The Skill

The plate on the margin left a question open. The wage, it argued, tends to the value of the worker's marginal product — but the marginal product of a surgeon is many times that of a porter, and the theory said nothing about why. What makes one worker's last hour worth so much more than another's? The dominant answer of the last half-century is a single transforming idea: that skill is capital.

Before the 1960s, economists meant by capital the machines, the buildings, the land — the physical equipment of production. Theodore Schultz and Gary Becker proposed that the most important capital of a modern economy is lodged instead in people: in their knowledge, their training, their health, their skills. Education and training, on this view, are not consumption, to be enjoyed and used up; they are investment — outlays made now, in money and in forgone earnings, that raise a worker's productivity, and so their earnings, later, exactly as a firm's investment in a machine raises its output and its returns.¹ The worker becomes a small capitalist whose capital is themselves, and the wage of a skilled worker is in part a yield — the return on the years and the money sunk into acquiring what they know.

Jacob Mincer turned the idea into an instrument. He wrote down a spare equation relating a worker's earnings to their years of schooling and their years of experience, and showed that this short formula captured a remarkable share of the variation in what people are paid.² The human-capital earnings function became one of the most estimated relationships in all of economics, and its two coefficients — the return to a year of school, the return to a year of experience — became the basic measuring rods of the worth side of the wage.

What, then, is a year of schooling worth? The answer, repeated across decades and countries and datasets with unusual constancy, is that each additional year of education is associated with something like seven to ten percent higher earnings.³ A half-century of estimates has not shifted the order of magnitude; it is among the most robust regularities the discipline possesses. And by the logic of the margin, this is just what one should expect if schooling raises the value of the work: more education, a higher marginal product, a higher wage.

Experience tells a subtler story, and it is the more revealing of the two.

Earnings do not rise with experience in a straight line. They climb steeply in the early years and then flatten, sometimes turning down near the end — the concave profile Mincer's curves made famous. The shape is exactly what the investment view predicts: a worker accumulates human capital fastest when young, through formal training and the steep early learning of a job, and ever more slowly thereafter, as the returns to further learning diminish and old skills begin to depreciate.⁴ It is the law of diminishing returns from the plate on the margin, now turned on the worker's own self-improvement: each year of experience, like each additional worker, adds a little less than the year before.

The profiles also fan apart, and this is the fact that matters most for the volume. The gap between the earnings of the schooled and the unschooled does not merely exist; it widens across a career. More schooling buys not only a higher starting wage but a steeper climb and a higher ceiling. Skill, in the language of the opening plates, raises the top of the bargaining range — and raises it by more the longer the worker remains in the market. The worth side of the wage, which the margin described in the abstract, takes on here its concrete human shape: education and experience, lifting the value of the work.

Becker drew a distinction within human capital that proved unexpectedly deep, and that quietly connects this plate to the one on the single buyer. Some skills are general: useful to many employers — literacy and numeracy, a professional qualification, the ability to write or to code. Others are specific: useful only, or chiefly, to the employer who provided them — familiarity with this firm's idiosyncratic systems, its customers, its particular way of doing things.⁵

The distinction settles who pays for the training and who keeps its return, and the logic is elegant. General skills the worker must pay for, typically by accepting lower wages while training, because an employer who paid could never recoup it: the newly skilled worker, now more productive everywhere, could simply leave for a rival willing to pay the full value of skills it had not funded. So the worker bears the cost of general training and pockets the return — the higher, portable wage the skill commands anywhere.

Specific skills are different, and the difference bears on power. Because a firm-specific skill is worth little to any other employer, the worker cannot credibly threaten to take it elsewhere, nor can the employer easily be undercut for the worker's labor; the investment and its return are therefore shared between them. The sharing binds the two together — it is one reason wages tend to rise with tenure, why long-serving workers are loath to leave and their employers loath to lose them.⁶ But notice what specific capital does to the bargaining range. It lowers the worker's outside option — the floor of their personal bargain — without lowering the value of their work to this employer. The gap between the two widens, and into it steps the very employer power the eighth plate described: a worker who would forfeit the value of their specific skills by leaving is, to that extent, captive, and a captive worker can be paid less than they are worth. Skill, which raises the ceiling, can in its specific form also deepen the well of monopsony power that holds the wage down.

Against the whole edifice stands a rival so simple, and so corrosive, that economics has never fully digested it. What if education does not raise productivity at all — but merely reveals it?

Michael Spence set out the logic in 1973. Suppose workers differ in some ability employers cannot observe directly and cannot cheaply test. Suppose, too, that the more able find school easier — that earning a credential costs the talented less, in effort and in struggle, than it costs the rest. Then education can serve as a signal: the able acquire the diploma because it is cheap for them to acquire, the less able do not because it is dear, and employers, unable to see ability directly, rationally pay for the credential as a proxy for it. On this account the schooling need not have taught the worker anything useful at all. It has only sorted the able from the rest — and the wage premium for education is the price of a sorting device, not the return on a skill.⁷

The two theories are maddeningly hard to tell apart, because they make the same prediction about the individual: the educated earn more, whether schooling built their productivity or merely flagged it. They diverge only on the deeper question — whether the schooling did anything — and the evidence is genuinely mixed. For signalling: the “sheepskin effect,” the disproportionate jump in earnings that comes from completing a degree rather than from the same years of study left unfinished, which is hard to square with learning and easy to square with the credential.⁸ For human capital: the many natural experiments in which schooling, when forced on people who would not have chosen it, does raise their later earnings and their measured skills. The honest verdict is that the premium is some of each — part built skill, part revealed sorting — in proportions no one has pinned down. But the stakes of the split are high. To the extent the premium is real skill, educating more people enriches the whole society; to the extent it is mere signal, mass education is partly a positional arms race, costly to each and barren to all, in which everyone runs harder to hold the same place.

If the skill premium is a price — and whether it rewards built skill or revealed ability, it is at least that — then like any price it is set by supply and demand, and like any price it moves. Its movement over the last half-century is one of the central facts of modern inequality, and it has a name: the race between education and technology.

The framing goes back to Tinbergen and was built into a full account by Goldin and Katz.⁹ The demand for skilled labor, they argued, has risen more or less steadily for a century, pulled up by technologies — most recently the computer — that complement educated workers while substituting for the routine work of the less educated.¹⁰ The supply of skilled labor has risen too, as each generation went to school longer than the last. The skill premium is the price that clears this market, and it rises or falls with which side of the race is winning: when the supply of the educated outruns the demand for them, the premium falls; when demand outruns supply, it climbs.

For most of the twentieth century, supply won, and the premium drifted down or held steady. Then, around 1980, the race reversed. The growth of college-going slowed while skill-biased technology accelerated; demand pulled ahead of supply; and the college wage premium roughly doubled in two decades — from around forty percent to near seventy by the year 2000. That rising premium is one of the principal engines of the great widening of the American wage distribution, the worth-side counterpart to the power-side story of the union's collapse: the ceiling rose, and rose fastest for those already near the top of it.

But a price set by a race can stop rising when the race changes, and this one has. Since about 2000 the college premium has plateaued, and lately slipped — to around seventy-five percent by the early 2020s, a little below where it stood at the turn of the century. The reasons fit the framework precisely: the supply of graduates kept growing, while the demand for them stopped racing ahead, the technical change of the new century proving, on the evidence, roughly neutral between the educated and the rest rather than relentlessly favoring the educated as the previous decades' had. Tight labor markets, meanwhile, lifted the wages of workers without degrees. And as the premium flattened, the cost of a degree went on climbing, so that the net return to college — the premium set against the rising price of obtaining it — has quietly eroded.¹¹ The lesson is the one the plate keeps circling back to: the reward to a skill is not a fixed property of the skill. It is a price, set by how scarce the skill is — and scarcity can fall.

Three limits hem the theory in, and naming them is the price of taking it seriously.

The first is a problem of measurement that haunts the headline number. The return to schooling — the seven to ten percent a year — is a correlation, and those who get more schooling are no random sample: they tend to be more able, more driven, better placed, in ways that would have raised their earnings even without the extra years. Some of the measured return to education is therefore really a return to the ability that produced the education, not to the education itself — the problem of ability bias. The cleaner studies, which lean on twins, or on compulsory-schooling laws, or on other accidents that push people into more education for reasons unrelated to their ability, generally do find a genuine causal return — schooling does raise earnings — but a smaller one than the raw correlation reports.¹² The worth side is real, but it has been somewhat oversold.

The second limit is the one the whole volume presses. Skill raises the ceiling of the bargaining range; it does not fix where, beneath that ceiling, the wage falls. A highly skilled worker has a high marginal product — a high ceiling — and may still be paid well below it if they face a single dominant employer, or lack the power to claim it, or belong to a group for whom the floor is set low. Skill tells you how high your ceiling stands. It is silent on whether you reach it. The two questions are independent, and running them together — supposing that the skilled are well paid and the well paid skilled — is among the most common errors in all thinking about wages.

The third limit has surfaced from the data only lately, and it is quietly devastating to the purest form of the theory. If the wage were simply the return to a worker's own human capital, two workers with identical skills would earn the same wherever they worked. They do not. A large body of evidence now shows that the same worker earns systematically more at some firms than at others — that a substantial part of what anyone is paid is a premium attaching to their employer rather than to themselves.¹³ Where you work, it emerges, can matter as much as what you know. And a wage that depends on which firm happened to hire you is no pure return to skill; it is, in part, a share of the rents that firm commands — which is to say, once more, a question of power, and not only of worth.

Human capital is the most humane-sounding of all the theories of the wage, and that is exactly what makes it the most dangerous when it overreaches. It says that your pay reflects your skills, and that your skills are yours to improve — that the wage is, in the end, a measure of what you have made of yourself. There is real truth in this, and real dignity. Education and training genuinely do raise the value of work; the return to schooling is no illusion; a person can, by learning, lift their own ceiling. A volume that denied as much would be false to the evidence.

But the same comforting logic, pressed too far, becomes an instrument for naturalizing every wage as deserved. If pay is the return to skill, then low pay is low skill — a deficiency in the worker, rather than in their bargaining position, or their employer, or the structure of their market. The language of human capital can quietly convert the verdict of power into the verdict of worth: the underpaid are simply the under-skilled, and the remedy is forever more training and never more power.¹⁴ This is where the volume must hold the line. Skill raises the ceiling — genuinely, measurably, and to the worker's lasting credit. It does not set the wage. The wage is where the bargain comes to rest beneath that ceiling, and where it rests is decided by the things this plate has kept brushing up against and the plates to come will name outright: the single buyer, the missing union, the discrimination that lowers the floor, the firm that keeps the rents.

To ask of any wage how much is skill is to ask only how high the ceiling stands. The harder question — the volume's question — is how much of the room beneath that ceiling the worker was able to claim.