Discussion of the gender pay gap generates a predictable argument in which two sides talk past each other, because they're citing different statistics without saying so.
Distinguishing them resolves most of the confusion and leaves a more interesting question.
The two main measures
The unadjusted gap compares average or median earnings of all women against all men in a workforce, without controlling for anything.
This is the figure usually quoted in headlines and in mandatory reporting requirements. It measures the overall difference in what men and women earn.
The adjusted gap compares earnings after controlling for factors including occupation, industry, hours, experience and education.
This figure is substantially smaller, typically a small number of percentage points, and is the one cited by people arguing the gap is overstated.
Both are accurate measures of different things. The argument is largely about which question matters.
What the adjusted gap leaves out
The critical point. Controlling for occupation removes from the measurement the question of why occupational distribution differs.
If women are concentrated in lower-paid sectors, controlling for sector removes that from the gap. Whether that concentration reflects free choice, constrained choice, or the systematic undervaluing of work done predominantly by women is precisely the interesting question, and adjusting removes it.
There's evidence relevant to this. Studies examining occupations that have shifted in gender composition over time have found that pay tends to fall, relative to other occupations, as an occupation becomes female-dominated — which is difficult to explain purely by the nature of the work.
So the adjusted figure answers "are women paid less for the same job", which is worth knowing. The unadjusted figure captures something broader about how work is distributed and valued.
The motherhood effect
Research using long-run administrative data has consistently identified childbirth as the dominant driver of the earnings divergence between men and women.
The pattern found across multiple countries: earnings of men and women follow similar trajectories until the arrival of a first child, at which point women's earnings drop sharply and do not fully recover, while men's are largely unaffected.
The magnitude varies by country and appears related to policy — parental leave design, childcare availability and cost, and cultural norms about caregiving.
This has shifted how economists discuss the issue. The gap is substantially a parenthood gap rather than a gender gap in a general sense, which points at different remedies.
Other contributing mechanisms
Hours and flexibility. Research on the structure of pay in some professions has found that earnings rise disproportionately with long and inflexible hours. Occupations where output is not substitutable between workers — where being present at specific times matters — show larger gaps.
Which suggests that some of the gap is a premium on availability rather than on productivity, and that jobs designed to be less dependent on individual presence show smaller gaps.
Negotiation and progression. Evidence here is mixed and more nuanced than popular accounts suggest, discussed elsewhere.
Discrimination. Audit studies using matched applications have found differential treatment in some contexts, with results varying by field and by study.
What reporting requirements have achieved
Several countries now require larger employers to publish gender pay gap figures.
Evaluations have found modest narrowing in affected organisations, with some evidence that the mechanism operates partly through reputational pressure.
Criticism of these schemes is reasonable: publication without required action allows organisations to report a gap indefinitely, and narrative explanations accompanying figures are frequently unaudited.
There's also a perverse possibility — that a gap can be reduced by reducing high earnings or by outsourcing lower-paid roles, neither of which helps anybody.
What appears to work
Interventions with some supporting evidence.
Pay transparency. Studies of transparency requirements have generally found narrowing of gaps, apparently through reducing the scope for individually negotiated variation.
Salary history bans. Prohibiting employers from asking about previous pay, on the reasoning that it perpetuates existing gaps. Evaluations have found some positive effects.
Non-transferable parental leave for fathers. Reserved leave that cannot be transferred has been found to increase take-up substantially, with associated effects on the division of subsequent caregiving. Countries with such policies show different patterns.
Affordable childcare. Consistently associated with maternal labour force participation.
Structured pay-setting. Reducing manager discretion in individual pay decisions reduces variation attributable to factors other than role.
The more useful question
Rather than arguing about the size of the number, the productive questions are about mechanism.
Why does the arrival of children affect one parent's earnings and not the other's, and what policy choices produce that?
Why does work become less well paid when women do more of it?
Why do the highest-paying roles require a pattern of availability that's incompatible with caregiving, and does that reflect genuine requirement or accumulated convention?
Those questions have answers that can be acted on. The argument about which statistic to cite mostly doesn't.