Wage Discrimination & Pay Equity Expert Witness

Our economists and statisticians are retained by plaintiff and defense counsel in Equal Pay Act, California Equal Pay Act, Title VII, Section 1981 and FEHA matters to answer a defined question: after accounting for the factors that legitimately explain pay, hiring, promotion or termination, does the outcome still differ by sex, race, ethnicity, age or another protected characteristic, and by more than chance would produce? We work from the employer’s own HRIS and payroll records, state every modeling choice that moves the result, and report what the data support and what they do not. A statistical disparity is evidence to be weighed with the rest of the record. It is not, by itself, a finding of discrimination, and we do not present it as one.

What our work includes

The assignment differs by claim and by stage, so the scope is set in writing before the engagement letter. It may include:

• Compensation analysis. Employee-level compensation analysis of base pay, total cash compensation and each pay component; descriptive statistics by job, level, location, tenure band and demographic group; multiple regression analysis of the wage gap with job, level and position controls, tenure and experience controls, geographic controls, and performance controls where appropriate; cohort and comparator analysis for Equal Pay Act and California Equal Pay Act claims; and gender, race and ethnicity disparity testing where it is relevant to the allegations.
• Selection analysis. Hiring analysis from applicant-flow or labor-market data, promotion analysis against the pool of eligible employees, job assignment and placement analysis, and termination and reduction-in-force analysis against the pool from which each decision was drawn.
• Inference and robustness. Statistical significance, reported in standard deviations and probability values; economic significance, reported in dollars, percentage points or positions; and sensitivity and robustness analysis across the alternative specifications the other side is likely to run.
• Damages. Individual back pay and front pay, and class-wide damages modeling that computes the shortfall for each class member and period from the same models used on liability questions.
• Review of the opposing expert’s report, replicated from its workpapers and tested one choice at a time.

How economists analyze pay discrimination

The sequence below is the one we follow on every engagement, whether the question is one plaintiff’s pay or a workforce of thousands. Each step produces something counsel can read before the next begins.

1. Data. HRIS and payroll extracts, applicant and promotion records and the compensation policies in force, requested in a data letter written to the claims pleaded so that nothing has to be re-requested after the model is specified.
2. Data validation. Files from different systems are reconciled into one employee-by-period dataset. Duplicate records, gaps in the period, unmatched employee IDs and inconsistent job codes are counted and reported, not silently dropped. The validated file and its log are the foundation of everything that follows and are produced with the report.
3. Descriptive analysis. Pay, headcount and selection rates by job, level, location and group, before any modeling. Raw gaps and their composition tell counsel where the disparity sits and whether the data can carry the theory.
4. Model specification. The comparison group, the pay measure, the period and the controls are chosen from the record and the statute, and the reasons are written down. Where a control is itself a product of the challenged decisions, such as a performance rating, a starting salary or a job assignment, it is designated for testing with and without.
5. Regression and statistical testing. The regression or selection-rate test is run, and the disparity is reported with its standard error, standard deviations and probability value, at the level where decisions were made and in the aggregate. How the regression is specified, read and attacked is set out in regression analysis in pay discrimination class actions.
6. Robustness testing. The result is re-run under alternative comparison groups, levels of pooling, control sets, pay measures and periods. Counsel sees which choices move the result and by how much before anyone is deposed.
7. Damages. Where liability is established or assumed, the same models produce the shortfall for each class member and period, individual back pay and front pay, and the statutory components kept separate so counsel can assemble the award.
8. Expert opinion. A report that states the data, the choices, the results and their limits, written so that the other side can rerun it, and testimony that stays within what the analysis shows.

The claims, and what each asks of the analysis

• Equal Pay Act. Under 29 U.S.C. § 206(d), the comparison is between men and women performing equal work requiring equal skill, effort and responsibility under similar working conditions, and a differential is lawful only if it rests on seniority, merit, quantity or quality of production, or a factor other than sex. The analysis identifies the comparators, measures the gap and tests whether the employer’s stated factors account for it. In the Ninth Circuit, prior salary is not a factor other than sex (Rizo v. Yovino, 950 F.3d 1217 (9th Cir. 2020) (en banc)), which changes which controls belong in the model.
• California Equal Pay Act. Labor Code § 1197.5 covers sex, race and ethnicity, compares employees doing “substantially similar work” viewed as a composite of skill, effort and responsibility, and requires that the factors the employer relies on be applied reasonably and account for the entire wage differential. Prior salary cannot justify a disparity. The regression therefore has to show how much of the gap each factor explains, not only whether a gap exists.
• Title VII, Section 1981 and FEHA. Pattern-or-practice claims ask whether disparate treatment was the employer’s regular practice (International Brotherhood of Teamsters v. United States, 431 U.S. 324 (1977)); disparate impact claims ask whether a specific neutral practice falls more heavily on a protected group (Griggs v. Duke Power Co., 401 U.S. 424 (1971); 42 U.S.C. § 2000e-2(k)). Title VII sex-based pay claims are not limited to equal work (County of Washington v. Gunther, 452 U.S. 161 (1981)), so the comparison group is broader and has to be defined with care.
• Age discrimination and reductions in force. Selection-rate and termination analyses under the ADEA and FEHA, including the pools from which a layoff was drawn and the criteria said to have governed it.
• Individual matters. The same tools answer whether one plaintiff was paid or promoted differently from comparable colleagues, and what the loss was. For the statutory remedies in a single-plaintiff pay claim, see Equal Pay Act damages.

Hiring, promotion and termination analysis

Selection decisions leave a different record from pay and call for different tests. For each decision we identify the pool of people who could have been selected, the people who were, and the rate at which each group was chosen. The tests, and the applicant-flow and availability pools they are run on, are described in analyzing promotion and hiring disparities.

• Hiring. Applicant-flow analysis compares selection rates across groups, with the four-fifths guideline in 29 C.F.R. § 1607.4(D) as a screen and tests of statistical significance as the evidence. Where applicant data are incomplete, we say so and use the labor-market availability analysis the record supports.
• Promotion. Who was eligible, who was selected and how long it took, measured against the pool of employees who met the posted requirements, tested at the level where decisions were made and combined across units with stratified methods so that a pattern spread across many small units is neither manufactured nor missed (Paige v. California, 291 F.3d 1141 (9th Cir. 2002)).
• Job assignment and placement. Who is placed in which job, shift, department or location at hire and afterward. Assignment often drives pay and promotion, so it is tested first when the theory is channeling.
• Termination and reductions in force. Involuntary separations compared across groups by stated reason, time employed and the selection criteria the employer says it applied. Discipline records also feed pay and termination decisions, which matters when they are proposed as a control.

When the group is small, we use exact tests suited to small samples and tell counsel when the statistical evidence will not carry weight on its own.

Class and collective actions

Pay equity and discrimination claims are brought as Rule 23 classes, as Equal Pay Act collective actions under 29 U.S.C. § 216(b), and in California as representative actions. In each form the question for the economist is whether the disparity can be shown with evidence common to the group and whether damages can be computed by a method that applies group-wide. In Wal-Mart Stores, Inc. v. Dukes, 564 U.S. 338 (2011), the Supreme Court held that a regional pay disparity could not by itself establish a uniform store-by-store pattern where store managers held the discretion, which is why we test the data at the level where decisions were made as well as in the aggregate. Under Comcast Corp. v. Behrend, 569 U.S. 27 (2013), the damages model offered at certification has to measure the damages that flow from the theory being certified, and under Tyson Foods, Inc. v. Bouaphakeo, 577 U.S. 442 (2016), representative evidence may be used class-wide when each member could have relied on it in an individual action.

What we deliver at the certification stage is a specified model rather than a description of one: the pay or selection equation, the controls, the level of aggregation, the data it runs on, and the shortfall it produces for each class member and period. Whether that showing satisfies commonality, predominance or the similarly-situated standard is a legal question for counsel and the court. We show the results at each level of aggregation so that argument is made on an informed basis. See also our note on class certification, our six questions for any statistical expert report, and our analysis of whether damages were computable on common proof.

Class-wide damages

Where liability is established or assumed, the same models support the damages calculation. For pay claims the shortfall is the difference between actual pay and the pay the model predicts absent the disparity, computed for each class member and pay period, with the benefits and retirement contributions that follow from pay. For promotion claims it is the difference between the earnings path of the position not received and the path actually followed. For terminations it is back pay and front pay, reduced by mitigation where the law requires it; see back pay vs. front pay.

Which remedies are available, for what period and subject to what limits differs by statute: the Equal Pay Act, enforced through the FLSA’s remedies (29 U.S.C. § 216(b)), provides the unpaid wages and an additional equal amount as liquidated damages; Labor Code § 1197.5(h) provides the balance of the wages, interest and an equal amount as liquidated damages; Title VII limits back pay to two years before the charge (42 U.S.C. § 2000e-5(g)(1)) and caps compensatory and punitive damages by employer size (42 U.S.C. § 1981a(b)(3)), while Section 1981 and FEHA carry no such caps; and since the Lilly Ledbetter Fair Pay Act each discriminatory paycheck restarts the Title VII charge-filing period (42 U.S.C. § 2000e-5(e)(3)(A)). Those are questions for counsel. We compute each scenario counsel identifies and keep the statutory components separate so the award can be allocated among class members on a stated basis. The class-level computation, period by period and statute by statute, is worked through in calculating class-wide damages in equal pay litigation.

What we request from counsel

For the conflict check, party names only. For scoping, a short call on the claims, the proposed class or comparator group and the deadline. Once engaged, the data letter typically asks for:

• HRIS or census data for every employee in the relevant jobs and period, with employee ID, sex, race and ethnicity as recorded, date of birth where age is at issue, hire and termination dates, and termination reason codes.
• Payroll data by pay period, with every component separately: base rate or salary, overtime, bonuses, commissions, equity grants and employer benefit contributions.
• Compensation history, including starting pay, each pay change and its stated reason, and pay-band or grade assignments over time.
• Job titles, job codes and levels, with job descriptions and any job-family or grade structure, and the history of each employee’s movement through them.
• Locations, departments and reporting lines, including the decision-maker for pay and promotion where the system records one.
• Promotion history and, where selection is at issue, the requisition, applicant and eligibility records for each decision.
• Performance information: rating scales, ratings by period and the policy that connects ratings to pay.
• Compensation policies: pay-setting and merit guidelines, bonus plans, seniority and merit systems, and any pay equity audit the employer has performed.
• Any expert report already served, with its data and workpapers, and the protective order, which we sign before receiving confidential material.

If a record you expect us to need does not exist, that changes the approach rather than ending it, and we would rather know at scoping. You receive a written scope and phase budget before any engagement letter.

Reviewing the opposing expert’s analysis

Most disputes between statistical experts are not about arithmetic. They are about which employees were compared, which pay components and periods were included, which controls were used, whether the data were pooled or split, and whether a significant coefficient was treated as a large one. We replicate the opposing analysis from its own workpapers, identify the choices that drive its conclusion, and show what happens when each one is changed. If you have received a report, send it to us for review. See also five problems to look for in an opposing damages report.

What the statistics can and cannot show

A statistically significant disparity shows that chance is an unlikely explanation for the difference. The controls in the model account for the lawful factors in the record. What remains is evidence for the trier of fact to weigh with the testimony and documents; the statistics do not decide whether it reflects discrimination, and a report that says otherwise has overreached. Significance also depends on sample size, so we report the practical size of the disparity alongside the standard deviations, and we tell counsel when a group is too small for the statistical evidence to carry weight on its own.

Who leads the work

Pay equity and discrimination analyses are led by Roman Garagulagian, Ph.D., principal economist (Ph.D. in Economics, Claremont Graduate University; Vice President–West and board member, National Association of Forensic Economics), and Britni Wilcher, Ph.D., an affiliated economist who quantifies compensation disparities in employment discrimination and class action matters, conducts pay equity analyses that identify compensation disparities by demographic group, and, before litigation consulting, led federal regulatory analyses at the Department of Labor’s OSHA that withstood judicial review.

Every engagement is led by the economist responsible for the opinion. Analysts may support the work, but the designated expert owns the methodology, assumptions, analysis, report and testimony. We do not designate one expert and hand the file to another. This practice sits alongside our broader employment litigation support and wage-and-hour and PAGA damages work, and draws on the same large-dataset and statistical sampling capability.

Getting started

Start with a conflict check using party names only; we answer the same business day. If the other side has already served a statistical report, send us the report for review. When the data are ready, upload the case files securely.

Frequently asked questions

Does a statistically significant pay disparity prove discrimination?

No. It shows that chance is an unlikely explanation for the gap that remains after the lawful factors in the model are accounted for. What explains that remainder is decided on the whole record. Our reports say so, because an expert who claims more than the statistics show loses the point on cross-examination.

How many standard deviations do courts look for?

Courts often refer to two or three standard deviations, following Hazelwood School District v. United States, 433 U.S. 299 (1977), and Castaneda v. Partida, 430 U.S. 482 (1977). That is a guide, not a line that decides the case. We report the exact figure, the size of the gap it measures, and how both change under alternative specifications.

Which controls belong in a pay regression?

The factors the employer says legitimately drive pay and that the record can measure: job, level, location, tenure, experience, education and, where appropriate, performance. A regression need not include every conceivable variable to be probative (Bazemore v. Friday, 478 U.S. 385 (1986)), but a control that is itself a product of the challenged decisions can hide the disparity, so we run the model with and without each contested control and report both.

Should the data be pooled or analyzed job by job?

It depends on where decisions were made and on the legal theory. Pooling can reveal a pattern too small to detect in any one unit; it can also create or hide a disparity when groups are spread unevenly across units. We run both, combine the unit-level results with a stratified test, and explain the difference.

Can you work from the employer’s own pay equity audit or pay data reports?

Both are useful for scoping and for testing the employer’s characterization of its workforce. A pay equity audit is usually run under privilege and built to find and fix gaps, not to be reproduced in litigation; California pay data reports under Government Code § 12999 summarize pay by job category and group. A litigation analysis runs on the underlying employee-level records, which we request in the first data letter.

How early should the economist be involved?

Before the data requests are final. The fields requested determine what can be tested later, and an early look at the produced data shows which claims the records can support before positions harden. Early scoping is also when an exposure range can be estimated from headcount, pay bands and the proposed class period.

What if the data are incomplete?

Gaps are counted and reported, and we state how each one is handled and what the result would be if it were handled differently. Where applicant data were never kept, hiring analysis turns to labor-market availability; where payroll history is short, the period is narrowed and the limitation is stated. Incomplete data change the approach; they do not end it.

Do you work for employers as well as plaintiffs?

Yes. We are retained by plaintiff and defense counsel, and the method does not change with the side of the caption. For defendants that often means rebuilding the plaintiff’s model exactly and testing only what the records do not support; for plaintiffs, a model built to survive that review.