Your sister does the same kind of work as the two men who sit either side of her, and she is paid less than both. She noticed it slowly, then all at once, the way these things surface: a salary band mentioned in passing, a bonus that landed heavier on the desk next to hers. What she cannot stop turning over is the timing. For years the three of them tracked more or less together. Then her daughter was born, she took the leave, she came back, and somewhere in that stretch the lines pulled apart and never rejoined. She wants to know what happened, and the first thing to see is that at least three different things are hiding inside the single word “gap.”
The claim she has heard, the one printed on placards and in headlines, is that women earn some fixed fraction of what men earn for the same work. The number moves around; a familiar version puts it near 80 cents on the dollar. That figure is real, it is measured carefully, and it does not mean what the slogan says it means, because the number and the phrase attached to it come from two different measurements that the slogan fuses into one. Pulling them apart is the whole task of this chapter. Once the pieces are separated, most of what looks like a single mysterious penalty turns out to be the accumulated result of which jobs men and women hold, how many hours they work, and, running through both, what happens to a woman’s career the year a child arrives. What that leaves unexplained is smaller than the slogan and harder to interpret than either side usually admits.
Two numbers, and why neither is the discrimination number
Start with the 80-cent figure. It is a ratio of medians: take the annual earnings of the woman standing at the midpoint of all women who work full time, year round, take the same for men, and divide. In the United States that ratio has sat in the high seventies to low eighties of cents for some time; the 2024 reading is 80.9. It is a broad comparison, and its breadth is the point. It makes no attempt to compare a woman and a man doing the same job for the same hours with the same experience. It puts the typical full-time woman’s earnings against the typical full-time man’s, and so it folds together every difference between the two groups at once: that men and women are distributed differently across occupations and industries, that they average different hours and different years of continuous experience, and whatever is left after all of that. Call this the raw gap. It answers a real question, which is how the typical woman’s pay compares to the typical man’s among full-time, year-round workers. It does not answer the question in the slogan, which is about the same work.
To get at the same-work question, economists do something different. They take a large sample of workers, measure each person’s pay and their measurable characteristics, and statistically hold those characteristics equal, asking how much of the gap survives once you compare men and women who match on education, experience, industry, occupation, and the rest. The gap that remains after that matching is the adjusted gap, and it is always smaller than the raw one. The temptation is enormous to call the adjusted gap the discrimination and the rest fair. That step is a common and consequential mistake, and it is wrong in both directions at once.
It is wrong in one direction because the leftover, the adjusted gap, is not clean. Some of it may be discrimination, but some of it may be productivity differences the data never measured, so the residual can be larger than the true unequal-pay-for-equal-work component. It is wrong in the other direction because the pieces the adjustment removes are not automatically innocent. When the procedure “explains” part of the gap by the fact that women are concentrated in lower-paying occupations, it has explained the gap in the narrow statistical sense and settled nothing about why women are concentrated there. If women are steered away from higher-paying fields, or crowd into others because those others accommodate the second shift at home, then discrimination and constraint are sitting inside the explained part, quietly counted as accounted for. The economists who built the standard decomposition say this plainly: conditioning on occupation and industry is itself a contestable choice, because those variables may carry the very forces the exercise is trying to isolate. So the adjusted number is not a floor on unfairness and the explained number is not a certificate of fairness. Neither number is the discrimination number, because no single number is.
What the pieces add up to
With that warning in hand, the decomposition is genuinely useful, because it shows the size of each measurable piece instead of leaving the gap as one undifferentiated lump. The figure below draws it from a survey that follows the same American workers year after year and records their actual work histories rather than guessing at experience from age.
Read the sizes rather than the labels. In this data the largest single measured piece is occupation: men and women hold different jobs, and those jobs pay differently, and that difference alone accounts for roughly a third of the gap. Industry accounts for roughly another sixth, and years of actual experience for a little less than that. Education runs the other way now, because women in this sample are the better-educated group, so their schooling works to shrink the gap and its reversal shows up as a small step widening the part left over. When all the measured factors are netted out, a bit under two-fifths of the original gap has no measured explanation at all. That leftover is where the word “discrimination” is usually stamped, and it is exactly the piece the previous section warned cannot bear the stamp cleanly in either direction.
The question underneath the question
The decomposition does not close the case. It relocates it. Saying that occupation and hours account for most of the gap only pushes the question back a step: why do men and women end up in different occupations, and why are the hours structured the way they are? Those are not raw facts of biology handed to the labour market. They are outcomes, and the volume’s opening chapter gave the vocabulary for reading them, because pay turns on which position each worker bargains from rather than on the worth of the work. Which field a person can enter, how continuously they can stay in it, and whether the hours bend around a home are all part of that position, and they are not distributed evenly between the sexes. So the “explained” part of the gap is not the part that needs no explaining. It is a different explanation, one about sorting and constraint rather than about a manager choosing to pay a woman less for identical work at the same desk. The decomposition tells us the gap runs mostly through the doors people walk through and the hours they can offer. It does not tell us those doors and hours were freely chosen.
Which raises the obvious question of when the sorting happens. If you could watch a large group of men and women across the exact years their careers diverge, you would not have to infer the story from a single snapshot. You could see it move.
The year the gap opens
That is what a second body of research does, and it locates the moment the gap opens. Using the administrative records of an entire country, economists tracked the earnings of men and women across the years surrounding the birth of their first child, lining everyone up by the year the child arrived rather than by calendar year. Before the birth, the earnings of the mothers and fathers to be move together. At the birth, they split, and they never rejoin.
The size of that finding is easy to miss. In this Danish data, the entire long-run gap between men and women can be traced, to a first approximation, to what children do to mothers’ careers and not fathers’. The researchers put a number on it: the share of the remaining man–woman earnings inequality attributable to children rose from around 40 percent in 1980 to around 80 percent by 2013; the other sources of the gap shrank while the child penalty held, and children did not come to matter more in absolute terms. This is why the chapter’s two figures belong together. The American decomposition says the gap runs mostly through occupation, industry, and experience; the Danish event study, tracking annual earnings in a different country, watches differences of the same kind open at the first birth, when a mother’s hours contract, her promotions slow, and her job often shifts toward an employer that accommodates a child. The sorting the first figure measures is, in large part, the kind of sorting the second figure watches happen.
Two boundaries keep this from being oversold. The first is that it is descriptive. The event study shows that the break coincides with the child, and the short-run jump is hard to attribute to anything else; the long-run level rests on the assumption that fathers make a fair comparison group for mothers, which is defensible and not airtight, and the design cannot by itself separate an employer marking a mother down from a mother trading pay for hours she needs. It identifies the moment, not the full mechanism. The second is that the magnitude is a Danish magnitude, measured in registers that cover an entire population. Comparable studies find child penalties that differ a great deal across countries when defined the same way, as the average penalty across the fifth through tenth years after the birth: near 21 percent in Denmark, closer to 31 percent in the United States, and above 60 percent in Germany. The mechanism travels; the size is local.
Back to the sister, and the number on the placard
Return to your sister with the pieces separated. The 80-cent figure she has heard describes something true about the whole labour market, that women’s pay in aggregate runs well below men’s, and it is driven mostly by the differences the decomposition lays out; where the trace has been run in full, those differences open at the arrival of a child. The same-work version of her situation, a woman and a man matched on job, hours, and experience, shows a smaller gap, and even that smaller gap cannot be cleanly labelled, because it may hide unmeasured differences or, just as easily, hide the discrimination that pushed her toward a lower-paying track in the first place. Her specific case may be any of these. She may be paid less than the men beside her for genuinely identical work, which is unlawful and real and happens. She may be on a track that pays less because it demanded less travel after her daughter was born. The pieces can also combine, and what the evidence can actually tell her is which questions to ask rather than a single verdict to recite. The one answer it rules out is the slogan she started with, because that slogan glues a whole-market number to a same-work claim and calls the join a measurement.
Women earn 77 cents on the dollar for the same work.
Oversimplified Moderate confidence
The claim welds two different measurements together and presents the join as one fact. The cents figure, whether it is quoted at 77, 80, or thereabouts, is a ratio of median annual earnings between women and men who work full time, year round. It is real and carefully measured, and it is the reason the ruling is not the reverse of the claim: women in aggregate do earn substantially less than men, so there is a true and large gap at the bottom of this. But “for the same work” imports a different measurement, the gap that remains after men and women are matched on job, hours, and experience, and that adjusted gap is considerably smaller than the raw cents figure. Attaching the raw number to the same-work phrase overstates the same-work gap by counting differences in occupation, hours, and experience as if they were unequal pay for identical work. The deeper trouble is that neither number is the discrimination number the slogan wants. The adjusted residual is not clean: it can be inflated by productivity the data never observed, or deflated because the sorting into lower-paying jobs, which the adjustment credits as explained, may itself be where discrimination and constraint are doing their work. Most of the gap runs through which jobs and hours each parent ends up with, and those differences open, where the full trace has been run, at the first birth. So the claim is not false, and it is not backwards; it takes a real, large, whole-market gap and mislabels it as a same-work penalty, while hiding what actually drives the gap, which is mostly about which doors open and when a career bends around a child.
Sources
- The measured decomposition: F. Blau and L. Kahn, “The Gender Wage Gap: Extent, Trends, and Explanations,” Journal of Economic Literature 55(3), 2017, Table 4 (PSID, full-time workers 25–64) — the 2010 raw gap of 0.2314 log points (women about 79 percent of men per hour) splits into about 62 percent accounted for by occupation, industry, experience and other measured factors and about 38 percent unexplained, with the residual larger still under a human-capital-only specification. The authors themselves caution that conditioning on occupation and industry is contestable, so the “explained” share is not thereby justified and the residual is not a clean discrimination measure. Units are log points of hourly wages, not cents. Verified against IZA Discussion Paper 9656.
- The distinction between the raw and the adjusted gap: the raw ratio is the Census Bureau’s comparison of median annual earnings between women and men working full time, year round (the source of the 77- and 80-cent figures; 80.9 cents in the 2024 data); the adjusted gap holds measured characteristics equal and is smaller. Neither is a direct measure of unequal pay for identical work.
- The timing signature: H. Kleven, C. Landais and J. Søgaard, “Children and Gender Inequality: Evidence from Denmark,” American Economic Journal: Applied Economics 11(4), 2019 — women’s earnings drop almost 30 percent at the first birth and settle about 20 percent below, men’s show no visible change, and the long-run child penalty at ten years is 19.4 percent; the share of gender earnings inequality attributable to children rose from about 40 to about 80 percent between 1980 and 2013 in Denmark. Comparable cross-country penalties, defined as the average over years five to ten, run near 21 percent in Denmark, 31 in the United States, and above 60 in Germany.
- Confidence is moderate under the rubric, which scores the weaker of evidence directness and construct match. Evidence directness is reasonable: the decomposition is well measured and the child-penalty event study identifies the timing sharply. The binding, weaker leg is construct match. The claim fuses a raw all-worker ratio with a same-work assertion, and the nearest same-work measurement, the adjusted residual, cannot itself be equated with discrimination in either direction. Because the number the claim quotes and the thing it asserts are different constructs, and the cleanest same-work construct is itself not a clean discrimination measure, the construct mismatch is what caps the confidence.
Why the pieces matter more than the number
The pay gap is a case where a single headline number does more to end thought than to start it. The raw ratio is true and worth knowing, and so is the fact that it is not a same-work figure, and so is the finding that, where the full trace has been run, most of the gap leads back to the year a first child arrives. Holding those together is harder than repeating the slogan or dismissing it, and it points somewhere the slogan never does: at the structure of jobs and hours and care that routes mothers and fathers onto different tracks, long before any manager sits down to set two salaries. That structure, and in particular the care work that goes unpaid and uncounted while it reshapes a career, is a chapter of its own later in this volume, on why the same task counts in the economy’s ledger when a nanny is paid to do it and vanishes from the ledger when a parent does it instead.
The next flashpoint keeps the leverage lens but turns it on a different fear. Where this chapter asked why two groups already in the workforce are paid differently, the next asks what happens to everyone’s wage when the workforce itself is enlarged from outside. The worry that new arrivals push your pay down is, at bottom, the same worry about the supply of replaceable labour that runs under all of these questions, and whether immigration lowers wages is where the volume goes next.