Analytics Methodology

Pay equity analysis tests whether employees of different demographic groups — primarily by sex, and by race/ethnicity where EEO-1 data is provided — are paid differently after accounting for legitimate, job-related pay factors.

PaySymmetry™ uses two complementary lenses:

• Descriptive comparison: an employee’s pay is compared to the median of their peer group (employees of the same pay grade and job title in the reference gender group). This produces a concrete, employee-specific benchmark.
• OLS regression: a statistical model that isolates the pay difference associated with gender while simultaneously controlling for multiple legitimate pay determinants. PaySymmetry™ uses this for the org-wide adjusted gap.

The unadjusted gap is the raw difference between a group’s median pay and the reference group’s median pay, expressed as a percentage of the reference median, with no controls applied; the adjusted gap is what remains after the regression controls for job grade, job title, work location, tenure, and other mapped factors.

Such analyses support employer obligations and defenses under Title VII of the Civil Rights Act, the Equal Pay Act (EPA), and a growing body of state pay-transparency and pay-data-reporting laws — including California SB 1162, Colorado’s Equal Pay for Equal Work Act (EPEWA), and New York S9427A — covered in the Jurisdictional Coverage section.

Users upload a workforce file (CSV or Excel) and map their columns to PaySymmetry™ fields in the upload wizard.

Required fields

• Employee identifier
• Annual salary (or pay rate)
• Gender
• Pay grade
• Job title
• Work location

Optional fields

• Tenure
• Performance rating
• Race/Ethnicity (EEO-1 categories)

Column mapping

The wizard auto-suggests mappings via name hints and lets the user confirm or override each mapping. An unmapped optional column is simply ignored and the corresponding control is dropped from the model.

Validation rules

• Rows with non-positive salary (≤ 0) are dropped before analysis.
• Race/Ethnicity values are validated against the 7 standard EEOC EEO-1 categories; non-standard values generate a warning but are not dropped.
• Employees missing performance data are retained via a _missing_ sentinel rather than dropped, preserving sample size.
• Minimum peer-group size (default 3) is enforced downstream — see Peer Group Construction.

Data is processed per-organization in isolation, with row-level org separation enforced at the database layer.

Specification

PaySymmetry™ uses Ordinary Least Squares regression with the natural log of annual salary as the dependent variable:

log(salary) = β₀ + β₁·gender + β₂·pay_grade + β₃·job_title
              + β₄·location + β₅·tenure + β₆·performance + ε

pay_grade, job_title, and location enter as categorical (dummy-coded) controls and are always included. tenure and performance are included only when those columns are mapped.

Race/ethnicity is not a regression control. When an EEO-1 column is mapped, PaySymmetry™ reports pay by EEO-1 category descriptively — headcount, mean and median pay, and each category’s median pay gap against the reference-gender median — which is what the California SB 1162 report draws on. The adjusted gap is therefore not adjusted for race/ethnicity.

Why log-linear?

With a log-transformed outcome, the gender coefficient is interpreted proportionally: exp(β₁) − 1 is the approximate percentage pay difference associated with gender, holding the controls constant. This matches how pay gaps are reported (as a percentage) and stabilizes variance across the salary range.

What the gender coefficient means

β₁ (reported as exp(β₁) − 1) is the adjusted pay gap — the average pay difference associated with the non-reference gender group after accounting for all modeled legitimate pay determinants. A negative value means the non-reference group earns less on average.

Optional gender × pay-grade interaction

When enabled, the model appends a gender × pay_grade interaction term so the gap can vary by grade. The per-grade gap is back-calculated as exp(β_gender + β_interaction,grade) − 1. This reveals whether a disparity is concentrated at senior levels (a “glass-ceiling” pattern) versus uniform across grades.

Optional: Structured Job Evaluation as a substitute control

Organizations that already maintain a structured job-evaluation score for their roles — produced by an external, point-factor methodology such as Mercer IPE, WTW, or Hay — can optionally build a Job Catalog mapping each job title / pay grade pair in their workforce data to that score, then substitute it for the separate pay_grade and job_title controls above in a given analysis run. PaySymmetry™ does not compute, assign, or certify job-evaluation scores itself — it stores and applies a score the organization already holds from its own methodology.

The substitution only takes effect when every employee’s job title / pay grade pair has a matching Job Catalog entry; any gap in coverage falls back to the standard pay_grade and job_title controls for that run rather than blending the two approaches. Because it changes the model specification, an adjusted gap computed this way is not directly comparable to one computed with the standard controls on the same data — the results page discloses which specification was used for a given analysis.

Note: Structured Job Evaluation is available on the Professional and Complete plans. It is a regression control option, not a methodology for scoring jobs — organizations remain responsible for the scores they supply, and this option does not, by itself, satisfy the job-evaluation requirements of any specific pay equity statute. Consult qualified counsel for your jurisdiction’s requirements.

Statistical significance

Assessed at the conventional two-tailed p < 0.05 threshold.

R² interpretation

R² is the share of salary variation explained by the model’s controls. A higher R² means the model accounts for more of the legitimate pay structure. R² is a measure of model fit, not of fairness — a high R² does not by itself imply the absence of a pay gap.

A peer group is the set of reference-gender employees who share the same pay grade and the same job title (the grade × job-title cell). The peer-group median salary is the benchmark used for individual flagging and remediation costing.

Minimum peer-group size

Default is 3 reference-gender employees; configurable from 1 to 5 in the upload wizard. If a peer group contains fewer than the minimum number of reference-gender employees, its median is treated as undefined (set to NaN) because a median from 1–2 people is not a reliable benchmark.

What happens to a too-small group

Affected employees are excluded from the remediation estimate and counted in excluded_no_peer_group; the results surface a note explaining the exclusion. Setting the minimum to 1 reverts to no exclusions.

Note: performance ratings are not used to define peer groups (grade + job title only), even when performance is included as a regression control.

PaySymmetry™ uses a dual-lens approach to identify potential pay disparities:

1. Org-wide signal (regression): the OLS adjusted gender coefficient establishes whether a statistically significant disparity exists across the workforce after controlling for legitimate pay factors.
2. Individual signal (peer median): an individual employee is flagged when their salary falls below their peer-group median (same grade × job title, reference-gender employees, subject to the minimum peer-group size).

The analysis applies a configurable gap threshold (gap_threshold_pct, default −5.0%): employees whose shortfall relative to the peer median exceeds this threshold are surfaced as flagged.

Employees whose peer group is too small to produce a reliable median are not flagged for remediation — see Peer Group Construction.

Using both lenses together means a flag reflects both a workforce-level statistical pattern and a concrete, employee-specific benchmark gap, making each flag explainable in plain terms.

Adjustment cost per employee

The cost to raise a flagged employee to their peer-group median:

adjustment_needed = max(0, peer_median − salary)

Employees already at or above their peer median require no adjustment.

Total remediation cost

The sum of adjustment_needed across all flagged employees with a valid peer median. Employees excluded for small peer groups are not included in this total.

Budget optimizer (greedy, cheapest-gap-first)

When a fixed budget is supplied, the optimizer sorts flagged employees by adjustment_needed ascending and funds each to their peer median until the budget is exhausted — maximizing the number of employees resolved per dollar. It runs without re-running the regression and persists no state.

What-if simulation

Users can propose new salaries per employee in the What-If Sandbox; the platform recomputes the resulting gap impact without re-running OLS. Simulation results are not persisted.

Remediation figures are planning estimates, not legally binding obligations. Adjustments must be implemented outside the platform.

PaySymmetry™ produces pay-transparency and pay-data reports for the US jurisdictions below; all reports are generated from the same underlying analysis.

All three reports are produced from a single analysis run via the Compliance Hub and are rendered to PDF.

• OLS assumptions: linearity (in the log of salary), independence of observations, and that the included controls capture the legitimate pay structure.
• Omitted-variable bias: the model can only control for factors present in the uploaded data. It does not control for variables that were not provided — e.g. specific skills, certifications, market premiums, negotiated offers, or team-level effects — and a gap may shrink or grow once such factors are added.
• Independence caveat: standard OLS assumes employee observations are independent; nested structures (employees within managers or teams) can violate this. Hierarchical modeling is on the roadmap but is not part of the current standard model.
• Intersectional analysis: when the Race/Ethnicity (EEO-1) column is mapped, PaySymmetry™ additionally reports an org-level gender × race/ethnicity cohort table — headcount, median pay for cohorts of 10 or more employees, and a regression-adjusted gap for each cohort. Cohorts under 5 employees are suppressed entirely, and an additional cohort is suppressed alongside a small one where needed to reduce the risk that its size can be inferred by subtraction from other published figures — both to prevent identification of individuals. This is a secondary breakdown; it does not change the headline adjusted gap.
• Small groups: estimates for thinly populated grades or peer cells are inherently less stable. Analyses with fewer than 30 employees may produce unreliable estimates.
• Not a legal determination: the model does not determine whether any pay difference is legally discriminatory — that is a legal determination requiring qualified counsel.

A qualified analyst and legal counsel should review results before they are used for any legal filing or litigation defense.

Analysis storage

Analyses, their configuration, and results are stored per-organization with row-level isolation enforced at the database layer.

Audit log

User and system actions are recorded — including the event type, user, resource, timestamp, and IP address.

SOC 2 evidence export

A downloadable ZIP package contains:

• audit_log.csv — all audit log rows
• active_users.csv — current users with role and last login
• deletion_receipts.csv — analyses whose underlying files were purged, with reason and timestamp
• manifest.json — generated-at, org id, export id, row counts, and SHA-256 checksum

Retention & purge

Organizations may enable automated retention. A nightly retention job (runs daily at 02:00 UTC) purges stored files for organizations that have opted in; each purge produces a deletion receipt added to the audit trail.