Work
How Your Wage Is Decided
Introduction
Most people sell their time for money. How that price gets set is a harder question than it looks. Ask people why they earn what they earn and the answers arrive in two flavors. The first is moral: I work hard, I studied for this, I earned it. The second is fatalistic: that is just what the job pays. Both are ways of not looking at the mechanism. The first treats a price as a verdict on your character. The second treats it as weather.
The mechanism is neither. Your wage sits somewhere inside a band. The top of the band is the most your employer could pay you and still be better off employing you than not. The bottom is the least you would accept before walking away to your next best option. Those two numbers are almost never close together, and the gap between them is usually large enough to hold several different lives. Where inside that band you actually land is not decided by effort or merit. It is decided by bargaining power: who has better options, who has better information, and who can more easily walk away.
This page is about that band - what sets its ceiling, what sets its floor, and what moves you within it. It is worth saying plainly at the start that this is not a cynical account. Productivity matters, skill matters, and the enormous differences in living standards between countries are real and mostly not about bargaining at all. But between two people doing the same job with the same competence, the pay gap is usually a bargaining story, and pretending otherwise leaves you unable to explain most of what you observe at work.
The Band, Not the Number
Start with the ceiling. An employer hires you because you are expected to add more to the business than you cost. Call that added value your contribution. No rational employer pays more than that for long, because doing so means losing money on every hour you work. This is the useful half of what economists call marginal product: it explains why nobody in a failing corner shop earns an investment banker's salary regardless of how brilliant they are. The ceiling on your pay is set by the value of the system you are plugged into, not by your personal qualities alone.
Now the floor. You will not work for less than your next best option, adjusted for how unpleasant the alternatives are. If you can walk across the street to an identical job at the same pay, your floor is high. If the nearest comparable employer is two hundred miles away, your visa is tied to your current employer, or your specific skill is useful to exactly three companies in the country, your floor collapses. Notice that none of these have anything to do with how good you are at the work.
Between those two lines there is room. Sometimes a great deal of room. A software engineer might generate several hundred thousand dollars of value a year for an employer while their realistic outside option pays $140,000. Any wage between those numbers leaves both parties better off than not trading. Both sides know this, which is why compensation conversations feel adversarial even when everyone is being pleasant: you are not discovering a correct number together, you are dividing a surplus. The single most useful thing to understand about your own pay is that the number is not a measurement. It is an outcome of a negotiation that started before you walked in the room.
Where Bargaining Power Comes From
Bargaining power is not a personality trait, though confident people often capture more of the band than timid ones. It is mostly a matter of position, and it comes from four places.
The first is outside options. How many employers could plausibly want you, how quickly, and at what pay. This is why the biggest raise in many careers comes from changing jobs rather than from performing well in the one you have: an offer letter converts an abstract claim about your value into a number your current employer has to beat. It is also why non-compete clauses, occupational licenses that do not travel across a border or between regions, and work permits tied to a single employer weaken your hand. Each one removes options without removing any of your ability. Employers defend non-competes as protection for trade secrets and training investments; after Oregon and Hawaii banned them for hourly and tech workers, pay rose 2-4%.
The second is information. Your employer usually knows the whole pay distribution of their organization and you usually know your own number and some rumors. That asymmetry is worth real money, which is why pay secrecy is so common and why pay-transparency laws have measurable effects. The evidence here has an uncomfortable wrinkle worth naming: transparency among coworkers reliably narrows gaps, but several studies find it does so partly by pulling high pay down rather than pulling low pay up, because an employer who must extend any raise to everyone bargains harder with each worker. Transparency about pay at other employers has tended to raise pay. Fairer and higher are not the same intervention.
The third is concentration on the employer side. Economists call a labor market with few buyers a monopsony, and it does to wages roughly what monopoly does to prices, in the opposite direction. José Azar, Ioana Marinescu, and Marshall Steinbaum, measuring the concentration of job postings in over 8,000 local occupational markets, found the average market highly concentrated by antitrust standards; posted wages were 5-17% lower at the 75th percentile of concentration than at the 25th. Kevin Rinz finds local concentration falling on average since the 1970s, though, so it cannot explain slow wage growth nationally. A hospital system that has absorbed every hospital within an hour's drive is not being villainous when nurses' pay stagnates. It simply no longer has to compete for them.
The fourth is coordination on the worker side. A single worker threatening to leave is an inconvenience. Every worker threatening to leave at once is an existential problem, which is the core mechanism behind unions, professional associations, and licensing bodies. This is why union membership correlates with higher pay for members, and also why the same institutions can raise the wages of insiders by restricting entry to outsiders. Both effects are real. How much of the premium comes from offsetting employer power and how much from shutting outsiders out is an open empirical question that depends heavily on the sector.
Why "You Are Paid What You Are Worth" Is Half True
The textbook claim is that in a competitive market, workers are paid their marginal product - the value of what one additional worker adds. As a statement about large aggregates over long periods it holds up reasonably well. Across countries and across decades, average compensation and average productivity track each other closely enough that no serious account of wages can ignore the connection.
As a statement about you, it is close to useless, for a reason that is usually skipped. Marginal product is almost never observable for an individual. Modern production is joint: an engineer's output depends on the codebase, the platform, the sales team, and a hundred people who built the tooling. There is no experiment that isolates your contribution, which means the number cannot be measured, which means it cannot be paid. What actually happens is that organizations construct a proxy - a level, a band, a title, a market survey - and then negotiate within it. The proxy is anchored to productivity in the way a map is anchored to terrain, and it inherits every bias of whoever drew it.
This explains something that otherwise looks like a paradox. Wage differences within an occupation, at the same firm, in the same city, are large and persistent - and they track measurable differences in output only loosely. The same work, done equally well, is paid differently depending on when you were hired, what you asked for, what you earned previously, and how badly they needed someone that month. None of that is measurement error. It is what happens when a negotiated number is dressed as an assessed one.
The Largest Wage Fact in the World
Everything above concerns the spread between people doing similar work in the same place. Zoom out and it is dwarfed by something else. The same person, with the same skills, the same work ethic, and the same years of schooling, earns radically different amounts depending on which country they are standing in. Michael Clemens and colleagues compared workers matched on country of birth, schooling, experience, and sex: for the median of 42 developing countries, a worker in the United States earns about four times as much as an observably identical worker at home, after adjusting for living costs. The ratio stays large even allowing for unusually able movers, and the authors find it far larger than pay gaps by race or sex within a country.
The mechanism is that much of productivity is not carried in your head. It is carried in the surrounding system: capital, machinery, reliable electricity, functioning courts, deep supplier networks, the ability to enforce a contract. A skilled welder in a country with no reliable power grid and no bankable customers produces less value per hour than the same welder in a shipyard with both, and no amount of personal excellence closes that gap. How the gap between rich and poor countries splits between place and skill is still argued. Tracking migrants' wages before and after they moved, Lutz Hendricks and Todd Schoellman put place at 40-50% of it. Even so, this is some of the strongest evidence that wages are a property of systems and not only of individuals, and it is the fact most often missing from conversations about who deserves what.
It also reframes a common intuition about immigration. If the same worker is several times more productive on one side of a border than the other, then the question of who is allowed to cross is not a minor administrative matter. It is one of the largest determinants of income in the world economy. That observation does not by itself settle what any country's policy should be. Immigration's effects on existing workers' wages, on public budgets, and on social cohesion are serious questions, set out on the Migration page. But it does mean any account of why pay differs that stops at the national border is leaving out the biggest term in the equation.
What People Do All Day, and How That Changed
For nearly all of recorded history, the answer to "what is your job" was farming, and by a wide margin: estimates for five European countries around 1400-1500 put 57-76% of their workers in agriculture. The dissolution of that arrangement is the largest change in what human beings do with their waking hours that has ever happened, and it is still under way. The world crossed the point where fewer than half of workers were in agriculture within living memory.
The American line is the one worth sitting with. About 1.5% of US workers are in agriculture, and the United States remains one of the world's largest food exporters. Agricultural employment collapsed while agricultural output rose. That combination is the central lesson of the whole transition, and it is the answer to a question people ask about every new technology: the jobs left and the sector did not. Something similar happened to manufacturing employment in every rich country, though there trade played a part too: in one prominent estimate, Chinese imports explain a quarter of the US factory job losses of 1990-2007. Still, "manufacturing employment fell" and "manufacturing fell" are different claims, and policy arguments often blur them.
Where did those workers go? Overwhelmingly into services - and in many developing countries, largely straight from farm to service work, with factory jobs peaking earlier and at far lower incomes than they did in today's rich countries. That matters for wages, because services are where measured productivity growth is slowest and hardest to define. It is straightforward to double the output of a loom. It is not obvious what it would even mean to double the output of a nurse, a teacher, or a barber without changing the thing being sold.
What the Number on the Offer Letter Hides
A wage is a price for a bundle, and the bundle contains far more than hours. Two jobs paying identically can differ enormously in what they actually cost the person doing them: night shifts, physical danger, the risk that the whole industry disappears, the requirement to live somewhere specific, whether you can be reached at nine on a Sunday. Economists call the pay differences that compensate for these compensating differentials, and the theory predicts that unpleasant jobs must pay more to attract anyone.
The theory is elegant and the world only partly obeys it. In practice, many of the most dangerous, dirtiest, and least autonomous jobs pay the least, not the most. The theory only works when workers have real choices; where the floor of the band has collapsed, unpleasantness gets absorbed rather than compensated. So the observed pattern is roughly this: within a labor market where people do have options, compensating differentials show up. Where they do not, the differential runs backward, and bad conditions and low pay arrive together. That inversion is a common sign of a labor market that is not competitive, though part of it is sorting: workers whose skills command high pay can also insist on better conditions, so counting conditions tends to widen measured inequality.
The bundle also increasingly contains things that never appear as pay. Employer-provided health coverage in the United States is the largest example: it is a substantial part of total compensation, it is largely invisible on a pay stub, and it can reduce mobility, because leaving the job means leaving the insurance. Any benefit that you lose by moving is, from the employer's side, a device that lowers your floor. That is not usually the intention. It is reliably the effect.
The Shrinking Slice
Of everything an economy produces, some share goes to people who work and some goes to people who own. For most of the postwar period, economists treated the split as one of the few stable numbers in the discipline - close to a constant of nature. Since roughly 1980 the measured labor share has declined in the United States and many other economies, rich and developing. The decline is not enormous in percentage-point terms, but at the scale of a national economy a few points is an extraordinary amount of money moving from one group to another.
Why it happened is properly contested, and the candidate explanations imply very different responses. One holds that capital goods, especially information technology, got cheap enough that firms substituted machines for people wherever the two were interchangeable. A second holds that a handful of superstar firms with very low labor shares came to dominate their industries, so the aggregate fell without the typical firm changing much - this is the account associated with David Autor and co-authors. A third, argued at length by Anna Stansbury and Lawrence Summers, points at declining worker bargaining power directly: weaker unions, more concentrated employers, offshoring as a credible threat. A fourth says much of the decline is housing and measurement. Matthew Rognlie's critique of Thomas Piketty's account finds that the long-run rise in the net capital share came entirely from housing. Germán Gutiérrez and Sophie Piton, correcting for how countries count dwellings and the self-employed, find the non-housing labor share stable or rising in every major rich economy except the United States and Canada.
The current reading is that several of these are contributing and that the weights are unsettled. What is not in dispute, for the United States at least, is the direction, and the direction is what people feel. Since 1973 median compensation has grown much more slowly than output per hour, and since 2000 average compensation has lagged too. A worker who feels that pay stopped keeping up is not misremembering; economists argue over how much of the gap is a shift to capital and how much is widening inequality among workers.
Credentials: Building Skill or Sending a Signal
People with more education earn more. Almost nobody disputes the correlation; the argument is about what produces it. The human-capital account, associated with Gary Becker, says education makes you more productive, and employers pay for the added productivity. The signaling account, associated with Michael Spence, says education mostly sorts: finishing a hard degree demonstrates traits you already had - persistence, conscientiousness, raw ability - and the certificate is valuable because it is expensive to fake, not because of what you learned.
Both mechanisms are real, and separating them empirically is hard. People who complete more education differ from those who do not in ways that also affect earnings. The cleanest evidence comes from natural experiments: changes in compulsory schooling laws, lottery admissions, students who just cleared a cutoff against those who just missed. These generally find real returns to additional schooling, which is a point for human capital. But the sheepskin effect cuts the other way. Finishing the final year of a degree raises earnings far more than the preceding year did, which is hard to explain if the value is the learning rather than the certificate. Yet one of the cleanest tests of the certificate alone, comparing students who barely passed and barely failed a high-school exit exam, finds little earnings gain from the diploma itself.
The practical consequence of the signaling component is credential inflation. If a degree is partly a sorting device, then as more people obtain one, its sorting power falls, and the requirement ratchets upward - jobs that once needed a high-school diploma now ask for a bachelor's degree without the work having changed. That is a treadmill: individually rational to run on, collectively wasteful, and very difficult to stop, because the first employer to drop the requirement bears the cost of sorting candidates themselves.
Machines and the Lump of Labor
The oldest error in this subject is the assumption that there is a fixed quantity of work to be done. If that were true, any task a machine takes would be a job a person loses. Economists call it the lump-of-labor fallacy, and two centuries of evidence say the name is earned. Mechanized agriculture did not produce permanent mass unemployment. It produced everyone who is not a farmer. Automation raises output per worker, which lowers prices, which raises real incomes, which creates demand for things that did not previously exist.
That is the correct long-run answer and it is also frequently used to wave away a real problem. The aggregate has always recovered; specific people frequently have not. Daron Acemoglu and Pascual Restrepo's work separates the displacement effect from the reinstatement effect - automation removes tasks from workers, and new tasks are created that workers do - and finds that the balance between them is not fixed by nature. It depends on what the technology is good at and on what firms are given tax and regulatory reasons to do. Their argument, with Andrea Manera, that the US tax code favors capital over labor and so tilts firms toward automating tasks that people did perfectly well is contested but well specified.
A more useful framing of the current wave is that the unit of displacement is the task, not the job, and almost every job is a bundle of tasks. When a technology takes some of the bundle, the effect on the person holding the job depends on whether their remaining tasks become more valuable or less. Historically, machines took the physically demanding and routine parts and made human judgment more valuable. What is new about the present wave is that it is aimed at exactly the judgment-and-language work that previously absorbed displaced labor. That does not license a confident prediction of mass unemployment; the evidence so far points both ways. Danish administrative data show no measurable effect of AI chatbots on earnings or hours two years after ChatGPT's launch. In US payroll data through June 2026, employment of 22-to-25-year-olds in the most AI-exposed occupations was 19% below where it would be had it kept pace with less-exposed work, mostly through fewer hires; the authors call the pattern descriptive, not causal. The new target does mean the historical reassurance rests on a pattern whose mechanism may not hold this time, and saying so is not alarmism - it is reading the argument carefully enough to notice what it actually assumed.
What Follows From This
If wages are a band rather than a number, several things that look like personal failures turn out to be features of the system, and a few things that look fixed turn out to be within reach.
The reachable part is your floor. Much of what reliably raises pay works by widening your outside options rather than by improving your performance. That means acquiring a skill that more than one industry wants, building a reputation that travels outside your employer, keeping the ability to move, and finding out what the market actually pays rather than what your last employer decided. The uncomfortable corollary is that loyalty is usually unpriced. An employer who knows you will not leave has no mechanism that forces them to pay you as though you might.
The structural part is the ceiling, and it is mostly not personal. If the surrounding system produces little value per hour, no individual excellence will fix that, which is why the same effort is rewarded so differently in different countries, industries, and firms. Choosing what system to plug into generally does more for lifetime earnings than any amount of optimizing inside a system that cannot pay. That sounds like advice, but it is really a statement about where the causation runs.
The question this page opened with has a short answer. Your wage was not measured, it was negotiated, inside a band whose ceiling belongs to the system you work in and whose floor belongs to the options you have. Reading a pay number as a verdict on your worth misreads what kind of thing it is. It is a price, and prices are about scarcity and alternatives, not about merit. Which raises the obvious next question: if so much of what determines pay happens inside organizations rather than in open markets, why do organizations exist at all?
