Two Types of Pay Gap
Evenpay calculates two complementary pay gap metrics. Understanding the difference is key to reading your results.
Unadjusted (raw) pay gap — the simple difference in average pay between men and women. No explanation of why the gap exists. This is the number required for EU Pay Transparency Directive reporting.
Adjusted pay gap — the portion of the gap that remains after controlling for legitimate pay factors like job grade, experience, and education. This is the number that tells you what needs investigating.
Metric | What it tells you | Example |
Unadjusted gap | Overall average pay difference | "Women earn 12% less than men" |
Adjusted gap | Unexplained difference after controlling for factors | "After accounting for job grade, experience, and tenure, women earn 4% less" |
A raw gap of 12% might sound alarming — but if 8 percentage points are explained by differences in job grade, experience, and tenure, the unexplained gap is 4%. That 4% is what requires action.
Unadjusted Gap: Mean vs. Median
Evenpay uses both mean-based and median-based formulas depending on the context:
Comparison level | Formula used | Why |
Overall organization gap | Mean-based | Standard EU formula |
Tier-based gaps | Median-based | Less sensitive to extreme values |
Job level, org unit, job family, position | Mean-based | Standard comparison |
Historical / regulatory reporting | Both | Full picture |
A statistical significance test (t-test) is run for all gaps to confirm whether the observed difference is real or could be due to chance.
Comparison Levels and the Pay Gap Matrix
Evenpay calculates unadjusted pay gaps at multiple levels. Together, these comparison levels form a pay gap matrix — letting you identify pay gaps across various dimensions of the organization rather than relying on a single number.
Tier-based gaps (primary) — Tiers represent groups of employees doing "work of equal value." How you define tiers is flexible — you can group by job levels, job grades, job grade combined with job family, or any other criteria that reflect your organization's view of role equivalence. For example, IC3, M2, and P2 might all map to the same tier if you consider them equal in value. Gaps are calculated within each tier, comparing men and women. This is the comparison level most relevant to the EU Pay Transparency Directive.
Job level gaps — More granular than tiers. Useful for diagnostics when you want to drill deeper within a specific tier.
Organization unit gaps — By department. Helps you see where in the organization gaps concentrate.
Job family gaps — By function (engineering, sales, finance, etc.). An alternative lens to organizational structure.
Position gaps — By specific position title. The most granular level.
Used together, these levels give you a comprehensive picture: the tier view shows your directive-relevant gaps, while the other levels help you pinpoint exactly where the issues are and what's driving them.
How the Adjusted Gap Is Calculated
The adjusted gap uses the Blinder-Oaxaca decomposition — an internationally recognized standard for pay equity analysis. Here's what happens under the hood:
Step 1: Salary transformation
Salaries are log-transformed before analysis. This is standard practice in labor economics because salary differences are typically proportional (percentage-based), not absolute. It also means regression coefficients can be read directly as approximate percentage effects.
Step 2: Regression model
Evenpay runs a multiple regression on log-transformed salaries using your enabled pay factors as independent variables:
log(salary) = β₀ + β₁(Job Grade) + β₂(Experience) + β₃(Tenure) + … + ε
Gender is deliberately not included in the model. It's only used afterwards to group employees and compare outcomes.
Step 3: Blinder-Oaxaca decomposition
The total pay gap is decomposed into two parts:
Explained portion — differences in characteristics (e.g., men are in higher job grades on average)
Unexplained portion — the remaining difference that can't be attributed to legitimate factors. This is the adjusted gap.
The decomposition trains three separate models: one on men only, one on women only, and a pooled reference model on all employees. The explained gap is calculated from differences in average factor values between groups, and the unexplained gap captures the rest.
The adjusted gap percentage is then calculated as:
Adjusted Gap (%) = (exp(unexplained log gap) − 1) × 100
Good to know: Factor contributions can be negative. For example, if women in your organization have more experience on average than men, experience actually counteracts the gap rather than explaining it.
Shapley Value Attribution: Why Each Factor Matters
Beyond the Blinder-Oaxaca decomposition, Evenpay uses Shapley values to show exactly how much each factor contributes to the difference between the raw and adjusted gap.
Think of it this way: if your raw gap is 12% and your adjusted gap is 4%, something explains that 8 percentage point difference. Shapley attribution tells you how much of that 8pp comes from job grade, how much from experience, how much from education, and so on.
The values always sum to the total adjustment, giving you a complete and fair breakdown.
Pay Factors
Organizations choose which factors are considered legitimate pay determinants through the Pay Philosophy settings. Each factor can be independently enabled or disabled.
The current system-defined factors include:
Factor | Type | What it captures |
Job Grade | Ordinal | Level in the job evaluation system — usually the strongest predictor |
Experience | Continuous | Years of professional experience |
Tenure | Continuous | Time with the organization |
Skills | Continuous | Skill count or aggregated skill score |
Skill Fit | Continuous | Match between employee skills and position requirements (%) |
Education | Ordinal | Highest education level achieved |
Certifications | Continuous | Number of professional certifications |
Certification Fit | Continuous | Match between certifications and position requirements (%) |
Work Location | Categorical | Employee's work arrangement |
Organization Unit | Categorical | Department or division (opt-in) |
Job Family | Categorical | Functional role grouping (opt-in) |
Performance | Continuous | Weighted average of recent performance reviews |
You can also add custom factors based on your organization's own fields if your compensation philosophy includes criteria beyond the defaults.
How factors are handled
Continuous and ordinal factors (like experience, tenure, and job grade) go straight into the regression model.
Categorical factors (like organization unit, work location, and job family) are encoded using dummy variables. Small categories are automatically merged to keep the model stable, and the reference category is always the most common value.
If a factor has too much missing data across your employee population, it's automatically disabled for that analysis run. You'll see which factors were disabled and why in your results.
Salary Configuration
You can control exactly what goes into the salary figure used for analysis:
Salary components — Choose which parts of compensation to include (base salary, bonuses, allowances, etc.)
Session overrides — Adjust component selection for a specific analysis without changing your org-wide defaults
Normalization — All salaries are converted to a common monthly basis. Annual salaries are divided by 12, hourly rates are converted using contracted hours
FTE adjustment — Salaries are adjusted for full-time equivalence before analysis
Employees with zero or invalid salary data are automatically excluded.
Evenpay supports two salary modes: Agreed salary (current contractual salary — the default for standard reporting) and Historical salary (for tracking how gaps evolve over time using stored snapshots).
Analysis Modes
Statistical Mode (default)
Pure Blinder-Oaxaca regression based on data and enabled pay factors only. No manual adjustments. This is what you use for standard reporting.
Justification-Aware Mode
This mode incorporates documented justifications — approved business reasons for individual pay differences recorded by managers or HR. The system adjusts salaries by justified amounts before running the regression.
This answers: after accounting for both statistical factors and documented reasons, what gap remains?
The mode tracks documentation coverage, addressed vs. unaddressed gaps, and compliance readiness.
Justification-adjusted gaps (tier level)
At the tier level, Evenpay also calculates a justification-adjusted gap: the raw gap minus documented justifications, floored at zero. This gives you an operational view that combines statistical analysis with your documented evidence.
Cost to Fix
For each comparison level, Evenpay estimates the annual cost to close the gap. It's calculated as the total amount needed to bring each underpaid woman's salary up to the men's median, annualized. This helps you quantify what action would actually cost.
Alert Thresholds
Evenpay flags gaps at configurable severity levels: low, medium, high, and critical. You can set the percentage thresholds for each level and define the minimum group sizes required for a comparison to be flagged.
Reading Your Model Quality Indicators
Every analysis includes quality metrics so you (and your advisors) can assess reliability:
Metric | What it tells you | What to look for |
R² (R-squared) | How much of salary variation the model explains | ≥ 0.70 is strong, ≥ 0.50 is acceptable, < 0.50 means important factors may be missing |
Adjusted R² | R² corrected for number of factors | Prevents inflated scores when many factors are enabled with few employees |
p-value | Whether the model's fit is statistically significant | < 0.05 means the results are unlikely due to chance |
F-statistic | Overall explanatory power | Higher = stronger. Combined with p-value for full picture |
Max VIF | Whether factors overlap too much (multicollinearity) | > 5 is moderate concern, > 10 means factors may be redundant |
Observations-per-factor ratio | Whether you have enough data for the number of factors | At least 10 employees per factor is recommended |
Each individual factor also reports its own coefficient, standard error, significance (p-value), confidence interval, and VIF — so you can see exactly which factors are pulling their weight.
Data Requirements
For the analysis to run, you need:
A minimum number of employees with valid salary data
At least one employee of each gender
For the full Blinder-Oaxaca decomposition, each gender group needs more employees than the number of enabled factors plus one
If you don't have enough data for the decomposition, the system will tell you why and continue with the prediction model alone.
Factors with too much missing data are automatically disabled for that run. You'll always see which factors were disabled and how many employees were excluded.
Relationship to FPI and EPTD
Evenpay's methodology is designed to meet and exceed the requirements of the FPI Universal Fair Pay Check, which requires an adjusted gender pay gap calculation based on multiple regression using the Oaxaca-Blinder method. FPI does not prescribe a specific formula, transformation, or variable set — Evenpay's approach is compatible with these requirements.
For the EU Pay Transparency Directive, Evenpay supports the required reporting on gender pay gaps and comparison of pay between workers doing the same work or work of equal value. The adjusted analysis is an additional method that supports investigation of underlying drivers.
Features like Shapley attribution, cost-to-fix estimation, model diagnostics, and justification-aware analysis are Evenpay's own analytical enhancements that go beyond what any standard requires.