Sample. Facts, counts and results are illustrative and do not describe a real matter or 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

Age-based disparate impact in a reduction in force — statistical assessment

To	[Counsel], [Firm]
From	Karim Souidi, M.Sc. (Econometrics), NorthLaw.ai
Matter	[Plaintiff] v. [Employer] — sample file NL-2026-0000 (ADEA / 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

In the March 2025 reduction in force, employees aged 40 and over were selected for layoff at 2.5 times the rate of younger employees (20.3% vs. 8.1%). The gap is 3.6 standard deviations from what a neutral process would produce, well beyond the two-to-three standard deviation threshold courts treat as significant. After adjusting for tenure and department, the odds of selection for the protected group remain about 2.4 times higher (95% CI 1.3 – 4.6).

The employer's likely rebuttal, that selection tracked performance ratings, weakens but does not eliminate the disparity: with ratings included, the adjusted odds ratio falls to 1.8 and its confidence interval crosses one. The dispute therefore turns on whether the ratings themselves are a legitimate, age-neutral criterion. That is a factual and legal question, not a statistical one, and it is where settlement leverage lies for both sides.

2. Question presented

Whether selection for the reduction in force was statistically associated with age; whether that association survives adjustment for the factors the employer says it used; and how robust the finding is to the employer's most likely explanation.

3. Data

The employer's RIF roster as produced in discovery: 412 employees in the affected business unit, 58 selected. Details in Appendix B.

Group	Employees	Selected	Selection rate
Under 40	210	17	8.1%
40 and over	202	41	20.3%
Total	412	58	14.1%

4. What the numbers show

Figure 1. Selection rates by age group. The dashed line is the overall rate; a neutral process would produce roughly that rate in both groups.

Measure	Result	Interpretation
Selection rate ratio (40+ ÷ under 40)	2.5	Protected group selected 2.5× as often
Standard deviations from a neutral process	3.6	Exceeds the 2–3 SD range cited in Hazelwood and Castaneda
Adjusted odds ratio (tenure, department)	2.4 (1.3 – 4.6)	Disparity persists after legitimate factors
Adjusted odds ratio (+ performance rating)	1.8 (0.9 – 3.7)	Attenuated; interval includes 1

Full results, including the four-fifths screen, exact test and sensitivity analyses, are in Appendix A.

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

Strong: the raw disparity is large, statistically significant by the standard courts use, and survives adjustment for tenure and department. The sample is big enough that this is not a small-numbers artifact.
Exposed: the four-fifths rule, a screen many defendants cite, is passed (0.87) because retention rates are high in both groups. Counsel should be ready to explain why the standard-deviation test, not the four-fifths screen, is the appropriate measure for a layoff of this size; the EEOC guidelines themselves say so.
The pivot: performance ratings. Ratings are strongly associated with selection and moderately associated with age (employees 40+ average 0.4 points lower). If ratings are accepted as legitimate and age-neutral, the adjusted disparity is no longer statistically significant. If ratings were assigned after the RIF was planned, or by the same managers who made selections, they are part of the challenged process and should not be adjusted for.

6. Limits of this analysis

Statistical disparity is evidence of impact, not proof of intent or of liability; both depend on facts and law outside this memo.
The roster was supplied by counsel from discovery and not independently verified.
Whether performance ratings are a legitimate criterion is a legal and factual question; this memo shows what follows under each answer, not which answer is right.
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. Full measures of disparity.

Measure	Result	Basis
Expected selections, 40+ (neutral process)	28.4	Hypergeometric expectation; observed 41
Standard deviation of expectation	3.53	Hypergeometric variance
Standard deviations from expectation	3.6	(41 − 28.4) ÷ 3.53
Two-sided p-value (Fisher's exact)	< 0.001	Less than 1 in 1,000 under a neutral process
Four-fifths rule (retention rate ratio)	0.87	79.7% ÷ 91.9%; passes the 0.80 screen
Logistic regression: age 40+ (tenure, department)	OR 2.4 (1.3 – 4.6)	n = 412; robust SE
Logistic regression: + performance rating	OR 1.8 (0.9 – 3.7)	n = 408; four missing ratings excluded
Rating difference, 40+ vs under 40	−0.4 points	Adjusted for tenure and department

A2. Sensitivity. Using age 50 as the cut-off raises the selection-rate ratio to 3.1 and the standard-deviation measure to 4.2. Excluding the two departments closed in their entirety (all employees selected regardless of age) lowers the raw ratio to 2.2 and leaves the significance conclusion unchanged. Results are not sensitive to the handling of the four missing ratings (dropped vs. imputed at department mean).

A3. Method.

Selection rates computed from the produced roster; standard-deviation and exact tests under the hypothesis of age-neutral selection (hypergeometric model).
Logistic regression of selection on protected-group status with tenure and department, then additionally with performance rating; 95% confidence intervals from robust standard errors.
Sensitivity to age cut-off, department closures and missing data as reported in A2.
All code and intermediate tables are retained and available to counsel; nothing depends on undisclosed data.

Appendix B — Data

Source: employer's RIF roster produced in discovery (Bates [range]). Fields used: date of birth, hire date, department, job level, 2024 performance rating (1–5), selection indicator. Population: 412 salaried employees in the affected business unit as of the RIF date; 58 selected. Four records with missing ratings were retained for unadjusted analyses and excluded from rating-adjusted models. Age computed at the RIF date. No records were excluded for any other reason. Data were supplied by counsel and not independently verified.