Pay, gaps and fairness

The Equality Act 2010 (Gender Pay Gap Information) Regulations 2017, SI 2017/172, require every employer with 250 or more employees to publish six figures each year, on a snapshot date of 5 April for private employers, within twelve months. All six are about hourly pay or bonus, and all six are averages across all men and all women, whatever their jobs.

1 and 2: the mean and the median hourly pay gap, as a percent of men's pay. The median is the one to read first: it is the middle woman against the middle man, and it is not pulled by a handful of very high earners the way the mean is. 3 and 4: the mean and the median bonus gap, on the same logic, over the year before the snapshot; 75.8 percent of 2024 filings report a median bonus gap, the rest pay no bonus to one sex or either. 5: the share of each sex who received a bonus. 6: the share of each sex in four equal-sized hourly pay quartiles, which is the pyramid in the figure, and the one figure that shows where a gap comes from rather than how big it is.

Every one of the six is a statistic about averages across roles. None of them compares a woman and a man in the same job. That comparison is equal pay, it is a legal duty, and a filing cannot show whether it is met. What the filing can show is how the workforce is shaped, and that turns out to be most of the story.

For reporting year 2024, 11,315 employers filed; 11,312 of the filings are complete and in range, and 11,301 have at least one man and one woman across the four bands, which is what it takes for a gap to be decomposable. Across them, the median hourly gap has a median of 8.4 percent (2023: 8.9). 77.6 percent of employers report a positive median gap, men paid more; 8.5 percent report exactly zero. The mean gap's median is 11.0.

The pyramid at the median employer runs 56.0 percent women in the bottom band, 52.7 in the lower middle, 47.0 in the upper middle, 40.0 in the top, and 81.3 percent of employers have a higher share of women at the bottom than at the top. That single fact is the reason the median gap is positive at most employers: it does not require any woman to be paid less than a man in the same band. It requires the bands to be paid differently, which is what bands are, and the sexes to be unevenly spread across them, which they are at four filings in five.

Take the median employer's pyramid, 56.0 / 52.7 / 47.0 / 40.0, and suppose every band pays the same to everyone in it. The women's average is still lower than the men's, because more of the women are in the lower-paid bands. How much lower depends on how steeply pay rises up the bands. Nobody publishes band pay, so the deep-dive track fits a ladder across all usable employers from the gaps they reported: 1 / 1.20 / 1.45 / 2.80, meaning the top band pays about 2.8 times the bottom at a typical filer. Under that ladder, composition alone explains about a third of the variation in reported mean gaps across employers (R2 0.353), and the composition-only gap has a median of 9.8 percent against a reported mean-gap median of 11.0.

The plain reading: at a typical UK filer, most of the reported gap is the pyramid. That is not a defence and it is not an accusation. It says where the gap is: in who holds which band, not in what the band pays. It also says where a plan has to act if it wants the number to move, which is Part 3.

Subtract the composition-only gap from the reported gap and what is left is the residual: pay differences inside bands, plus whatever the fitted ladder gets wrong for that employer. Its median is 2.3 points, with a wide spread, and 33.8 percent of employers report a gap below their composition-only gap, which means within-band pay at those employers runs in women's favour, or their own ladder is flatter than the fitted one. The filing cannot tell those apart. It knows nothing about pay within a band, nothing about hours, occupation, tenure or part-time status, and nothing causal.

Measuring within-band pay properly needs the salary table and a regression: pay on grade, tenure and sex, with the method that splits a gap into "explained by measured differences" and "not explained", named Oaxaca-Blinder after the two 1973 papers. Deep-dive session 6 does it on constructed Apex salaries, with the warning that "explained" is not the same as "justified".

A gap made of composition moves when the composition moves, and composition is who is in which band. Adding people to the bottom band adds to the pyramid; adding them to the top band subtracts from it. A hiring plan that brings in women mostly at entry level, which is where most hiring happens, makes the reported gap wider at the next filing while the company is doing what it said it would. The plan is not wrong. The metric is measuring composition, and composition just got more pyramid-shaped.

The honest sentence for a plan is therefore about bands, not headcount: "we intend to change the share of women in the top band from 40.0 to X", with the date. Promotion into the band and hiring into the band both count; hiring into the bottom band counts the other way.

The arithmetic is modelled, and every assumption is on the figure: the median employer's band profile from the 2024 filings, the fitted ladder 1 / 1.20 / 1.45 / 2.80, equal pay within every band, and women added equal to 10 percent of headcount with nothing else changing. Under those assumptions the composition-only gap starts at 9.57 percent. Put the added women in the bottom band and it becomes 14.88. Put them in the top band and it becomes -3.18, a gap in women's favour.

None of that is a forecast for any real employer, whose ladder is steeper or flatter than the fit and whose within-band pay is unknown. It is a demonstration that the sign of the effect depends on the band, which is true under any ladder that rises. Deep-dive session 10 runs the same two arrows on a 3,000-employer sample of the real filings, live, where the numbers come out as 9.79, 15.07 and -2.92 on the sample's own profile.

The Office for National Statistics publishes the UK gender pay gap from the Annual Survey of Hours and Earnings: for April 2024, 7.0 percent for full-time employees and 13.1 percent for all employees (2023: 7.5 and 14.2). It is tempting to hold 8.4 beside 7.0 and conclude that the median filer is slightly worse than the country. The two numbers are not comparable. The ONS figure is the median hourly earnings of all women in the economy against all men, one calculation across everyone. The 8.4 is the median of 11,315 employer-level medians, each computed inside one workforce. An economy where every employer had a zero gap could still have a national gap, if women worked at lower-paying employers; and the reverse. The two numbers answer different questions and both are worth knowing. Write which one a slide is quoting.

Goldin 2014, "A Grand Gender Convergence: Its Last Chapter", American Economic Review 104(4) 1091-1119, argues that most of the historical gap has closed and that the remaining gap is driven by occupations that reward long and particular hours disproportionately: jobs where being available at specific times, and for many of them, is paid at a premium rather than pro rata. That is a statement about the structure of pay across occupations, not about any employer's filing, and it is one explanation for why the top band of the pyramid looks the way it does at so many employers. The filing cannot see hours or occupation, so it cannot test this; a company that wants to know whether its own top band is shaped by hours has to go and look.