Sample. Facts, counts and results are illustrative and do not describe a real employer. The structure, disclosures and limits are exactly what a delivered memo contains: a short memo for counsel, followed by technical and data appendices for the other side's expert.

Settlement Analytics Memorandum

Gender pay equity — regression analysis of base compensation

To	[Counsel], [Firm]
From	Karim Souidi, M.Sc. (Econometrics), NorthLaw.ai
Matter	[Plaintiffs] v. [Employer] — sample file NL-2026-0001 (Equal Pay Act / Title VII / state analogue)
Date	24 September 2026
Purpose	Settlement and mediation support. Privileged; prepared at the request of counsel. Not prepared for filing or testimony.

1. Bottom line

Among 1,240 salaried employees in the produced 2025 payroll, women earn 14.2% less than men on average. After adjusting for the factors the employer says determine pay (job level, tenure, performance rating, location and department), the gap is 3.8% (95% CI 1.6% – 6.0%), statistically significant and stable across every specification tested. The adjusted gap widens at higher pay levels, and about 42% of the raw gap is attributable to women being placed in lower job levels for comparable tenure and ratings, which is a separate question from unequal pay within level.

Annual base-pay exposure from the adjusted gap is roughly $1.6 million (interval $0.7M – $2.5M). If job-level placement is itself found to be affected by gender, the exposure rises to about $4.6 million per year. The employer's strongest argument is that job level is a legitimate factor; the plaintiffs' strongest argument is that it is not fully legitimate. Both are quantified below.

2. Question presented

Whether the employer's base compensation differs by gender after accounting for legitimate, non-discriminatory factors; how robust that finding is to the choice of factors and model; and what the difference implies in dollars for a settlement discussion.

3. Data

Payroll and HRIS extract produced in discovery: 1,240 salaried employees active on 31 December 2025 (540 women, 700 men), with base salary, job level, tenure, performance ratings, location, department and degree. Details in Appendix B.

4. Results

Specification	Female coefficient	95% CI	p-value
Raw gap, no controls	−14.2%	−17.9 to −10.5	<0.001
Employer's full model (tenure, location, department, rating, degree, job level)	−3.8%	−6.0 to −1.6	<0.001
Same, excluding job level (if level is a tainted variable)	−10.9%	−14.0 to −7.8	<0.001

Figure 1. The female coefficient with 95% intervals across six specifications (Appendix A, Table A1). No interval reaches zero.

Where the raw 14.2% gap comes from	Points	Interpretation
Job level differences	5.9 (42%)	Women are in lower levels than men with similar tenure and ratings
Tenure, location, department	2.7 (19%)	Legitimate if applied neutrally
Performance rating, degree	1.8 (13%)	Ratings average 0.2 points lower for women
Unexplained (within-level gap)	3.8 (27%)	Not accounted for by any produced factor

5. Where the case is strong and where it is exposed

Strong: the within-level gap of 3.8% is significant in every specification, is not driven by executives, and is largest at the top of the pay distribution (5.1% at the 75th percentile vs. 2.9% at the 25th), consistent with a glass-ceiling pattern rather than noise.
Exposed: 3.8% is a modest gap in absolute terms, and the employer will argue it reflects unmeasured factors (prior salary, negotiation, external hires). None appears in the produced data; if the employer relies on them, counsel should request the fields and we re-run the model.
The pivot: job level. Treating level as a legitimate control removes 42% of the raw gap. A separate model of level placement shows women are 31% less likely (OR 0.69, 95% CI 0.55 – 0.87) to be in a given level or higher than men with the same tenure, rating and degree. If level is tainted, it should not be controlled for, and the relevant gap is 10.9%.

6. Exposure

Scenario	Gap applied	Annual base-pay exposure	Interval
Within-level gap only (employer's framing)	3.8%	$1.6M	$0.7M – $2.5M
Within-level gap + placement effect (plaintiffs' framing)	10.9%	$4.6M	$3.3M – $5.9M
Midpoint, level placement partially legitimate	~7%	$2.9M	$2.0M – $3.9M

Exposure = gap × female base payroll ($42.1M). Excludes bonus, equity, benefits, interest and liquidated damages; look-back periods vary by statute and are for counsel to apply.

7. Limits of this analysis

A regression gap is evidence of a difference not explained by the produced factors; it is not proof of intent, and unmeasured legitimate factors could reduce it.
Whether job level is a legitimate control is a legal and factual question. This memo quantifies both answers; it does not choose between them.
Data were supplied by counsel from discovery and not independently verified.
Aggregation across departments assumes a common pay policy; if policies differ materially by unit, class-wide analysis should be revisited.
Prepared for settlement discussions; not an expert report under FRCP 26(a)(2)(B).

Karim SouidiM.Sc. Econometrics · DASCA Senior Data Scientist · NorthLaw.ai · karim@northlaw.ai

Appendix A — Technical results

Table A1. Regression of log base salary on gender and controls. Coefficients are approximately percentage differences; heteroskedasticity-robust standard errors in parentheses.

Specification	Female coef.	SE	95% CI	p	R²
(1) Raw gap, no controls	−14.2%	(1.9)	−17.9 to −10.5	<0.001	0.04
(2) + tenure, tenure², location, department	−11.6%	(1.7)	−14.9 to −8.3	<0.001	0.31
(3) + performance rating, degree	−10.9%	(1.6)	−14.0 to −7.8	<0.001	0.38
(4) + job level (employer's full model)	−3.8%	(1.1)	−6.0 to −1.6	<0.001	0.81
(5) As (4), level × department fixed effects	−3.5%	(1.2)	−5.9 to −1.1	0.004	0.84
(6) As (4), excluding levels 7–8 (executives)	−3.3%	(1.0)	−5.3 to −1.3	0.001	0.79
Quantile regression, specification (4)
25th percentile	−2.9%	(1.3)	−5.4 to −0.4	0.023	—
50th percentile	−3.7%	(1.2)	−6.1 to −1.3	0.002	—
75th percentile	−5.1%	(1.5)	−8.0 to −2.2	<0.001	—

Table A2. Ordered logistic model of job-level placement. Outcome: job level (1–8); controls: tenure, tenure², performance rating, degree, location, department. Female: OR 0.69 (95% CI 0.55 – 0.87), p = 0.002; proportional-odds assumption not rejected (Brant test p = 0.21).

A3. Decomposition. Oaxaca-Blinder decomposition of the raw log-pay gap using specification (4) coefficients, pooled reference; components reported in the memo body sum to 14.2 points within rounding.

A4. Sensitivity. Direction and significance of the female coefficient are unchanged when: performance rating is lagged one year; salary is modelled in levels rather than logs; the nine imputed ratings are dropped; each of the four sites is excluded in turn; standard errors are clustered by department. Using 2024 payroll instead of 2025 gives a within-level gap of 4.1% (95% CI 1.9 – 6.3).

A5. Method.

OLS on log base salary with heteroskedasticity-robust standard errors, adding controls in the order the employer's compensation policy describes them.
Fixed-effects and executive-exclusion variants to test whether the gap is an artefact of a few cells.
Quantile regression at the 25th, 50th and 75th percentiles.
Oaxaca-Blinder decomposition to attribute the raw gap to measured factors and an unexplained residual.
Ordered logistic model of job-level placement to assess whether level is a tainted variable.
Exposure from the adjusted gap and female payroll, with intervals from the coefficient's confidence interval. Code and intermediate tables retained and available to counsel.

Appendix B — Data

Source: payroll and HRIS extract produced in discovery (Bates [range]). Population: 1,240 salaried employees active on 31 December 2025 (540 women, 700 men); hourly and contract staff excluded as outside the pay policy at issue. Fields: base salary, gender, job level (1–8), hire date, 2024 and 2025 performance ratings (1–5), location (4 sites), department (9), highest degree. Nine records with missing 2025 ratings carried the 2024 rating; no other imputation. Tenure computed at 31 December 2025. Female base payroll: $42.1M. Data were supplied by counsel and not independently verified.