Calculating Class-Wide Damages in Equal Pay Litigation

Class-wide damages in an equal pay or pay equity case are built from two inputs: the adjusted pay disparity the statistical analysis measures, and each class member’s own pay history inside the period the governing statute allows. The first is one number, or a small set of them. The second is a payroll file. The work of the damages expert is to combine them, period by period and member by member, under rules that differ by statute, and to present the result in a form that can be tested at certification, used at mediation, and divided at settlement. This note sets out how that is done. The individual Equal Pay Act measure that it scales up is described in how a forensic economist measures a pay disparity, and the practice is on our pay equity and wage discrimination page.

From a regression coefficient to a dollar figure

There are two routes from the regression to damages. The first applies the class-wide adjusted disparity to each member’s actual pay in each pay period: if the model says women were paid 3.2 percent less than otherwise similar men, each woman’s but-for pay in each period is her actual pay divided by 0.968, and her back pay for the period is the difference. The second route matches each member with comparators in the same job, level and location and measures her shortfall against their pay directly, period by period. The first is simpler, uses every member’s own pay, and follows directly from the common proof; the second is more individualized and is often preferred where the class is small, the jobs are few, or the court has signaled that a single percentage will not do.

Either route can be defended, and either can be attacked, on the same ground: whether it measures the damages the liability theory implies. That is the requirement of Comcast Corp. v. Behrend, 569 U.S. 27 (2013), and it is why the damages model should be specified at certification rather than deferred. A model that cannot be described until after liability is found is a model the court cannot evaluate when it decides whether damages can be measured class-wide.

The statutory building blocks

The measure, the multiplier, the period and the people in the class all depend on which statute the claim is brought under. Most pay equity class actions plead more than one, and the model has to compute each separately because the answers differ.

Which of these apply, whether a violation was willful, and whether liquidated damages should be reduced are legal questions. The model does not decide them; it computes every scenario counsel identifies and keeps the components separate, so that a court or the parties can assemble the award under whichever rules govern.

Class and collective mechanics

The Equal Pay Act is enforced through the Fair Labor Standards Act’s collective action, in which each employee must opt in by filing a written consent, and the court may facilitate notice to potential members (Hoffmann-La Roche Inc. v. Sperling, 493 U.S. 165 (1989)). Title VII, § 1981 and the California statutes proceed under Rule 23, where members are in unless they opt out. The two can be combined in one case, and most circuits permit it. For damages the difference is practical: in a collective the class size is unknown until the opt-in period closes and each member’s limitations window depends on when her consent was filed, so the model has to be built to accept a roster rather than a count, and to compute each member’s window from her own consent date.

Three Supreme Court decisions frame what the model must show at certification. Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), rejected a “Trial by Formula” in which a sample of class members’ claims would be tried and the results extrapolated, because it would deprive the employer of the chance to raise individual defenses. Comcast requires the damages model to fit the liability theory. Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442 (2016), allows representative evidence where each member could have relied on it in an individual action, and left open how an aggregate award is to be distributed so that members who were not injured do not recover. In a pay case those cases point to the same design: a model that produces a class-wide figure from common proof and can also be read member by member.

The pattern-or-practice framework of International Brotherhood of Teamsters v. United States, 431 U.S. 324 (1977), supplies the procedure. Liability is tried first on the class-wide evidence. If a pattern or practice is found, individual relief is determined in a second stage in which each member is presumed to have been a victim and the employer bears the burden of showing otherwise. Franks v. Bowman Transportation Co., 424 U.S. 747 (1976), and Albemarle Paper Co. v. Moody, 422 U.S. 405 (1975), hold that back pay and seniority relief follow presumptively once liability is established. The damages model is built for that second stage: it has to produce each member’s figure, state the assumptions behind it, and be adjustable when the employer shows that a particular member’s pay was set for a reason the model did not capture.

Building the model, member by member

The computation runs over a grid: one row for each class member and each pay period from the start of her limitations window to the end of the damages period or her separation, whichever comes first. For each cell the model needs her actual pay, by component, and her but-for pay.

Pay components that move with base pay

A disparity in base pay is rarely the whole loss. Overtime is paid at a regular rate built on base pay, so a lower base means lower overtime for the same hours. Bonuses expressed as a percentage of salary, employer retirement contributions that match a percentage of pay, and equity grants sized to salary all carry the disparity forward. Each component is computed separately, from the plan documents and the payroll detail, and summed; a model that applies the percentage to base pay alone understates the loss for every member who received those components.

Compounding

Raises are usually set as a percentage of current pay, so a lower starting salary compounds. Applying the class-wide percentage to each period’s actual pay captures this automatically if the disparity is constant in percentage terms. Where the regression shows the gap widening with tenure, or where the claim is specifically about starting salaries, the alternative is to construct a but-for salary path: correct the starting salary and apply the employer’s own raise percentages forward. The two methods are reconciled in the workpapers and the report says which it used and why.

Promotions, transfers and separations

A member who was promoted mid-period moves to a new job and level, and her comparators change; the regression’s controls handle that if the model is estimated on employee-years rather than a snapshot. A member who left stops accruing back pay at separation unless a termination claim is also pleaded, in which case the back pay and front pay analysis takes over from the separation date. Members hired during the period enter the grid at hire. Each of these events is recorded in the personnel data the model runs on, which is one reason the data request asks for effective-dated job history rather than a current roster.

Limitations windows, interest and multipliers

Each claim’s window is applied to each member separately: an Equal Pay Act opt-in whose consent was filed in March 2026 has a different window from one whose consent was filed in September, and both differ from the Title VII window measured from the charge. Interest is computed on each period’s shortfall from the date it was owed. The liquidated multiplier under the federal and California Equal Pay Acts applies to the wages, not to the interest, and the model carries the three figures, wages, interest and liquidated damages, as separate columns. Title VII back pay carries interest; the statutory caps apply to compensatory and punitive damages, not to back pay, and are applied per claimant.

Aggregation and allocation

Summing the grid gives the class-wide figure for each scenario. The report presents the scenarios as a table rather than a single number: by statute, by willfulness assumption, by class period, and with and without the contested specification choices in the underlying regression. A mediation model built this way lets both sides see what each legal ruling is worth in dollars, which is usually what moves a negotiation.

At settlement the same grid produces the allocation. A common design assigns each member points equal to her pay-periods inside the window weighted by pay, so that the fund is divided in proportion to each member’s modeled loss; variants add weight for tenure or for members who opted in to the collective. Whether the settlement is a common fund or claims-made is counsel’s decision; the model supplies the schedule either way, and a claims-made structure needs an estimate of the claims rate to size the fund.

The defense version

Defense counsel need the same model for a different purpose: exposure. The scenario table shows the range from the most favorable set of rulings to the least, and identifies which rulings matter most. Two legal defenses have direct arithmetic consequences. Under § 260 of the FLSA, a court may reduce or deny liquidated damages if the employer proves good faith and reasonable grounds for believing it complied, which halves the Equal Pay Act figure. Under California’s § 1197.5, the employer must show that its factors, reasonably applied, account for the entire differential; a model that shows what share of the raw gap each factor explains is the evidence on that question, and it is the same decomposition the plaintiffs’ expert produces, read from the other side. An early exposure model also tells the employer what a prospective pay adjustment would cost, which is a business decision the litigation model informs but does not make.

A worked illustration

The figures are invented and describe no matter; they also hold pay constant and ignore interest, to keep the arithmetic visible. A certified class of 400 women earns an average of $78,000. The regression’s adjusted disparity is 3.2 percent, so the class-wide back pay is about $998,000 for each year of the period. The table shows what the statutes do with that single input.

The same 3.2 percent produces figures from $2.4 million to $6.0 million depending on the statute, the willfulness finding and the opt-in rate, before interest and before any good-faith reduction. In a real model each row is the sum of 400 individual computations with different hire dates, pay histories, components and windows, and the spread between scenarios is usually wider than this. That spread, not the point estimate, is what the parties negotiate over, and it is why the model is built as a grid that can be re-run under each ruling rather than as a single calculation.

Where to start

For a damages scoping, send the certification order or the proposed class definition, the complaint and the payroll data dictionary, and we will tell you what the production can support and what the first scenario table will look like. If you have received a class damages model from the other side, send it for review; the first step is to rerun it. For a new matter, start with a conflict check using party names only. The certification questions the model has to answer are summarized in our note on class certification, and the common-proof case study shows the same question decided in a wage-and-hour class.

Frequently asked questions

Are Equal Pay Act liquidated damages automatic?

The statute provides for an equal additional amount as liquidated damages, and courts award it unless the employer proves that it acted in good faith and had reasonable grounds to believe it was complying, in which case the court may reduce or deny the award under 29 U.S.C. § 260. Whether the employer has made that showing is a legal question; the model carries liquidated damages as a separate column so the figure can be included or removed.

Is Title VII back pay limited to two years?

Back pay cannot accrue from more than two years before the EEOC charge was filed, and it continues from the charge forward until judgment or reinstatement. Under the Lilly Ledbetter Fair Pay Act each discriminatory paycheck is a new violation for charge-filing purposes, so a charge filed years into a disparity is timely, and back pay reaches two years before it where the earlier practices were similar or related.

Can class damages be computed with a single formula?

A class-wide percentage applied to each member’s actual pay is a formula, and it is often the right starting point because it follows directly from the common proof. The model must still fit the liability theory under Comcast, must be readable member by member so that individual defenses can be heard under Dukes, and must be adjustable in the second stage of a Teamsters proceeding. A formula that cannot be opened up to the individual level is the kind the courts have rejected.

How is interest handled?

Interest is computed on each period’s shortfall from the date the pay was owed, at the rate the governing law and the court apply, and is carried as a separate column. Under the federal Equal Pay Act, courts generally do not award both prejudgment interest and liquidated damages on the same wages; under California’s § 1197.5 the statute provides for both.

What about members who received raises or promotions later?

Their damages are computed on their actual pay in each period, so a later raise or promotion changes the base to which the disparity applies but does not erase the shortfall in earlier periods. If the claim is that the disparity originated in starting salary and compounded through percentage raises, the model constructs a corrected salary path and applies the employer’s own raise percentages to it.