People & Exec SupportFree
Pay Equity Analysis and Gap Report
An 8.6 point raw gap fell to 1.6 once level was controlled for. Level assignment was where most of the gap lived.
River builds the comparator groups before any control variable is applied: job code, level, location and scope, taken from the work rather than from job titles. Groups too small to compare get a count and nothing else. Then each explanatory factor is applied in a stated order and the residual reported after each one, so you can see exactly how much of the gap tenure absorbed and how much level absorbed. In the worked example level absorbed 4.7 of 8.6 points.
Then the part that changes the answer. A factor is only allowed to explain a difference if the factor itself is distributed evenly. Marloe Logistics had women at L3 and above at 31 percent against 44 percent for men at equal tenure in the same job families, so controlling for level converted a promotion gap into an explained gap. That is the specific reason most of these analyses end at no statistically significant difference and change nothing. The merit cycle version runs on money not yet committed.
Third, bonus, commission and equity are analysed apart from base, and eligibility is tested separately from amount. Base was within a point in three of the four groups where the bonus gap was 9.2 percent on the same plan and the same rating. Written for people leads and finance partners who need a number they can act on, so the output is fourteen named people, a per-person figure and a remediation cost at three targets.
Controlling for the wrong thing first is how a gap disappears
A pay equity analysis has two halves and almost all the effort goes into the second one. The first half decides who is compared with whom, and it is the half that determines the answer. EEOC guidance is explicit about the order. Identify the pool of comparators first, then consider the factors that might explain a differential, because qualifications and experience require their own analysis before they count as legitimate (EEOC Compliance Manual Section 10).
The same logic applies to every control you add afterwards. Tenure explains a difference only if tenure is not itself the product of the thing you are looking for. Level explains a difference only if people doing the same work sit at the same level. Under the Equal Pay Act a differential has to rest on a seniority system, a merit system, a system measuring quantity or quality of production, or a factor other than sex (29 CFR 1620.13). A control that fails its own test is not one of those.
Base salary is also only part of the answer. Non-base compensation can be discriminatory even when base is not, and it needs two separate tests: who is eligible for the plan, and how much eligible people actually receive. At Marloe the bonus plan was open to everyone at every level, so eligibility was clean, and the amounts inside it carried a 9.2 percent gap on identical plans and identical ratings. A base-only analysis returns nothing there.
How it works
Send payroll and roles
Base, bonus, equity, job code, level, location, tenure, ratings and the demographic fields you already hold.
Comparator groups get built
From the work, not the job title. Groups below the stated population threshold are reported as not compared.
Controls earn their place
Level, tenure and location are each tested for even distribution before being allowed to absorb any part of the gap.
Names, not percentages
The residual as a list of people with a figure each, and remediation priced at three defensible targets.
What you get
- Comparator groups built from job code, level, location and scope, with the groups too small to compare printed
- An ordered decomposition showing the residual after each control, so you see what every factor absorbed
- A distribution test on every control, because a factor that is itself uneven cannot explain anything
- Bonus, commission and equity analysed apart from base, with eligibility tested separately from amount
- The residual expressed as named people and money per person, never as one company-wide percentage
- Remediation costed at three targets, with the cheapest defensible one named and the reasoning shown
- A written findings memo naming the causes, so the same gap does not reappear next year
Common questions
Why not just run a regression?
You should, and it is the last step rather than the first. A regression with level on the right-hand side answers a question about pay within level, which is a fine question and not the one you asked. Run the distribution test on level first, and if level fails, report the level effect separately instead of letting the model absorb it.
Can I group people by job title?
No, and this is where most analyses go wrong before they start. Titles are the least consistent field in an HRIS: two people with the same title do different work, and two doing the same work carry different titles. Groups come from job code, level, location and scope, which is what a job description should have defined.
Is performance rating a valid control?
Only if the ratings are consistent between raters, which is a separate question with its own answer. Run the distribution test on ratings the same way, and if one manager's Exceeds is another's Meets, the rating is carrying the disparity rather than explaining it. That is what a calibration session is for. The reviews behind those ratings come from review writing.
What happens to groups too small to analyse?
They get a headcount, the words below threshold, and nothing else. No ratio, no direction, no significance test. Twenty of 34 groups fell below the threshold here, covering 82 people, and all twenty are named so next year nobody assumes they were cleared. A differential computed from four people invites a decision it cannot support.
How much does fixing it cost?
Priced at three targets so the decision stays yours. To the comparator median, 214,600 across fourteen people. To the bottom of each comparator range, 96,300, which is 0.4 percent of payroll. Flagged rows only, 61,800 across nine. The middle one is recommended, because the midpoint over-corrects and creates a new outlier.
Do bonus and equity really need their own analysis?
Yes, because non-base compensation can be discriminatory even when base is not, and it takes two tests. First, who is eligible for the plan. Second, among eligible people, how much they get. Marloe's bonus plan was open to everyone and still carried a 9.2 percent gap. Employer-paid benefits are a separate problem, handled by invoice reconciliation.
Who has to run one of these?
Federal contractors must evaluate their compensation systems for gender, race or ethnicity-based disparities (41 CFR 60-2.17). Everyone else runs it because the alternative is an employee running their own version against the pay scale you now have to publish in job postings (California Labor Code 432.3).
Pay Equity Analysis and Gap Report
Fill in the form and your workspace opens with the work already underway.