on issue certification that leaves causation individual, or on the class period.

Worked example one: the antitrust overcharge

Beatriz Salgado represents a proposed class of direct purchasers of an industrial chemical, alleging a price-fixing conspiracy over six years. Her expert, an industrial organization economist, builds a regression estimating but-for prices using a benchmark period before the alleged conspiracy.

The plaintiff's model. Transaction-level data on 340,000 purchases by 1,900 customers. Controls for input costs, capacity utilization, demand, and customer-specific characteristics. The model estimates an average overcharge of 9.4%.

The defence attack, in the order it lands:

Member-level results. The defence expert requests and obtains the model's fitted values for each customer. Under the plaintiff's own specification, 22% of customers show a negative overcharge — they paid less than the but-for price. The defence argues these members are uninjured and that individualized inquiry is required to identify them.

Benchmark contamination. The defence shows that the pre-conspiracy benchmark period includes eighteen months during which, on the plaintiffs' own documents, at least some coordination was occurring. If the benchmark is contaminated, the but-for price is inflated and the overcharge understated — or the model is simply unreliable.

Aggregation. The chemical is sold in three grades to two customer types under both contract and spot arrangements. The defence expert runs the model separately for each combination and shows materially different results, including two segments with no statistically significant overcharge.

False positives. Applied to a two-year period four years before the alleged conspiracy began, the model generates a 6.1% "overcharge."

Beatriz's response, and it is the right one:

She has her expert address the negative results directly rather than dismissing them. He explains that individual-level fitted values from an aggregate model are not estimates of individual injury — they reflect estimation error around a common effect — and offers an alternative specification that produces positive overcharges for 96% of customers.

She narrows the class: two grades rather than three, contract purchasers only, and a class period beginning eighteen months later, which removes the contaminated benchmark.

She confronts the false positive result: her expert shows that the earlier period included a documented capacity shock that the model's controls do not capture, and that adding a control eliminates the result.

The outcome. Certification granted for the narrowed class. The narrowing cost roughly 30% of the potential class recovery and saved the case.

Worked example two: the mislabeling class

Owen Fitzhugh defends a food manufacturer against a class alleging that a "no artificial ingredients" claim was false. The plaintiffs' expert offers a conjoint analysis estimating a 12% price premium attributable to the claim.

The attack.

Supply side. Owen's expert demonstrates that the conjoint measures willingness to pay and says nothing about the market price that would have obtained. The actual market includes competitors, capacity constraints, and retailer pricing decisions, none of which the model incorporates. A survey of consumers cannot establish what price the market would have set.

Real-world evidence. Owen's expert compares the actual retail prices of the product with and without the claim on the label across a period when the labelling changed in some regions. The observed price difference is under 2%.

Survey design. The conjoint presented the "no artificial ingredients" attribute in every choice task, focusing respondents on it in a way no shopping environment does. The defence runs a replication with the attribute embedded among twelve others and obtains a premium of 3%.

Class composition. Sales data shows that 40% of purchases were made at a promotional price, and a further segment through a channel where the label was not visible at the point of sale.

The plaintiff's response is to argue that conjoint has been accepted in many courts, which is true and insufficient — the question is whether this conjoint, applied to this claim, reliably measures the premium.

The outcome. The court excludes the conjoint model under Rule 702, finding that its failure to account for supply-side factors made it incapable of measuring a market price premium, and denies certification for want of a class-wide damages methodology tied to the liability theory.

Worked example three: the price impact hearing

Hana Kobayashi is lead counsel for a securities class alleging that a bank's repeated statements about its conflict-of-interest controls were false, and that the truth emerged when enforcement actions were announced, causing a 12% decline.

The certification hearing is, in substance, a two-day trial on price impact.

The defence case, built on Goldman Sachs:

The alleged misstatements — "we have extensive procedures to identify and address conflicts" — are generic. The alleged corrective disclosures are highly specific enforcement actions concerning particular transactions. The mismatch between the generality of the statements and the specificity of the disclosures is, under Goldman Sachs, important evidence that the price decline did not result from the alleged misstatements.

The defence event study shows no statistically significant price movement on any of the dates the statements were made — consistent with the statements having no price impact.

On the disclosure dates, the defence expert identifies confounding information: two of the three dates also carried earnings news and an analyst downgrade.

The plaintiff's case:

Hana's theory is inflation maintenance: the statements did not raise the price, they prevented it from falling. On that theory, the absence of a price increase on the statement dates proves nothing.

Her expert shows that the enforcement actions concerned precisely the conduct the statements denied, and that analyst commentary on the disclosure dates tied the decline to the revelation about conflicts specifically.

On confounding, her expert isolates the intraday price movement following the enforcement announcement, before the earnings release.

The burden. After Halliburton II and Goldman Sachs, the defendant bears the burden of persuasion on price impact by a preponderance. That allocation matters: where the evidence is genuinely in equipoise, the plaintiff wins.

The outcome. Certification granted. The court finds the mismatch argument weakened by the specificity of the analyst commentary tying the decline to the conflicts issue, and holds the defendant has not carried its burden. The case settles four months later.

Managing the expert relationship

The expert is the case at certification, and how counsel works with them determines the quality of the product.

Engage early and on the right question. An expert brought in after the complaint is filed and asked to support a theory is in a worse position than one who helped frame the theory against the available data. The best models are built by people who saw the data first.

Give them the data problems. Transaction data is messy — inconsistent product codes, missing fields, merged entities, changed systems. An expert who discovers these in a deposition rather than in the workroom is in trouble.

Test the model against the attack. Before the report is served, have the model run under alternative specifications, on subsets, and on periods when no conduct is alleged. If it breaks, it is better to know.

Require member-level output. Assume the court will ask. An expert who says the model does not produce individual estimates is giving a defence argument its best line.

Watch the drafts. Rule 26 protects draft reports and most communications between counsel and a testifying expert, but the protection has limits — facts and data provided by counsel and assumptions counsel supplied that the expert relied on are discoverable. Manage the file accordingly, and do not put in writing anything you would not want read aloud.

Prepare for the hearing, not the deposition. Certification hearings increasingly involve live testimony, sometimes with concurrent evidence — both experts on the stand together. An expert who performs well in a deposition and poorly under a judge's direct questioning is a liability.

Consider a second expert. In large cases, a consulting expert who is not testifying can stress-test the testifying expert's work without creating discoverable material. The cost is real and so is the benefit.

Appellate review under Rule 23(f)

Rule 23(f) permits a court of appeals to grant permission to appeal an order granting or denying certification, in its discretion, on a petition filed within fourteen days.

What gets granted. Petitions raising an unsettled legal question of general importance; petitions where the certification order effectively ends the case — a "death knell" for the plaintiff on denial, or irresistible settlement pressure on the defendant after grant; and petitions where the district court's order is manifestly erroneous.

The fourteen days are jurisdictional and are not extended by a motion for reconsideration in most circuits. This has cost parties their appeal.

What the record must contain. The court of appeals reviews the district court's rigorous analysis, and a district court that made no findings on a contested expert dispute is vulnerable. Both sides should make sure the record contains the expert evidence, the objections, and a request for findings. A defendant that did not move to exclude has weakened its position on appeal.

Standard of review. Abuse of discretion for the certification decision; abuse of discretion for evidentiary rulings under Joiner; and de novo for legal questions, including whether the district court applied the correct standard. The most successful appellate arguments are framed as legal error — the court applied the wrong predominance standard, or failed to resolve a factual dispute it was required to resolve — rather than as disagreement with the weighing.

Where the doctrine is unsettled

Three questions divide the courts and are worth tracking, because they determine forum value.

When must every class member's Article III standing be established? TransUnion held that every member must have standing to recover damages, but did not say when that must be shown. Some courts require a showing at certification; others hold it can be addressed at judgment or through the claims process; others distinguish between named plaintiffs and absent members. The answer materially affects the uninjured-member argument.

Does the presence of uninjured class members defeat predominance, and at what threshold? Olean declined a bright line. Some courts have suggested a de minimis tolerance; others have denied certification where the proportion was substantial. There is no accepted number, and expert quantification is therefore decisive.

Must predominance be satisfied for the case as a whole before issues may be certified under Rule 23(c)(4)? Circuits divide. The more permissive approach allows issue certification where common issues predominate within the certified issues, which makes (c)(4) a far more powerful tool.

A fourth, quieter question: how far the Comcast damages-model requirement extends beyond antitrust. The broad reading has not prevailed, but it is still argued, and the answer varies by circuit and by claim type.

A closing observation about what certification is for

There is a persistent argument, made from both directions, that certification practice has become detached from its purpose. Defendants say courts certify cases that could never be tried, producing settlements driven by exposure rather than merit. Plaintiffs say the rigorous analysis has become a merits trial conducted before discovery is complete, with a burden of proof that no individual plaintiff would face.

Both complaints have force, and the doctrine as it stands is a compromise that satisfies nobody. What it does provide is a clear instruction to practitioners: the certification decision turns on whether the plaintiff can prove injury with evidence common to the class, and that question is answered by experts under an admissibility standard that is applied seriously.

Everything else — the pleading, the class definition, the discovery plan, the choice of theory, the choice of expert — should be organized around that single question from the first week of the case. Plaintiffs who plead broadly and build the model afterward lose. Defendants who argue the merits instead of the model lose. The cases are won by the side that understood, early, exactly what would have to be proved with common evidence, and built everything toward it.

Related documents