Do minimum wages cause unemployment?
The textbook says yes. The data says maybe not. A 30-year war between theory and evidence.
The $7.25 question
AOC holds up a $7 croissant in a congressional hearing to make a point: the federal minimum wage is $7.25. You can barely buy a single pastry for an hour's labor. But is $15 the right fix — or just the loudest one?
AOC's argument is visceral and real: a single croissant costs $7, and the minimum wage is $7.25 — an hour of labor barely covers a pastry. But the economics behind setting the "right" number is one of the most fought-over questions in the discipline. To see why, you need the cleanest prediction in introductory economics — and you need to understand why it might be wrong.
Supply and demand in labor markets. A labor market works like any other market, except the buyers and sellers are reversed. Firms are the buyers — they demand labor. Workers are the sellers — they supply it. The demand curve slopes downward: the higher the wage, the fewer workers a firm wants to hire, because each additional worker must produce enough revenue to justify the cost. The supply curve slopes upward: higher wages attract more people into the workforce. Where these curves cross, you get the equilibrium wage and employment level.
The price floor prediction. A minimum wage is a price floor on labor. If the government mandates a wage above equilibrium, two things happen simultaneously: firms want fewer workers (moving up the demand curve) and more people want jobs (moving up the supply curve). The gap is a surplus — and in a labor market, a surplus of workers has a name: unemployment.
If the minimum wage $w_{\min}$ exceeds the equilibrium wage $w^*$, then labor demanded $L^d(w_{\min}) < L^d(w^*)$ and the difference $L^s(w_{\min}) - L^d(w_{\min})$ is involuntary unemployment. The size depends on the elasticity of labor demand: elastic demand (firms can easily substitute machines for workers) means a large employment drop; inelastic demand (the work requires humans) means a small one. But the direction is supposed to be unambiguous.
The size of the unemployment effect depends on how easily firms can replace workers. If a robot can do the job, even a small wage increase pushes firms toward automation. If the work genuinely requires human hands, the effect is small. But the textbook says the direction is always the same: more expensive labor means less labor hired.
A concrete example. Consider a small-town diner that pays its three dishwashers $8 per hour at the market equilibrium. The state raises the minimum wage to $12. The diner operates in a competitive market — it can't raise menu prices much without losing customers to the diner down the road. So the owner does the math: three dishwashers at $12 is $36 per hour; two dishwashers at $12 is $24 per hour. She lets one go and makes the remaining two work harder. Clean, simple, and — according to a generation of textbooks — the end of the story.
For most of the 20th century, this prediction dominated policy advice. A famous 1978 survey summarized the consensus: a 10% increase in the minimum wage reduces teenage employment by 1–3%. The debate was about the size of the number, not the sign.
"A $7.25 minimum wage is too low to buy a croissant in the building I work in. It costs $7 for a croissant. The American people are not being unreasonable when they say this is not enough."
— Alexandria Ocasio-Cortez, Congressional hearing, 2019
"Fight for $15!"
The most successful wage campaign in a generation turned a number into a movement. But $15 means very different things in San Francisco and rural Mississippi. The economics of "how much" turns out to be inseparable from "where."
The textbook vs. the real world
"A 10 percent increase in the minimum wage reduces teenage employment by 1 to 3 percent. This is one of the best-established empirical regularities in economics."
— Charles Brown, Curtis Gilroy & Andrew Kohen, Journal of Economic Literature, 1982
For decades, this was the consensus estimate. Every labor textbook cited it. Policy advisors repeated it. The logic was clean: price floors cause surpluses, and labor is no exception. The debate was about magnitude, not direction. Then someone decided to actually run the experiment.
"No business which depends for existence on paying less than living wages to its workers has any right to continue in this country."
— Franklin D. Roosevelt, 1933
Roosevelt's framing sidesteps the employment question entirely. If the goal is that full-time work shouldn't produce poverty, then employment effects are a cost to weigh, not a reason to abandon the policy. This is the moral backbone of the minimum wage movement — and it exposes the gap between "what does the model predict?" and "what should we do about it?" The economics can tell you the tradeoffs. It can't tell you which tradeoffs are acceptable.
Where this leaves us
At this level, the competitive model gives a powerful, clean prediction: binding minimum wages cause unemployment. The logic is the same as any price floor creating a surplus. For most of the 20th century, economists treated this as settled science. The open question — the one that would blow this consensus apart — is whether the model's key assumption actually holds: are low-wage labor markets competitive?
Theory says yes: unemployment is the inevitable cost of a price floor. But what if the labor market doesn't work the way the model assumes? What if employers have more power than the textbook gives them — and the minimum wage is correcting a failure, not fighting a market?
The Card-Krueger revolution
"The minimum wage debate is really a debate about market power. If the labor market were perfectly competitive, the textbook would be right. It isn't."
— Alan Manning, Monopsony in Motion, 2003
This insight flipped the textbook prediction. Manning showed you don't need a company town for employers to have wage-setting power — you just need the real world.
Manning's claim is radical in its simplicity. The textbook prediction — minimum wages cause unemployment — rests on one assumption: that many firms compete for workers, driving wages to the competitive level. If that assumption fails, the prediction flips. And Manning argued it fails almost everywhere in low-wage labor markets.
Classical monopsony. Joan Robinson formalized this in 1933. A monopsonist is the mirror image of a monopolist: instead of a single seller restricting output to raise prices, it's a single buyer restricting purchases to lower prices. A company town with one employer is the textbook example. The employer faces an upward-sloping labor supply curve: to hire one more worker, it must raise wages not just for the new hire but for all existing workers. This makes each additional hire more expensive than the wage alone suggests.
The monopsonist's marginal cost of labor exceeds the wage: $MCL = w + L \cdot \frac{dw}{dL}$. The firm hires where $MCL = MRP$ (marginal revenue product), which means fewer workers at a lower wage than the competitive outcome. Employment is $L_m < L^*$ and the wage is $w_m < w^*$.
Now introduce a minimum wage $w_{\min}$ between $w_m$ and $w^*$. The firm's labor supply curve becomes flat at $w_{\min}$ up to the supply at that wage. The marginal cost of labor drops to $w_{\min}$ for additional hires, because the firm no longer needs to raise all workers' wages to attract one more. Employment increases. The minimum wage has corrected the monopsony distortion.
Here's the key insight: if the employer is already restricting employment below the competitive level — hiring fewer workers to keep wages down — a minimum wage can actually increase employment. It eliminates the employer's incentive to keep hiring artificially low. The wage floor says "you can't hold wages down by restricting jobs anymore," and the employer responds by hiring more workers at the mandated wage. The minimum wage isn't fighting the market — it's correcting a market failure.
The "new monopsony." Before Manning, economists dismissed monopsony as a curiosity — relevant to coal towns in 1920, not modern cities with hundreds of employers. Manning's revolution was to show that you don't need a single employer. You need frictions. Search costs (it takes time and money to find a new job). Moving costs (workers can't easily relocate). Information asymmetries (workers don't know all available wages). Simple inertia (quitting is scary). All of these give employers wage-setting power, even in a city with a thousand firms.
The evidence for labor market power is now extensive. Azar, Marinescu, and Steinbaum (2022) found that labor market concentration — measured by how few employers dominate hiring in a given occupation and commuting zone — is associated with wages 5–17% below competitive levels. Dube, Jacobs, Naidu, and Suri (2020) ran experiments on Amazon Mechanical Turk and found that even in an online labor market with near-zero search costs, employers retain significant wage-setting power. The competitive labor market is the exception, not the rule.
"If you're an employer and you're paying someone $7.25, you're not competing for workers -- you're exploiting the fact that they have no better options."
— Arindrajit Dube, The Hamilton Project, 2014
The $15 debate with monopsony glasses on
If low-wage labor markets are monopsonistic, the Fight for $15 isn't fighting the market -- it's correcting a market failure. The textbook unemployment prediction assumes competition that may not exist.
How much monopsony power is there?
"Labor markets are pervasively monopsonistic. The perfectly competitive labor market of the textbook, where wages are determined by the intersection of supply and demand, is a theoretical construct with little empirical relevance for low-wage workers."
— Alan Manning, Monopsony in Motion, 2003
Manning's evidence was structural: he showed that labor supply to individual firms is far less elastic than labor supply to the market as a whole. A firm can cut wages by 10% and lose only a fraction of its workers, because search frictions, moving costs, and inertia keep people in place. That gap between firm-level and market-level elasticity is monopsony power. It doesn't require a company town. It requires the real world.
"The monopsony model has been used to rationalize any minimum wage result. Find a positive employment effect? Monopsony. Find zero? Monopsony. The model has become unfalsifiable, which means it explains nothing."
— Summary of Neumark & Wascher critique
This is a legitimate methodological concern. If monopsony can explain positive, zero, and negative employment effects depending on the size of the increase relative to the monopsony gap, it becomes hard to test. The response from Manning's camp: monopsony is testable — it predicts that employment effects should vary with labor market concentration, and Azar et al. confirmed exactly this pattern. Markets with more employer concentration show smaller (or positive) employment effects; competitive markets show the traditional negative effects. The theory makes specific, falsifiable predictions. The critics just haven't engaged with the right ones.
Where this leaves us
Monopsony theory gives us a specific, testable prediction: if employers have wage-setting power, moderate minimum wages can increase employment. The empirical evidence for widespread labor market power is now strong. The question shifts from "do minimum wages cause unemployment?" to "how much monopsony power exists in which markets, and at what wage level does the monopsony buffer run out?" That's an empirical question. And in 1994, two economists ran the most famous experiment in labor economics to answer it.
Theory now gives us two opposite predictions. Competitive markets: minimum wages cause unemployment. Monopsonistic markets: they might not. A beautiful theoretical impasse. Only data can break it. And the data came from an unlikely source: fast-food restaurants on opposite sides of the Delaware River.
The new consensus
"We found that the increase in New Jersey's minimum wage led to an increase in employment at fast food restaurants — the opposite of what the textbook predicts."
— David Card & Alan Krueger, 1994
The most famous natural experiment in economics. Card and Krueger compared fast-food restaurants across a state border and got a result that shook the profession to its foundations. Card won the Nobel Prize in 2021 partly for this work.
In April 1992, New Jersey raised its minimum wage from $4.25 to $5.05 per hour. Neighboring Pennsylvania didn't. Card and Krueger saw something no one had thought to exploit before: a natural experiment. Compare employment changes in fast-food restaurants on both sides of the state border — same regional economy, same consumer base, one policy difference.
Difference-in-differences. The method is elegant in its simplicity. Don't just compare NJ employment to PA employment — that could reflect pre-existing differences. Instead, compare the change in NJ employment to the change in PA employment. The comparison of changes cancels out anything that was different between the two states before the policy change, as long as those differences stayed constant.
The difference-in-differences estimator:
$$\hat{\delta} = (\bar{Y}_{NJ,\text{after}} - \bar{Y}_{NJ,\text{before}}) - (\bar{Y}_{PA,\text{after}} - \bar{Y}_{PA,\text{before}})$$Card and Krueger surveyed over 400 fast-food restaurants before and after the wage increase. Their $\hat{\delta}$ was positive: NJ employment rose relative to PA. The sign of the textbook prediction was wrong.
Card and Krueger surveyed over 400 fast-food restaurants in NJ and eastern PA before and after the wage increase. Their finding: employment at NJ restaurants did not fall relative to PA restaurants. If anything, it rose slightly. The textbook prediction appeared to be wrong. Not a little wrong — the sign was wrong.
Not a one-off. In the three decades since, the result has been replicated dozens of times. Dube, Lester, and Reich (2010) compared all adjacent counties straddling a state border where minimum wages differ — hundreds of county pairs across the entire United States. Result: no significant negative employment effects. The most comprehensive study to date, Cengiz, Dube, Lindner, and Zipperer (2019), examined every state and federal minimum wage increase in the U.S. from 1979 to 2016 — 138 events in total. Their innovation was a "bunching estimator": look at the entire wage distribution. Did jobs paying below the new minimum disappear? Yes. Did jobs paying at or just above the new minimum appear to replace them? Yes, in roughly equal numbers. Workers weren't losing jobs — they were getting raises.
But the debate isn't over. Neumark and Wascher (2007) argued that using payroll data instead of survey data shows employment did decline. Jardim et al. (2022) found that Seattle's aggressive $15 minimum wage reduced total hours worked even though headcount was stable — firms adjusted on the intensive margin (fewer hours per worker) rather than the extensive margin (fewer workers). Clemens and Wither (2019) found larger disemployment effects during the Great Recession, suggesting the economic context matters.
"A minimum wage is not a living wage. The question isn't whether the minimum wage causes unemployment -- it's whether it lifts workers out of poverty."
— David Card, Nobel Prize lecture, 2021
Minimum wage vs. living wage
Even if moderate minimum wage increases don't cause unemployment, do they actually reduce poverty? The minimum wage is a blunt instrument -- many minimum wage workers aren't in poor households (think teenagers with well-off parents). The EITC targets low-income families more precisely. Are we fighting about the wrong policy?
What does the evidence actually show?
"Across 138 prominent state-level minimum wage events between 1979 and 2016, we find that the number of low-wage jobs remained essentially unchanged. The missing jobs from below the new minimum are accounted for by the excess jobs above it."
— Doruk Cengiz, Arindrajit Dube, Attila Lindner & Ben Zipperer, Quarterly Journal of Economics, 2019
The Cengiz et al. bunching estimator is the strongest piece of evidence in the modern debate. Instead of comparing treated and untreated groups (which depends on finding a good control), they look at the shape of the wage distribution before and after a minimum wage increase. The "missing" jobs below the new minimum show up as "excess" jobs just above it. Net employment effect: approximately zero for moderate increases. This isn't one study — it's 138 natural experiments analyzed with a single, transparent method.
"The number of hours worked by low-wage workers fell by about 9 percent, while the number of low-wage jobs fell by only 1 percent. Workers kept their jobs but got fewer shifts."
— Summary of Jardim et al. findings on Seattle's $15 minimum wage, 2022
This is the most sophisticated critique of the "no employment effect" consensus. Jardim et al. studied Seattle's phased increase to $15 and found the employment effect hiding on the intensive margin: not fewer jobs, but fewer hours. Workers kept their positions but worked less. Total earnings for low-wage workers may have fallen even as the hourly wage rose. The CBO's 2019 estimate of a $15 federal minimum projected 1.3 million jobs lost but 17 million workers getting raises — a massive redistribution with winners and losers. The employment effects are real, but they show up in ways that simple headcount measures miss.
The verdict
The evidence has a clear message and a clear boundary. Moderate minimum wage increases — up to about 60% of the local median wage — have small or zero employment effects. This is one of the most replicated findings in modern labor economics, confirmed across dozens of studies, multiple countries, and three decades. The competitive model isn't wrong in general; its key assumption (competitive labor markets) just doesn't hold for many low-wage markets where monopsony power is real, widespread, and empirically important.
But this is not a blank check. Large increases — especially in low-wage regions where the minimum wage would far exceed the local median — could overwhelm the monopsony buffer and produce exactly the unemployment the textbook predicts. The Jardim et al. finding about hours reduction suggests effects may show up in subtler forms than simple job loss. And the long-run automation response remains largely unmeasured. The policy-relevant question isn't yes or no. It's how much, where, and measured how.
Where this leaves us
We started with AOC on the House floor, telling 30 million viewers that $7.25 is a starvation wage. Three stages later, here's what you now know:
- The textbook prediction is clean but conditional (Stage 1). In a competitive labor market, a binding minimum wage causes unemployment. The logic is identical to any price floor creating a surplus. For most of the 20th century, economists treated this as settled. It wasn't.
- Monopsony changes everything (Stage 2). If employers have wage-setting power — and the evidence says they do, pervasively, in low-wage labor markets — a moderate minimum wage can increase employment by correcting the monopsony distortion. Manning's "new monopsony" showed that search frictions, moving costs, and information asymmetries create employer power without requiring a company town.
- The data broke the tie (Stage 3). Card and Krueger's 1994 natural experiment, replicated dozens of times across countries and decades, showed that moderate minimum wage increases have small or zero employment effects. The Cengiz et al. bunching estimator — 138 events, one transparent method — is the most comprehensive confirmation. The empirical threshold is roughly 60% of local median wage. Below it, the monopsony buffer absorbs the increase. Above it, you're extrapolating beyond the evidence.
The next time someone tells you "minimum wages kill jobs" or "just raise it to $15," you have the tools to evaluate both claims. The first is the textbook prediction for competitive markets that don't describe most low-wage labor. The second ignores that $15 is moderate in Seattle and extreme in Mississippi. The right answer is a formula, not a slogan: index to local median wages, stay within the empirically tested range, and accept that the tradeoff between higher wages and potential job loss is real even if the consensus says it's small for moderate increases.
AOC was right that $7.25 is indefensible. The textbook was right that price floors have costs. The empirical revolution showed those costs are smaller than anyone expected — but not zero, and not uniform. That's not a cop-out. That's what it means to actually understand the question.