Guides  /  pay reporting

Gender pay gap analysis and reporting

An employer can pay every man and woman in the same job exactly the same and still report a gap approaching thirty per cent. That is not a loophole, it is what the measure is designed to reveal. This guide works the six statutory figures out from raw salaries, shows why the quartile table matters more than the headline, and sets out what the gap can and cannot tell you.

Saad Tariq Written and reviewed by Saad Tariq, Principal Statistician
Updated 4 November 202613 min read
What is the gender pay gap?

The difference between the average hourly pay of men and women across an entire workforce, expressed as a percentage of men's pay. It is not a measure of equal pay. Equal pay compares people doing the same or equivalent work and unequal pay has been unlawful since 1970. The gap compares everybody regardless of role, so it is driven mainly by how men and women are distributed across grades. UK employers with 250 or more employees must publish six specific figures every year.

Decomposing the gap, with the arithmetic

The published mean gap in this workforce is 28.9%. The useful question is how much of it is accounted for by measurable differences in what men and women in the organisation do, and how much is left over.

Fitting a log-pay model with grade, contracted hours and tenure, then attributing the gap to the difference in those characteristics:

ComponentPercentage pointsShare of the gap
Grade and seniority mix6.924%
Contracted hours2.48%
Tenure1.55%
Explained, total10.837%
Unexplained residual18.163%
Raw gap28.9100%

In money, the unexplained 18.1 percentage points is about £8,300 on the men's mean of £45,778.

What "unexplained" does and does not mean

Two misreadings are common and point in opposite directions.

✓Unexplained is not a measure of discrimination. It is what the variables in your model did not account for. Add a variable and it shrinks; the world has not changed.
✓Explained is not the same as justified. Grade mix accounts for 24% of this gap, which immediately raises why grade distribution differs. If promotion routes are the cause, the gap has been relocated rather than explained.

This is why controlling for grade is the most contested decision in any pay analysis. Controlling for it tells you whether men and women are paid differently within the same grade, which is an equal pay question. Not controlling for it tells you whether women are concentrated in lower grades, which is the pay gap question. Both are worth knowing, and reporting only the first while calling it the pay gap is the standard way of making a gap disappear.

Report both figures and say which model produced which. The method is the same multiple regression covered in the regression analysis guide, applied to log pay so coefficients read as approximate percentages.

The core distinction

Pay gap and equal pay are different things

Equal payGender pay gap
ComparesPeople doing the same or equivalent workAll men against all women, any role
Legal statusUnequal pay is unlawfulReporting is mandatory at 250+ staff; having a gap is not unlawful
Driven byPay decisions for individualsWho holds which jobs
Fixed byCorrecting individual payChanging progression, recruitment and retention

Conflating them produces two opposite errors. A company announces "we have no gender pay gap because we pay equally for equal work", which does not follow. Or a journalist reports a 28% gap as evidence that women are paid less for the same job, which also does not follow.

Both can be true, and both can be false

An employer can have a zero pay gap and a serious equal pay problem, if underpaid women happen to be spread evenly across grades. It can have a large gap and no equal pay problem at all. The two measures answer different questions and neither substitutes for the other.

The statutory set

The six figures you must publish

#FigureBased on
1Mean gender pay gap in hourly payOrdinary pay, snapshot date
2Median gender pay gap in hourly payOrdinary pay, snapshot date
3Mean bonus gapBonuses paid over the preceding 12 months
4Median bonus gapSame
5Proportion of men and women receiving a bonusSame
6Proportion of men and women in each pay quartileAll relevant employees, ranked by hourly pay

The snapshot date is 31 March for public bodies and 5 April for private and voluntary sector employers, with publication due within a year. Only "full-pay relevant employees" count for the hourly pay figures, which excludes anyone on reduced pay for leave at the snapshot date. That exclusion matters: leaving in people on statutory maternity pay would create an artificial gap that reflects leave policy rather than pay.

The formula throughout is the same.

gap = (men's figure − women's figure) / men's figure × 100

A positive number means men are paid more. A negative number is reported as such, not as zero.

Worked example one

Calculating the gap

An employer with 300 staff and three grades. Pay bands are identical for men and women in every grade. Nobody is paid differently for the same work.

GradePay bandWomenMen
Junior£24,000–£28,00011035
Mid£34,000–£42,0004540
Senior£55,000–£72,0001555
Total170130

Step 1. Mean pay for each group

women's mean = £32,528
men's mean = £45,778

Step 2. The mean gap

(45,778 − 32,528) / 45,778 × 100 = 28.9%

Step 3. The medians

women's median = £27,150
men's median = £40,550

Step 4. The median gap

(40,550 − 27,150) / 40,550 × 100 = 33.0%

What has just happened

A 28.9% mean gap and a 33.0% median gap, in an organisation where not one person is paid differently from a colleague doing the same job. Every pound of that gap comes from the fact that 42% of men are senior and 9% of women are.

The test that proves it

Hold every salary exactly where it is and rebalance the grades so men and women are distributed identically. The mean gap falls to roughly zero. Nobody's pay changed. This is why the gap is a measure of workforce structure, and why actions aimed at individual salaries rarely move it.

Worked example two

The quartile table

Figure 6 is the one most organisations treat as an afterthought and the one that actually explains the result.

Rank every relevant employee by hourly pay, split into four equal groups, and report the sex split of each.

QuartileWomenMen
Upper24.0%76.0%
Upper middle52.0%48.0%
Lower middle66.7%33.3%
Lower84.0%16.0%

The workforce is 57% female overall. The lower quartile is 84% female and the upper is 24%. That gradient is the gap, expressed as a cause rather than a symptom.

Read together, the two tell a complete story. The headline says there is a gap. The quartiles say it comes from progression, not from pay decisions. Those point to entirely different remedies, and only the second is actionable.

✓A steep gradient across quartiles means a progression or recruitment issue.
✓A flat gradient with a large headline gap suggests something within grades, which is an equal pay question and needs a separate audit.
✓An unusual single quartile often reflects one large occupational group, such as a predominantly female administrative function.

Reading the pair

Why mean and median disagree

The two figures are published together deliberately, and the gap between them carries information.

The mean is sensitive to extreme values. A handful of very highly paid people, who in most organisations are disproportionately male, pulls the men's mean up and widens the mean gap. The median ignores them entirely and describes the typical employee.

PatternUsually indicates
Mean gap much larger than medianA small number of very high earners, mostly men. An executive pay story.
Median larger than meanStructural difference through the middle of the organisation, not driven by the top.
Both similarA consistent difference across the whole distribution.

In the worked example the median gap of 33.0% exceeds the mean of 28.9%, which says this is not an executive pay problem. It runs through the middle of the organisation, where most people are.

Report both without choosing. Publishing only the more flattering figure is visible to anyone who knows the regulations require both.

The volatile figures

The bonus figures, and why they look extreme

Bonus gaps are routinely far larger than pay gaps, often above 50%, and three features of the calculation explain most of it.

✓Bonuses are not pro-rated for part-time work. Unlike hourly pay, the bonus figures use the actual amount received. A part-time employee on a pro-rated bonus appears to have received less, and part-time workers are predominantly women. This mechanically inflates the gap.
✓Bonus distributions are extremely skewed. A few large payments at the top dominate the mean. The median bonus gap is almost always the more informative of the two.
✓The proportions receiving a bonus matter separately. Figure 5 exists because a small gap among recipients can sit alongside a large difference in who gets one at all.

If your bonus gap is far larger than your pay gap, check part-time distribution before concluding anything. It is usually the main driver and it is an artefact of the methodology rather than a finding about bonus decisions.

The narrative

Explaining a gap without explaining it away

The regulations invite a supporting narrative. It is optional, it is the part everyone reads, and it is where credibility is won or lost.

A regression can decompose the gap into a part associated with measurable factors such as grade, tenure and hours, and a residual that those factors do not account for. The Oaxaca-Blinder decomposition is the standard method.

Two cautions that matter more than the technique.

✓"Explained" does not mean "justified". If grade explains the gap, the question becomes why grade distribution looks like that. Controlling for seniority can control away the very thing being asked about.
✓A small residual is not a clean bill of health. It means the factors you measured account for the difference. If progression itself is where the problem lies, the model will absorb it and report nothing.

The strongest narratives do three things: state the figures plainly without softening them, identify where in the structure the gap arises using the quartile data, and commit to actions that address progression rather than restating a commitment to equal pay, which is a legal obligation rather than an achievement.

Doing it

In R, SPSS and Stata

Reading the quartiles as representation ratios

The four percentages are easier to interpret against the overall female share of the workforce, which here is 42%.

QuartileWomenRepresentation ratio
Upper26%0.62
Upper middle38%0.90
Lower middle46%1.10
Lower58%1.38

representation ratio = quartile share / overall share = 0.26 / 0.42 = 0.62

Women are at 62% of their expected representation in the upper quartile and 138% in the lower. A ratio of 1.00 in every quartile would mean no vertical segregation and a mean gap close to zero. This single column is usually the most informative part of the published figures, and almost nobody computes it.

One arithmetic check worth running: the four quartile shares should average to roughly the overall share, since the quartiles are equal in size. Here (26 + 38 + 46 + 58) / 4 = 42%, which matches. If yours does not, the quartiles were built wrongly, usually by ranking within each sex separately instead of across the whole workforce.

R

The whole statutory calculation is a few lines on a tidy payroll extract.

library(dplyr)
pay <- pay %>% filter(full_pay_relevant)
s <- pay %>% group_by(sex) %>% summarise(mean = mean(hourly), median = median(hourly))
gap_mean <- (s$mean[s$sex=="M"] - s$mean[s$sex=="F"]) / s$mean[s$sex=="M"] * 100

For the quartiles, pay %>% mutate(q = ntile(hourly, 4)) %>% count(q, sex) %>% group_by(q) %>% mutate(pct = n/sum(n)). Note ntile splits as evenly as possible when the count does not divide by four, which matches the guidance. For decomposition, oaxaca::oaxaca().

SPSS

SELECT IF fullpayrelevant = 1.
MEANS TABLES = hourly BY sex /CELLS = MEAN MEDIAN COUNT.
RANK VARIABLES = hourly /NTILES(4) INTO payquartile.
CROSSTABS TABLES = payquartile BY sex /CELLS = COUNT ROW.

Apply the SELECT IF before anything else. Including employees on reduced pay at the snapshot date is the most common data error in this exercise and it inflates the gap in a way that is hard to spot afterwards.

Stata

keep if fullpayrelevant == 1
tabstat hourly, by(sex) stat(mean p50 n)
xtile payq = hourly, nq(4)
tab payq sex, row

For the decomposition, ssc install oaxaca then oaxaca lnhourly grade tenure parttime, by(sex) weight(1). Model log hourly pay rather than raw pay, since pay distributions are right-skewed and the log form gives coefficients that read as percentage differences.

Hourly pay is a defined quantity, not an obvious one

It is ordinary pay for the pay period divided by the employee's weekly working hours, converted to an hourly rate. It includes allowances and shift premia but excludes overtime, redundancy pay and benefits in kind. Getting this definition wrong at source makes every downstream figure wrong, and it is the part auditors check first.

Avoid these

Six mistakes that cost you credibility

MistakeWhy it matters
Claiming no gap because you pay equallyThey are different measures. An employer with identical pay bands can report a 28.9% gap, as the worked example shows.
Including staff on reduced payPeople on maternity or sick leave at the snapshot date are excluded from the hourly figures. Including them creates a gap that reflects leave policy.
Publishing only the flattering figureBoth mean and median are required. Choosing one is visible to anyone who knows the regulations.
Treating the bonus gap as a bonus decisionBonuses are not pro-rated in this calculation, so part-time working, which is predominantly female, inflates it mechanically.
Controlling for grade and declaring the gap explainedIf progression is the problem, adjusting for grade removes the finding rather than accounting for it.
Ignoring the quartile tableIt is the only one of the six figures that shows where the gap comes from, and so the only one that points at an action.
Do you need the adjusted figure as well as the published one?

The statutory figures are the start. Boards ask what explains the gap and what is left over, and controlling for grade answers a different question from not controlling for it. Both are worth having, provided it is clear which is which.

See statistical consultancy
Saad TariqSaad TariqPrincipal Statistician

Writing it up

How to report it

✓All six figures, unrounded beyond one decimal place, on your own website and the government portal.
✓The snapshot date and the number of relevant employees.
✓The quartile table given prominence, not buried after the narrative.
✓A plain statement of where the gap arises, in structural terms.
✓Last year's figures alongside, with any change described honestly rather than selectively.
✓Actions tied to progression and recruitment, with dates, not a restatement of equal pay commitments.

One sentence worth drafting carefully: "Our median gender pay gap is 33.0%. Pay bands are identical for all employees in the same grade; the gap arises because women make up 84% of our lower pay quartile and 24% of our upper quartile."

Common questions

Frequently asked questions

What is the difference between the gender pay gap and equal pay?

Equal pay compares men and women doing the same or equivalent work, and paying unequally has been unlawful since 1970. The pay gap compares the average pay of all men against all women regardless of role, so it is driven mainly by how the sexes are distributed across grades. An employer with perfect equal pay can report a large gap.

How do you calculate the gender pay gap?

Subtract women's average hourly pay from men's, divide by men's, and multiply by 100. Do it once with means and once with medians. In a worked example with identical pay bands but different seniority distributions, the mean gap was 28.9% and the median 33.0%.

Who has to report?

UK employers with 250 or more employees. The snapshot date is 31 March for public bodies and 5 April for private and voluntary sector employers, and publication is due within twelve months on both your own website and the government portal.

Why is our bonus gap so much larger than our pay gap?

Usually because bonuses are not pro-rated in this calculation. Part-time employees receive a pro-rated bonus and the figure uses the actual amount, so part-time working, which is predominantly female, inflates the gap mechanically. Bonus distributions are also heavily skewed, which is why the median bonus gap is more informative than the mean.

Should I exclude people on maternity leave?

Anyone on reduced pay at the snapshot date is excluded from the hourly pay figures, which includes most people on statutory maternity or sick pay. They still count for the bonus figures. Including them in the hourly calculation creates a gap that reflects leave arrangements rather than pay.

Can we adjust the gap for grade and seniority?

You can decompose it, and Oaxaca-Blinder is the standard method, but be careful what you conclude. If women are under-represented in senior roles, controlling for grade removes the gap and also removes the question. Explained is not the same as justified.

Is a negative gap a problem?

No, and it should be reported as a negative figure rather than as zero. It means women's average pay exceeds men's, usually because of occupational distribution in the other direction. The regulations require the actual number whichever way it runs.