Partner Pay Equity Analysis

Firm: 142-partner US multi-office firm; modified-lockstep + origination overlay. Three comp years (2023, 2024, 2025). Each row: partner_id, comp_year, legal_sex, gender_identity, race, cohort_year, tier, equity_points, practice_group, office, leadership_role, originations, billed_hours, total_comp, plus the four comp components (scale, discretionary, origination, capital return).

Audit scope: gender pay gap on total_comp; intersectional gender-and-race where cell sizes allow; specification curve across {with/without origination, with/without tenure, with/without practice-group fixed effects}; Oaxaca-Blinder on the 2025 cross-section; panel fixed-effects on the three-year window.

Privilege: under-counsel direction; output is privileged work product, not a public report.

Headline finding: On the 2025 cross-section, women partners earn meaningfully less in total compensation than men. Controlling for cohort, tier, equity points, practice group, office, origination credit, and hours worked, the gap narrows substantially but does not close — a residual difference remains that the standard controls cannot explain.

Robustness: The specification curve runs the model across hundreds of defensible variant combinations (with and without origination credit, tenure, practice-group fixed effects, and so on). The direction of the gap is stable: a clear majority of variants return the same sign, and the central estimate sits in a narrow band. The finding is not an artifact of one modeling choice.

Decomposition (Oaxaca-Blinder): A majority of the raw gap is "explained" by differences in observable composition between the men and women partner groups — most prominently differences in origination credit and cohort year. A meaningful minority of the gap is "unexplained" by those observables. The skill flags this as structural, not causal. Unexplained does not mean discriminatory; it means the standard pay-determinants the model can see do not account for the residual difference.

Distributional shape (quantile regression): The gap is not uniform across the partnership. It is largest at the top of the compensation distribution — concentrated in the equity-partner book — and smaller in the middle and lower bands. This pattern is consistent with the unexplained residual sitting in discretionary or origination-driven comp components rather than in scale-component comp.

Multi-year panel: Across the three-year window, the within-partner fixed-effects estimate of the gender coefficient is consistent with the cross-section. The result is not driven by a single year's hiring or attrition cohort.

Intersectional view: Cell sizes for Black women and Latina women partners are too small for stable point estimates in this firm's 2025 data (flagged INSUFFICIENT SAMPLE in two of the three years). The skill does not report a number where the sample cannot support one. The committee should treat the intersectional question as open, not as zero.

Reading note: The output is structured statistical evidence, not a verdict of discrimination. The decomposition's "unexplained" share is the right starting point for a privileged committee deliberation — it is not, on its own, a finding of fault. The numerical detail behind every claim above lives in the Excel workbook with formulas exposed; the Word memo carries the prose version with the same caveats.

Outputs require professional review.