Monopsony

The competitive picture of a labor market is clean enough to teach in an afternoon, and for most of a century it was taught in every first one. Employers are many and each is small — too small to move the wage — so each takes the going rate as given and hires until the value of one more hour of work equals the wage paid for it. Workers offer more hours as the wage rises. The two schedules meet, the market clears, and every worker is paid, to the penny, the value that the last hour of their labor adds.

Set a wage by law above that meeting point and the logic is unforgiving. At the higher wage employers want fewer hours than before, and workers offer more; the gap between them is people who would take the job at the legal wage and cannot find it. This was never advanced as cruelty. It was advanced as arithmetic — and it grounded the standard objection to every minimum wage and every limit on hours: whatever the humane intention, the worker at the margin pays for it with the job.

There had been a dissenting intuition, older than the model that buried it. Adam Smith, two centuries before the diagrams, had noticed that the two sides of the wage bargain were not evenly matched — that masters were fewer, could hold out longer, and were “always and everywhere in a sort of tacit, but constant and uniform combination” not to let wages rise.¹ The marginalist synthesis that followed was so clean, and so powerful, that it quietly overwrote the observation. The asymmetry Smith saw was assumed away, not answered.

And it was assumed away in a single stroke. Strip the competitive model to its load-bearing member and what remains is one number: the elasticity of the labor supply that an individual firm, by itself, faces — how freely workers flow toward it when it pays a little more, and away from it when it pays a little less. The clean picture sets that number to infinity. Raise your wage by a hair and the world’s workers queue at your door; cut it by a hair and they are gone to a rival by morning. Every humane or cruel thing the model has ever been made to say descends from that one silent setting.

Relaxing it does something more radical than add a special case. It reverses the order of things. Perfect competition is not the rule from which monopsony departs; it is the single frictionless point — labor infinitely mobile, information perfect, every worker indifferent between identical jobs — toward which real markets only tend and never arrive. Monopsony, in this light, is not a pathology that afflicts the occasional market. It is the general weather, and competition the rare cloudless day. The achievement of the next hundred pages of economic history was to ask what the wage looks like on every other day.

Joan Robinson asked it, in 1933, and gave the contrary case a name — monopsony, the single buyer, a word a Cambridge classicist supplied her as the mirror of monopoly.² Where a firm looms large in its labor market — the mill in the mill town, the one hospital a region’s nurses can reach — it cannot draw in one more worker at the old wage. To attract the next, it must raise the offer, and pay the raise to everyone already on the payroll. The true cost of that marginal worker is therefore greater than their wage; and a firm watching its costs stops hiring before the competitive point. It employs fewer people, and pays them less than the value their work adds. No one designs this; it falls out of the shape of the exchange.

The gap that opens has a precise name and a precise size. Pigou called the distance between what a worker produces at the margin and what the worker is paid exploitation — not a moral charge but a technical measurement, the wedge a buyer’s power drives between value and price.² Robinson supplied its arithmetic, and the arithmetic is unexpectedly stark. How much of a worker’s product is withheld turns on one quantity and one only: that firm-level supply elasticity. Where the elasticity is infinite, nothing is withheld and the worker is paid exactly their product; where it is four, the wage is four-fifths of the product and the remaining fifth is kept; where it is lower, less of the work’s worth reaches the worker still. The fraction withheld is one divided by the elasticity plus one — exploitation, in this exact and bloodless sense, shrinking only as labor grows freer to leave.

And then the turn that made the idea dangerous. A minimum wage set between the depressed wage and the competitive one does not, in this world, price workers out of work. It prices out the firm’s power to underpay. Once the wage is fixed by law, the firm can no longer push it down by holding hiring back — so it stops holding back, and hires more. Pay rises, and employment rises with it. The arithmetic does not merely soften. It inverts.

A floor set right prices out not the worker, but the power to underpay him.

The inversion has a limit, and the limit is the whole of the policy question. Begin at the depressed wage and raise the legal floor: employment climbs, because each increment strips away a little more of the firm’s power to hold hiring back, and the firm answers by hiring. But the climb does not continue forever. It peaks precisely at the wage a competitive market would have set — the point where the firm’s power is fully neutralized and not a cent of markdown remains — and beyond that point the floor is no longer disciplining power but pricing labor above its worth, and the old arithmetic returns to collect the jobs it was always owed. Employment, plotted against the minimum wage, is therefore not a line but an arch: rising, cresting at the competitive wage, then falling. The employment-maximizing minimum wage is exactly the wage that monopsony suppressed.

This is why the theory licenses no floor whatever, and why every serious disagreement is, beneath the noise, the same disagreement — not whether a minimum wage helps or harms, but where on the arch a particular market currently stands. For sixty years the whole construction remained a chalkboard curiosity, and for a plain reason.

Who, in a modern economy of a thousand employers, is a single buyer of labor? Fast food seemed the least promising place on earth to look — countless outlets, low barriers, the very picture from the textbook. Which is precisely what made it the right place to test.

In 1992 New Jersey raised its minimum wage and Pennsylvania, next door, did not. David Card and Alan Krueger surveyed several hundred fast-food restaurants on both sides of the line, before the raise and after, letting Pennsylvania stand in for the New Jersey that might have been.³ The model predicted lost jobs in New Jersey. They found none — employment held, and by their count edged upward.

The result was received as heresy and handled accordingly: reworked with payroll records by skeptical hands,⁴ defended by its authors, and then tested again across every pair of counties that face each other over a state line in America.⁵ It largely held. The deeper revolution was the method itself — reading the labor market’s natural experiments rather than its correlations, treating a state border as the boundary of an accidental trial — and that method, more than any single finding, would carry Card to a share of the 2021 Nobel.³

It would be tidy to read the missing pink slips as monopsony confirmed — tidy, and tempting, and premature. A minimum wage that fails to destroy jobs is consistent with employer wage-setting power, but it is consistent with much else besides: with firms passing the cost forward into prices, with the raise paying for itself in lower turnover and fewer vacancies left gaping, with efficiency gains as better-paid workers quit less and work harder, with a floor simply set too low to bind. The absence of disemployment is a necessary symptom of monopsony, not a sufficient proof of it. The proof, if it is anywhere, lies in the elasticity itself — which is where the inquiry turned next, and to which this account will return when the stakes of the turn are clear.

The evidence has since grown both larger and more careful, and it is worth stating precisely rather than as a slogan. The most comprehensive reckoning pooled a hundred and thirty-eight separate minimum-wage increases and asked a sharper question than the early studies had: not whether some headcount fell, but where the jobs paying below the new floor actually went.⁶ They reappeared, almost exactly, just above it — the jobs lost beneath the new minimum rematerializing as jobs gained at and around it, the total stock of low-wage work essentially unmoved, and unmoved even where the floor was high relative to local wages. The adjustment was real, but it ran along other seams: into prices, into thinner margins, into lower turnover, into capital quietly substituting for labor at the edges. One exception proved the rule. In industries exposed to outside competition — where a restaurant’s trick of passing the cost to its own walk-in customers is unavailable — some employment did fall, exactly where the theory says the cushion of local market power runs out.

The sharpest dissent came from Seattle, whose unusually steep increases were read by one team as cutting the hours of the lowest-paid.⁷ The finding is genuine and the argument over it unresolved, but it rests on a single city in the middle of an extraordinary boom, which makes the counterfactual — what Seattle’s low-wage hours would have done without the raise — unusually hard to pin, and a single case unusually hard to generalize from. The honest summary is not that minimum wages never cost jobs. It is that within the range legislatures actually choose, the measured employment effect clusters near zero, and the live argument has narrowed from the sign of the effect to its size, and to the height of the floor.

But the puzzle had only moved. There is no single buyer of fast-food labor; why should the market behave as though there were? Alan Manning’s answer, worked out at length in 2003, was that the single buyer was never the point.⁸ You need only friction.

Workers do not flow without cost to the highest wage. Searching takes time, information is poor, quitting is dear, and a life is anchored to a place — a school, a lease, a family. Each of these hands every employer a portion of the single buyer’s power, because nudging the wage up a little draws a few more applicants and keeps a few more workers who would otherwise drift off, while nudging it down sheds them slowly rather than all at once. The supply of labor to the individual firm slopes upward even with a hundred rivals in town. The formal engine had in fact been built a few years earlier — a model in which firms post wages and workers, searching one offer at a time, cannot costlessly find the best, so that identical workers end up sorted across firms paying different wages, none of which is the competitive one.⁹ Its fingerprint is the wage dispersion competition was supposed to erase: alike workers paid unalike by different firms, and firms that set a wage rather than meet one.

You do not need a single buyer. You need only friction.

This matters more than a change of microfoundation, because it shows that two very different stories arrive at the same observable. Whether a firm’s power comes from sheer fewness, in the manner of the company town, or from the search friction that lets a crowded market behave like a thin one, the first-order condition is the same and the markdown is the same: one over the elasticity of the labor supply the firm actually faces.¹⁰ The empirical program therefore need not first settle which story is true in a given market in order to measure how much power is present. It need only estimate that one elasticity.

And the estimates, by now numerous, point the same way while measuring subtly different things. One approach watches how the flow of applicants to a posted vacancy responds to the wage it offers; the elasticity comes back around four-tenths — raise the wage a tenth and you draw a few percent more applicants, not an unlimited queue.¹¹ That number is not itself the markdown, for applications are not hires and the supply curve a firm truly faces is a steeper and different object; but it is hard evidence that the curve slopes at all. The quantity that does enter the markdown comes from watching how workers actually move between firms in response to pay, and there the elasticities, pooled across the best studies and corrected for the tendency of striking results to find their way into print, settle around five to seven¹² — which, by Robinson’s arithmetic, puts the representative firm a tenth to a quarter below the marginal product of its workers. The decisive fact is the source of that wedge. Because it is built from friction and not from fewness, it does not wait for a thin market to appear: it persists where employers are many, in the thick of a city as much as in the isolation of a company town, and concentration where it exists only deepens it further. That is why monopsony, on these numbers, is not the affliction of the few. It is the climate.

Here the strongest objection to the whole turn must be met, because if it stands, none of the above is worth much. If a minimum wage that destroys jobs can always be explained away as a floor set above the competitive wage, and a minimum wage that does not can always be explained as monopsony, then the theory appears to accommodate every outcome and forbid none — and a theory that forbids nothing predicts nothing. The objection is serious, and the answer is exactly why the elasticity matters more than the employment studies. The wedge is not read off the policy result; it is measured before any policy is applied, from how workers separate and how applicants arrive and how movers respond to pay — quantities estimated independently of any minimum wage. From that independent measurement the theory makes a prediction the data could have refuted and did not: that the firm-level supply curve slopes upward, neither vertical nor flat, with a slope that pins the markdown near a tenth to a quarter. And it makes a further family of predictions, genuinely out of sample, about where the wedge should run deepest — predictions that are the entire substance of what follows. If the power is built from friction, the markdown should be largest where the friction is thickest: over the workers who can least freely leave, and in the markets held by the fewest hands. That it is found precisely there is not ornament on the theory. It is the theory surviving the only test that could have killed it.

The first of those predictions concerns who carries the markdown, and it casts the longest shadow. Friction is not spread evenly across workers, and where it gathers, the wedge deepens. The supply of women’s labor to the individual firm is, on the evidence, less elastic than men’s: a job more often tethered to a household location fixed by someone else’s; a career more often interrupted and resumed; a search more often bounded by a shorter commute and a narrower set of reachable employers.¹³ Each is a friction, and each hands the employer a larger share of the single buyer’s power over the workers it constrains.

Monopsony therefore predicts something the clean model cannot produce at all: a gap between men’s and women’s pay that owes nothing to any employer’s taste for discrimination and everything to the different elasticities the two sides face — women paid further below their product not because anyone prefers to underpay them, but because they can less freely leave. This is not an alternative to discrimination as an explanation; it is a second mechanism, structural rather than attitudinal, that would open a gap even in a world purged of prejudice, and that the usual controls for occupation and hours will not capture because it operates through the slope of the supply curve itself. The same logic names the remedies. A wage floor compresses the gap by stripping markdown disproportionately from those who carry the most of it; pay transparency compresses it by dissolving the very informational friction that let identical work command unalike wages in the dark. That the measured wedge runs larger over women — and over nurses and teachers, occupations thin in employers and thick in workers who cannot easily move — is monopsony’s signature written into the most familiar statistic in labor economics, and the first sign that the theory predicts not just that power exists but where.

The second prediction concerns where the markdown pools, and it points at structures the law can see and, when it chooses, dismantle. If the power is built from fewness and friction, its sources can be named, measured, and curbed — and four of them have drawn the law’s attention, each encounter revealing as much about the machinery of enforcement as about the economics it enforces.

The first is concentration. Many local labor markets, defined the way workers actually meet them — a particular occupation within a particular commuting zone — hold only a handful of employers, and where they do, wages run lower; the markdown rises with concentration and falls with the elasticity, just as the theory requires.¹⁴ The measurement is genuinely hard, because the thing that confounds concentration and wages is productivity, the way cost confounds concentration and prices in the market for goods, and disentangling the two has become a discipline of its own. Its practical edge is that merger review, long concerned only with what a combination does to the prices buyers pay, has begun to ask what it does to the wages a combined employer offers — the same antitrust apparatus, turned at last to face the other side of the firm.¹⁵

The second is the covenant not to compete — friction written directly into the contract, a clause forbidding a worker to carry their skills to a rival, and so a private manufacture of exactly the immobility that hands employers their power. Something like one American worker in five labors under one, and not the engineers and executives the device was supposed to be for, but sandwich-makers and hairdressers and warehouse hands; the covenants demonstrably depress both mobility and pay.¹⁶ In 2024 the Federal Trade Commission, reasoning in precisely these terms, issued a rule voiding nearly every non-compete in the country. It did not survive the year. A federal court in Texas set the rule aside with nationwide effect, holding that the agency had no power to make substantive competition rules of this kind, and that it had in any event acted arbitrarily; a parallel court rested on the related ground that a question of this magnitude belongs to Congress, not an agency.¹⁷ By late 2025 a reconstituted Commission had abandoned its appeals and conceded the rule’s defeat; by early 2026 it had struck the rule from the books. The economics had never been what was at issue. What defeated the rule was the law of who may decide — the outer limit of administrative power, not the inner merits of the wage.

The third is the bare agreement among employers not to hire one another’s workers — Smith’s tacit combination made explicit, monopsony reduced to a handshake — and the law’s encounter with it is the most instructive failure in the whole story. In 2016 the antitrust authorities announced that naked no-poach and wage-fixing agreements would no longer be treated as ordinary civil matters but prosecuted as crimes, per se violations of the Sherman Act, the buying side’s exact analogue to a price-fixing ring.¹⁸ The theory was sound, and the courts, asked at the threshold, accepted it: judge after judge refused to dismiss, agreeing that an agreement to carve up a labor market could be criminal on its face. Then the cases reached juries, and the juries would not convict. A staffing operator who had texted a rival to cut therapists’ pay walked on the antitrust count; a dialysis chain and its chief executive, charged with a years-long agreement not to poach, were acquitted on every count, the judge offering the government his condolences; an aerospace no-poach prosecution collapsed before it ever reached the jury room. The single conviction the campaign produced came not from a verdict but from a guilty plea. Something in the lay intuition about work — that an employer declining to raid a competitor’s staff is doing business, not committing a felony — proved more stubborn than the economics, and the criminal program was effectively spent within a few years of its launch. Where collusion on the buying side was in fact undone, it was undone elsewhere: the no-poach clauses that bound roughly three in five major franchise chains, forbidding one franchisee of a brand to hire another’s workers, were dismantled not by federal prosecutors but by state attorneys general, quietly, through settlement, in the very years the criminal cases were failing.¹⁹

The fourth is the newest and the quietest: information itself. If part of the employer’s power is that a worker cannot easily learn what the work is worth elsewhere — or even what the colleague at the next desk is paid — then pay secrecy is a friction like any other, and the spreading requirement that employers post salary ranges is an attempt to dissolve it. The logic is the logic throughout: lower the friction, raise the elasticity, shrink the wedge.

And so the question fell back to the states, where the firm’s own home jurisdiction had answered it long before the federal argument was born. California has voided the non-compete since the Civil Code of 1872, and in 2024 it tightened the screw — not merely declining to enforce the covenants but obliging employers to notify their workers, in writing, that the clauses they had signed were void from the first day.²⁰ What a federal rule reached for and lost in a single year, one state had been enforcing, without drama, since Reconstruction.

The four sources are not a list but a map, and the map of sources is the map of remedies. Where the power comes from fewness, the instrument is antitrust — merger review and the policing of collusion. Where it comes from manufactured immobility, the instrument is mobility law — the voiding of non-competes and no-poach pacts, the opening of pay. And where it comes from the irreducible friction of any real labor market — the cost of being a person tied to a place, which no contract can be struck down to abolish — the instrument is the floor itself. The three are not rival theories of what to do. They are answers to three different sources of one power, and a serious wage policy reaches for all three at once. What it cannot do is what the federal non-compete rule attempted and the criminal no-poach cases attempted: carry the whole weight on a single institution. The law’s reach over employer power has tracked, again and again, not the strength of the economics but the allocation of authority — federal rulemaking blocked, criminal juries unmoved, while a single state’s old statute did the work.

Raise the lens from the single firm to the whole economy and a larger question swims into view — one the wage debate had been circling without naming. For four decades the pay of the typical worker has seemed to drift apart from what that worker produces; labor’s share of national income has fallen across the rich world. If employers everywhere hold some power to pay below the value of the work, here, surely, is the culprit.

The claim must be made with care, and the care is itself instructive. At the level of the whole economy the link between productivity and average compensation has held up better than the slogan of “decoupling” admits; much of the apparent break is the widening distance between the average worker and the median one — that is, inequality — together with the accounting of one price index against another and the slow fall in labor’s share.²¹ The divergence is real, but it is several different things at once, and monopsony speaks to only some of them. To treat the wage-productivity gap as a single quantity with a single cause is to repeat, at the level of the aggregate, exactly the error the monopsony literature spent thirty years correcting at the level of the firm.

So monopsony is a suspect, not the verdict, and there are rival suspects with alibis of their own. One account locates the great divergence in the product market: a handful of remarkably productive “superstar” firms have come to command growing shares of their industries, and because such firms run on a thin share of labor, the economy’s tilt toward them drags the aggregate down — a story about market power, but power over customers more than over workers.²² A second locates it not in market structure at all but in the broad erosion of worker power: the decline of unions, the fissuring of the firm into subcontractors and franchisees and gig platforms, the rise of shareholder primacy, the standing threat of replacement.²³ Tellingly, the economists most associated with that second account doubt that measured concentration is large enough to be the main macro driver. Labor-market monopsony of the kind this plate has described is one strand in that larger fabric — pervasive, real, worth curbing — but it is a thread, not the weave, and the discipline of not mistaking the two is the same discipline the whole subject teaches.

The practical edge of the theory is sharpest in a question that looks merely technical. If the aim is to lift the incomes of low-paid workers, why fix a wage at all — why a minimum wage rather than simply topping the wage up through the tax code, as an earnings subsidy already does? In a competitive market the subsidy is the cleaner instrument, distorting nothing. Under monopsony the choice inverts, and instructively. A subsidy paid to workers draws more of them into the market, and a firm with power over the wage answers exactly as its incentives dictate: it lets the pre-subsidy wage fall, pocketing a share of a credit meant for the worker — a capture the evidence on the largest such program confirms, with part of every dollar landing on the employer’s side of the ledger and the wages of ineligible workers dragged down beside it.²⁴ A wage floor cannot be captured this way; it sets the very price the subsidy would have let slip. The two instruments are usually discussed as substitutes; under monopsony they are closer to complements, and the floor is what keeps the subsidy from leaking to the firm.

The same point, in older language, is the case for the union. A workforce bargaining collectively against an employer with monopsony power is not extracting a premium from a competitive market but recovering a markdown from a non-competitive one, and can raise the wage toward the competitive level — the level the firm’s power suppressed — without the job losses the textbook predicts. Countervailing power, the mid-century economists called it.²⁵ It is the arithmetic of the arch, approached from the workers’ side: the same climb toward the same peak, sought now across the bargaining table rather than legislated from above.

None of this licenses any wage floor whatever. The monopsony case is for a floor set between the depressed wage and the competitive one; lift it far above what labor would fetch in a market that genuinely cleared, and the old arithmetic returns to collect — the marginal worker is priced out after all. The loud disputes, over a wage of fifteen dollars or twenty, are arguments about where that line lies, and whether a given market sits beneath it or above. The honest account is neither the textbook’s nor its mirror image: the employer’s power over the wage is real, common, and larger than the clean model ever allowed — and it is also bounded, and the size of it is the entire question.

Notice, then, what the whole argument turns on and never sees. The competitive wage — the benchmark against which every markdown is measured and every floor judged too low or too high — is not a fact anyone can observe. No market hands it over. It is an inference, reconstructed from elasticities and natural experiments and the shapes of wage distributions, and reconstructed differently by careful people of good faith. Both sides of the loud disputes are making claims about the same invisible quantity: one says the floor still sits below it, the other that the floor has climbed above. This is why the question does not close, and will not. It is not, at bottom, an argument about what is happening to employment. It is an argument about an unobservable — and arguments about unobservables are settled, if they settle at all, not by looking but by the slow accumulation of method, and in the meantime by whoever holds the power to decide. The unobservability cuts at the policy as sharply as at the debate: a regulator who would set the floor at the very crest of the arch is aiming at a peak no instrument locates, so even the friendliest case for a minimum wage is a case for raising it in steps small enough to read the market’s answer before the next — not for certainty about where the peak lies.

What monopsony offered, and why it was received as heresy and defended as vindication, is the rarest thing the discipline has to give: a single policy that might be just and efficient at once. Almost everything else in economics forces the choice — you may redistribute, but at a cost to the size of the pie; you may grow the pie, but the slices fall where they fall. A minimum wage in a monopsonized market promises both at one stroke, lifting the low-paid and, by stripping the markdown, moving the market toward the allocation a competitive one would have reached: more efficient and more equal together. In that setting the minimum wage stops being a transfer and becomes a correction — nearer in kind to breaking a cartel than to taxing the comfortable to pay the poor — which is how it slips the iron law that redistribution must cost efficiency. That is precisely the trade-off the century’s arithmetic insisted was iron — and the deepest thing the theory does is suggest that in this corner of the world the trade-off was never iron, because the market was never competitive to begin with. The competitive market, here, was the special case; and a policy judged against a benchmark the market never met was bound to be misjudged.

Which returns the argument, finally, to the desk where it is actually lived. Behind every regular-rate computation, every meal-period premium, every overtime multiplier a wage-and-hour practice will ever litigate, sits a single number — the wage itself — that the technical work takes as given and builds upon. Monopsony’s claim is that this number was never simply given; it is, in part, the recorded outcome of a contest of power, a settlement between a worker who could not easily leave and an employer who knew it. And where that contest is fought directly — in the wage-fixing and no-poach class actions that became the civil sequel to the failed prosecutions, in the merger reviews that now count harm to workers as harm worth blocking — the case turns on an expert’s reconstruction of the very benchmark this plate has called unobservable: the wage that would have ruled had the power been absent.²⁶ The competitive wage is not a philosopher’s abstraction. It is a figure two economists are paid to dispute, to the dollar, before a jury. The whole delicate apparatus of wage-and-hour law — its premiums and multipliers and exemptions, computed to the cent — sits downstream of that prior, unsettled question and inherits its uncertainty without ever naming it. To compute the premium exactly is necessary work; but the figure it is computed on was always, and remains, contested ground — and to argue the small questions well is, whether one names it or not, already to have taken a position on the large one.