Reviewed: 9 October 2026
Pay gap analysis compares earnings across defined groups. Before calculating a percentage, decide whose earnings you are comparing, which pay components and period you use, and whether your statistic is a mean or a median. Different choices can produce different answers from the same records.
This guide walks through an original eight-worker fictional dataset. It contains no Indian market observations, actual employee information or current national wage-gap estimate. Its purpose is to show the calculation, a composition effect and the questions that a result leaves unanswered. It does not determine discrimination or legal compliance.
Define the statistic before collecting figures
The OECD gender wage gap indicator compares men's and women's median earnings relative to men's median. Its estimates generally concern unadjusted gross earnings of full-time wage and salary workers.
The ILOSTAT wages and working-time methodology describes an unadjusted gap based on average hourly earnings. Mean hourly and median earnings calculations should therefore carry different labels. Neither definition means that a single aggregate figure explains the cause of a difference.
For a workplace analysis, record the population and measurement choices explicitly. Are you comparing employees or also contractors and self-employed income? Does the measure include bonuses? Are amounts gross or net? Is it an hourly rate, monthly earnings or annual compensation? Which date or period does it cover? Mixing those quantities produces a number without a clear interpretation.
The exercise below uses equally weighted fictional workers, each with one supplied gross hourly rate in rupees for the same invented reference period. It does not convert annual packages into hourly pay or model taxes, benefits, hours worked or market salaries. The two category labels, L and H, merely organise the invented rates; they are not real occupations or assessments of job value.
Keep preparing
Continue with Sarkari Resume Templates₹299 — coaching के एक महीने से काफ़ी सस्ता / far cheaper than a month of coachingThe complete fictional dataset
| Worker | Group | Category | Gross hourly rate |
|---|---|---|---|
| M1 | Men | L | ₹100 |
| M2 | Men | H | ₹200 |
| M3 | Men | H | ₹200 |
| M4 | Men | H | ₹200 |
| W1 | Women | L | ₹100 |
| W2 | Women | L | ₹100 |
| W3 | Women | L | ₹100 |
| W4 | Women | H | ₹200 |
There are four records in each group and no missing rates. Every record has equal weight. The fixture uses two group labels to demonstrate a conventional comparison; it does not imply that every person's gender fits those labels or prescribe how an organisation should collect identity data.
The quantities are deliberately small and simple so readers can reproduce each step. They are not recommended wages, employee profiles or evidence about a particular employer.
Calculate the unadjusted mean gap
Add the four men's rates: 100 + 200 + 200 + 200 = 700. Divide by four: the men's mean is ₹175 per hour.
Add the four women's rates: 100 + 100 + 100 + 200 = 500. Divide by four: the women's mean is ₹125 per hour.
Using men's mean as the reference denominator:
Mean gap = (175 − 125) ÷ 175 × 100 = 28.57%, rounded to two decimals.
The absolute difference is ₹50 per hour. The percentage expresses that difference relative to the ₹175 reference mean. It does not say that every woman earns 28.57% less than every man, nor that a particular individual's rate differs from another's by that percentage.
Changing the denominator changes the statistic. The women's mean is 71.43% of the men's mean; the men's mean is 40% higher than the women's mean. These descriptions use different reference amounts and must not be interchanged. State the formula alongside the reported number.
Calculate the median gap separately
Sort each group's rates. With four observations, the median is the average of the two middle values.
- Men: 100, 200, 200, 200. The middle values are 200 and 200, giving a median of ₹200.
- Women: 100, 100, 100, 200. The middle values are 100 and 100, giving a median of ₹100.
Using men's median as the reference:
Median gap = (200 − 100) ÷ 200 × 100 = 50%.
The mean gap and median gap are both correctly calculated for this fixture. They answer different summary questions. Neither replaces the other merely because one looks more favourable. This median exercise is not a reproduction of the OECD's population dataset, and the mean exercise is not an ILO estimate for India.
Compare the categories without losing the overall result
Within category L, the supplied men's and women's rates are all ₹100. The group means are therefore equal and the category mean gap is zero. Category H has ₹200 in both groups and also has a zero mean gap.
| Category | Men's records | Women's records | Mean gap |
|---|---|---|---|
| L | 1 | 3 | 0% |
| H | 3 | 1 | 0% |
The aggregate mean gap remains 28.57% because the category mix differs. Three of the four men's records are in H; three of the four women's records are in L. The overall means use those observed counts, not equal category weights.
This explains the arithmetic of the fictional dataset. It does not explain why real people might be assigned to different roles or rates. Equal category averages do not establish fair access, equal work of equal value, an absence of bias or legal compliance. Broad categories could also hide differences within them.
An illustrative reweighting, with a precise label
Suppose we deliberately give L and H equal weight in each group. The illustrative comparison becomes:
- Men: 0.5 × 100 + 0.5 × 200 = ₹150.
- Women: 0.5 × 100 + 0.5 × 200 = ₹150.
- Gap between these reweighted means: 0%.
This calculation changes the weights; it does not change the underlying eight records. Call it an equal-category-weight illustration. It is not a validated adjusted regression, a causal estimate or proof that the aggregate disparity has been solved.
An actual adjusted analysis needs a documented model, justified variables, sufficient data and appropriate review. Choosing variables can change the question being answered. A resulting residual is not automatically discrimination, and a small residual is not automatically proof of fairness.
Check what happens when a record is missing
For a separate variation, suppose W4's rate is unavailable. Keep its missing status visible rather than replacing the unknown rate with zero.
If the clearly labelled available-record calculation excludes W4, the women's available-record mean is ₹100 from three records. The men's mean remains ₹175 from four records. The resulting available-record mean gap is:
(175 − 100) ÷ 175 × 100 = 42.86%.
That is not the complete-data result of 28.57%. The changed coverage must be reported. It does not reveal what W4 actually earned or whether the missingness was random. Including an invented zero would lower the women's mean further and misrepresent a missing observation as measured pay.
If the reference mean is zero, this percentage formula has a zero denominator and is undefined. Report that condition; do not force a value of zero, infinity or 100% into a summary table. Confirm unusual or invalid inputs against the measurement definition before calculating.
A reproducible calculation checklist
Before reporting a result, document these choices in a short methods note:
- Population: Who is included, excluded and counted once? State group sizes.
- Period and unit: Use comparable earnings periods and identify the currency and hourly, monthly or annual measure.
- Pay definition: Specify gross or net and the included components.
- Statistic and weights: State mean or median and whether records have equal or survey weights.
- Reference: Give the denominator and sign convention.
- Missing data: Explain exclusions and report their counts; do not silently use zeros.
- Grouping: Name categories and explain whether totals use the observed or changed composition.
- Interpretation: Separate what the calculation shows from questions it cannot answer.
For the original fixture, a complete concise result is:
In an eight-record fictional dataset with equally weighted gross hourly rates, the men's mean is ₹175 and the women's mean is ₹125. The unadjusted mean gap relative to men's mean is 28.57%. The median comparison gives 50%. Category-specific mean gaps are zero, while category counts differ. No cause, individual decision or legal finding is established.
That statement preserves both the result and its limits without replacing the analysis with generic negotiation advice.
Analysing other diversity dimensions
The same measurement discipline applies when comparing other defined groups, but do not infer a person's identity from a name, address or photograph. Define the question and use data through an appropriate authorised process. A two-group example does not establish how to combine multiple identities or classify people with missing information.
Intersectional analysis can involve smaller groups. A group mean based on very few records may be unstable and may expose individuals when combined with other details. Avoid publishing raw employee rows or treating this exercise's eight-record table as a suitable template for public disclosure of real pay data.
For actual organisational work, establish permitted access and reporting practices before using employee information. An aggregate statistic alone does not decide an individual's compensation, establish a legal entitlement or justify an employment action.
Questions to ask when reading a published pay-gap figure
Why do two reports show different percentages?
Check their population, period, pay components, hours measure, mean versus median, weights and denominator. The same label may conceal different definitions. Read the methodology before treating the figures as a trend or contradiction.
Does a positive gap prove discrimination?
It shows a difference under the stated measure. It does not by itself establish the causes, individual treatment or a legal conclusion. Further investigation may be warranted, but it requires evidence beyond this summary.
Does a zero gap prove equal opportunity?
No. Equal aggregate earnings can coexist with different access to jobs, promotion patterns or variation within groups. Our category exercise also shows why grouped and aggregate answers need to be read together.
Should I use the figure to choose a personal salary request?
An aggregate group gap is not a market quote for your role or a personal compensation recommendation. Keep individual offer comparisons separate and examine their actual terms. This article provides no salary band, negotiation uplift or promised outcome.
If you want to practise the data handling behind group counts and missing values, use the SQL interview exercises. Their dataset is a separate technical exercise. For recording project work accurately, the resume writing guide helps distinguish an analysis exercise from real employee-data experience.