Labor: Clark through the credibility revolution
A century from “wages equal what you produce” to a fast-food counter on the New Jersey side of the Delaware that proved the textbook wrong. This is the one thread no era-organized book hands you whole.
The competitive baseline
“If you raise the price of something, people buy less of it. Labor is no different. Mandate a wage above what a worker produces, and the job disappears. That’s not ideology — that’s Econ 101.”
— the standard supply-and-demand restatement of the competitive prediction, repeated in op-eds, hearings, and viral clips for a century
This argument feels like arithmetic. It descends, almost word for word, from a result a Columbia economist published in 1899 — and for most of the twentieth century the profession agreed it was settled. Where does it come from, and is it a theorem or a slogan?
The prediction is the surface of an apparatus. In The Distribution of Wealth (1899), John Bates Clark gave the first rigorous answer to a question classical economics had fumbled: why does a factor of production earn what it earns? His answer was marginal productivity. A firm hires one more worker only if that worker adds at least as much to revenue as the wage costs — so in equilibrium the wage equals the value of the marginal product. Each factor is paid exactly what the last unit of it contributes. The labor market is then a market like any other: a downward-sloping demand curve (the marginal-product schedule), an upward-sloping supply curve, and a clearing wage.
From that equilibrium the policy prediction falls out cleanly. A minimum wage set above the market-clearing level is a price floor on labor. At the higher price, firms want fewer workers and more workers want jobs; the gap is unemployment. The disemployment result is not a separate claim bolted on — it is what the marginal-product equilibrium says the moment you bind the wage above $w^*$.
The firm hires until the wage equals the value of labor’s marginal product:
$$w = MRP_L = P \cdot MP_L$$A binding floor $w_{\min} > w^*$ moves the firm back up its labor-demand curve, so quantity demanded falls: $L^d(w_{\min}) < L^d(w^*)$. The wedge is the predicted job loss.
A diner keeps a dishwasher on only as long as the dishes washed are worth more than the wage. Force the wage up past that point and the rational move is to let the dishwasher go — one fewer job. Scale that logic across every low-wage employer and you get the textbook’s confident sign: floors destroy jobs.
That is the founding rung of the entire thread, and everything after it is a response to it. The formal labor-demand apparatus — the marginal-product schedule, the competitive wage, the price-floor diagram — is worked out in Labor Economics §21.2 (Labor Demand). The intellectual lineage — Clark’s marginal-productivity step inside the wider marginalist break with the labor theory of value — is the spine of History of Economic Thought Ch.5 §5.1 (Three revolutions at once).
Clark at full strength
It is tempting to treat the competitive model as a strawman the modern field knocked down. That would be a misreading, and the whole thread depends on resisting it. Clark’s marginal-productivity theory was a genuine intellectual achievement. Before it, the classical economists had no coherent account of wages — Ricardo and Marx leaned on a labor theory of value that could not say why a surgeon earns more than a porter without circular reasoning. Clark dissolved the puzzle: a factor earns the value of what its marginal unit adds, no more and no less. The result is a theorem, not a conjecture. Under its assumptions — competitive output and labor markets, homogeneous labor, costless mobility — it is simply correct.
“The studies of professional economists … reduce to the conclusion that a 10 percent increase in the minimum wage reduces teenage employment by one to three percent … one of the more reliable findings in empirical economics.”
— Charles Brown, Curtis Gilroy & Andrew Kohen, Journal of Economic Literature, 1982 — the profession’s consensus restatement of Clark’s prediction
By 1982 the disemployment prediction was not a fringe slogan but the surveyed consensus of the field, stated with the confidence of an established empirical regularity. That is how strong the apparatus looked. And yet it carries one assumption that, once you stare at it, refuses to sit still: labor is homogeneous. Clark’s worker is a unit, scalable up and down. But the data are full of workers who are observably alike — same age, same industry, same region, same schooling — earning systematically different wages. A model in which identical units earn identical marginal products has no room for that. The anomaly is not a quibble; it is a hole at the center of the apparatus.
Where this rung leaves us
Marginal productivity gave labor economics its founding result and its cleanest prediction, and for most of the twentieth century the profession treated both as settled. The achievement is real and it survives — we will not be taking it back. But the homogeneous-labor assumption leaves an anomaly the equilibrium cannot absorb: why do similar workers earn different wages? Answering that required treating labor not as a uniform factor but as something workers actively invest in.
Clark’s worker is a unit of homogeneous labor earning its marginal product. But a surgeon and a janitor are not the same unit scaled up and down. What if the difference isn’t the worker — it’s what the worker has invested in?
Labor as investment
“College is a scam. Four years and six figures of debt for a piece of paper. Just learn a trade.”
— the perennial “is college worth it?” argument, restated every hiring season
Buried inside this argument is a model. Whether college “is worth it” is a question about the return on an investment — forgo earnings now, recover them later in higher wages. That framing did not exist in Clark’s apparatus. It was built in the 1960s to plug exactly the hole the competitive baseline left open.
The fix came from Chicago. Theodore Schultz, in his 1961 presidential address “Investment in Human Capital,” argued that education and training are not consumption but investment: they raise a worker’s productivity, and wages then embody the accumulated stock. Gary Becker’s Human Capital (1964) formalized the decision — a worker forgoes earnings today for higher earnings later, exactly like a firm choosing whether to buy a machine, and invests up to the point where the discounted return equals the cost. Jacob Mincer (1974) turned the idea into the workhorse equation of empirical labor economics, regressing log wages on schooling and experience.
The Mincer earnings equation:
$$\ln w = \beta_0 + \beta_1 S + \beta_2 X + \beta_3 X^2 + \varepsilon$$where $S$ is years of schooling and $X$ is labor-market experience. The positive $\beta_1$ is the return to a year of school; the concave experience term ($\beta_2 > 0$, $\beta_3 < 0$) captures on-the-job investment that tapers as a career matures.
Two identical eighteen-year-olds. One starts earning now; the other spends four years and real money on a degree, then earns more for the rest of a career. Human capital says the wage gap between them later is the return on that investment — and the same logic explains why earnings rise steeply early in a career, then flatten: you front-load the learning.
This is the rung Clark’s apparatus could not supply. Homogeneous labor has no place for a surgeon-versus-janitor gap; human capital makes the gap the visible return on accumulated investment, without abandoning the equilibrium discipline that wages still track productivity. The formal apparatus — the Mincer equation, the human-capital investment decision — is worked out in Labor Economics §21.3 (Human Capital and the Mincer Equation). The intellectual lineage runs through the Chicago school of the 1960s — Schultz, Becker, and Mincer reframing skill as capital — one of the most consequential exports of the postwar Chicago program.
Human capital at full strength
Make no mistake about where this rung sits in the thread: human capital is a retention, not a casualty. The Mincer equation is, to this day, the single most-estimated relationship in empirical labor economics — run on every household survey in every country, the first thing a labor economist reaches for to describe the wage distribution. The credibility revolution we are walking toward did not falsify it. It sharpened it. A careless version of this story implies the natural-experiment turn killed human capital; the careful version is that human capital was the apparatus the credibility revolution learned to estimate cleanly.
But human capital inherited Clark’s frictionless-competitive market wholesale, and with it two anomalies it cannot touch. First: a frictionless market clears, so it cannot explain why any involuntary unemployment exists in equilibrium — yet it plainly does, in every economy, in every year. Second: the same human capital should command the same wage everywhere, yet two workers with identical résumés routinely earn different wages at different firms across the street from each other. Explaining wage differences across skill levels is one thing; explaining wage differences between identical workers, and unemployment itself, is another. For that, the apparatus had to admit something both Clark and the Chicago school assumed away — that finding a job, and filling one, takes time.
Where this rung leaves us
Human capital solved Clark’s anomaly without breaking his equilibrium logic, and it survives intact as the workhorse of wage-variation analysis. This rung is a retention — the thread keeps it. But it carried over the frictionless-competitive assumption, and that assumption cannot produce equilibrium unemployment or explain why the same worker earns different wages at different firms. To get there, the apparatus had to take frictions seriously.
Human capital tells you why a surgeon earns more than a janitor. It can’t tell you why two janitors with identical résumés earn different wages at the diner across the street — or why either of them was ever unemployed at all. For that, you have to admit something the competitive model assumed away: finding a job takes time.
Frictions and market power
“Unemployment is not a sign that the market has failed to clear. It is what an equilibrium looks like when it takes time and effort for workers and jobs to find each other.”
— the framing recognized by the 2010 Nobel to Peter Diamond, Dale Mortensen & Christopher Pissarides
This sentence quietly dismantles the competitive baseline. In Clark’s world, a market that did not clear was a market with a problem. Diamond, Mortensen, and Pissarides showed that unemployment is a normal property of a labor market once you admit the obvious: matching workers to jobs is not instantaneous. And once you admit frictions, a second thing follows that the textbook never allowed — the employer, not the market, may be setting the wage.
Half one: search and matching. The Diamond-Mortensen-Pissarides apparatus replaces the frictionless market with a matching function — matches are produced from the pool of unemployed workers and the pool of vacancies, the way output is produced from labor and capital. At any moment some workers are searching and some jobs are unfilled; the steady-state level of that overlap is equilibrium unemployment. No sticky wages, no money illusion, no market failure required — frictions alone make unemployment a permanent feature of equilibrium, and the wage is split between worker and firm by bargaining over the surplus the match creates.
Half two: who captures the surplus. Frictions create room, and that room has an occupant. If quitting to find another job is costly and slow, a worker cannot costlessly flee a low-wage employer — so the employer faces an upward-sloping supply of labor to itself and can hold the wage below the competitive level. This is monopsony, formalized by Joan Robinson in 1933 and revived by Alan Manning’s Monopsony in Motion (2003), whose decisive move was to show you do not need a company town: ordinary search frictions hand every employer some wage-setting power. The single spine holds the rung together — frictions create the room; monopsony is who captures the surplus the frictions create.
Search and matching: matches are produced by a matching function over unemployed workers $U$ and vacancies $V$,
$$M = m(U, V)$$Monopsony: a wage-setting employer faces a rising supply curve, so the marginal cost of labor exceeds the wage,
$$MCL = w + L \cdot \frac{dw}{dL} > w$$and the firm hires fewer workers at a lower wage than the competitive benchmark — the mirror image of Clark’s diagram.
Two ideas, one mechanism. Because finding a job takes time, workers can’t all walk out the moment a wage is stingy — so some are always between jobs (that’s the unemployment), and the employer who knows you can’t leave easily can pay you less than you’re worth (that’s the market power). Friction is the cause; unemployment and underpayment are the two symptoms.
The formal apparatus for both halves lives in Labor Economics §21.4 (Search and Matching, DMP) and §21.5 (Monopsony). The lineage runs through the information-economics break with the perfect-information assumption — the move that made search theory possible — in History of Economic Thought Ch.11 §11.1, and the post-2008 market-power revival in Ch.17 §17.4 (the empirical turn).
The foundation, rebuilt
Take both halves at full strength. DMP is a Nobel apparatus that did something Keynes had only asserted — it made involuntary unemployment a clean equilibrium phenomenon, derivable from frictions alone, with no appeal to sticky wages or money. And the new-monopsony evidence is not a theoretical curiosity. Azar, Marinescu, and Steinbaum (2022) found that in concentrated local labor markets, wages run roughly five to seventeen percent below the competitive benchmark. Dube and co-authors (2020) ran wage-setting experiments on Amazon’s MTurk platform — about as close to a frictionless spot market for labor as exists — and still found employers with measurable power to set wages below the going rate. The competitive frictionless labor market, on this evidence, is the special case, not the norm.
Here is what makes this rung the pivot of the whole thread. With monopsony in the model, the sign of the minimum-wage effect flips: a wage floor set between the monopsonist’s wage and the competitive wage can actually raise employment, because it strips the employer of the power to suppress hiring. So theory now offers two opposite predictions — the competitive model says a floor destroys jobs, the monopsony model says it can create them — and no way to choose between them from the armchair. This is the apparatus that powers the live minimum-wage debate; for the full monopsony-mechanism treatment and the policy fight it drives, see the walkthrough Do minimum wages cause unemployment? The thread’s job here is narrower: to place monopsony where it belongs in the lineage — as the answer to the frictions anomaly DMP opened — not to re-litigate the policy magnitude.
Where this rung leaves us
Frictions and monopsony rebuilt the foundation rather than patching it. Unemployment became an equilibrium phenomenon; employer wage-setting power became the default rather than the curiosity; the competitive frictionless market became the special case. The apparatus now predicts that Clark’s disemployment result need not hold. But predicting that the competitive result might fail is not the same as demonstrating that it does. The theory had outrun the evidence — until two economists found a way to run the experiment.
Theory now offered two opposite predictions: competitive markets say minimum wages destroy jobs, monopsonistic markets say they might create them. An impasse like that can only be broken one way. Someone had to actually look at what happened when a wage floor went up — and the cleanest place to look turned out to be a fast-food counter on the New Jersey side of the Delaware River.
The experiment that broke the textbook
“to David Card for his empirical contributions to labour economics … and to Joshua Angrist and Guido Imbens for their methodological contributions to the analysis of causal relationships.”
— the 2021 Sveriges Riksbank Prize in Economic Sciences citation
Read that citation as the thread’s climax. The field’s founding prediction — the one descending straight from Clark — was overturned by the field’s own empirical turn, and the overturning was awarded the discipline’s highest honor. When you finally run the experiment, does the competitive prediction survive contact with the data?
The impasse at the end of Stage 3 was an empirical one, and it needed an empirical method. David Card and Alan Krueger (1994) supplied it. When New Jersey raised its minimum wage and neighboring Pennsylvania did not, they compared the change in fast-food employment on the New Jersey side to the change on the Pennsylvania side — differencing out anything constant across the border. The textbook predicted a fall. They found, if anything, a small rise. The competitive sign was wrong. In parallel, Joshua Angrist and Alan Krueger (1991) used quarter of birth interacting with compulsory-schooling laws as an instrument for years of schooling — identifying the return to education without the structural assumptions human-capital estimation had quietly leaned on. That second result closes a loop the thread has kept open: the credibility revolution did not falsify Stage 2’s human capital. It gave it a credible estimate.
The difference-in-differences estimator differences out fixed cross-border factors and the common time trend:
$$\hat{\delta} = \big(\bar{Y}_{NJ,\text{after}} - \bar{Y}_{NJ,\text{before}}\big) - \big(\bar{Y}_{PA,\text{after}} - \bar{Y}_{PA,\text{before}}\big)$$The instrumental-variable logic is the same discipline applied to a different problem: use a source of variation in schooling (birth quarter) that is plausibly unrelated to ability, so the estimated return is not contaminated by who chooses to stay in school.
The trick in both is to find a natural accident. New Jersey and Pennsylvania fast-food counters are nearly identical except for one law that changed on one side — so whatever else was going on cancels, and what’s left is the policy’s effect. Birth quarter is an accident of the calendar that nudges how much school the law forces on you — so it isolates the wage effect of schooling from the effect of being the kind of person who stays in school.
The formal apparatus — the diff-in-diff estimator, the instrumental-variable logic, the threats to validity that make these claims credible or not — is worked out in Econometrics Foundations §10.4 (Difference-in-Differences), §10.3 (Instrumental Variables), and §10.8 (Threats to Validity). The intellectual lineage — the credibility revolution as the empirical turn that landed first and hardest in labor — is the subject of History of Economic Thought Ch.12 §12.5 (Apparatus-origin: labor and the credibility revolution), with the broader post-2008 empirical turn in Ch.17 §17.4.
The climax, and the honest caveat
Card-Krueger was not one study; it was the opening of a method. The New Jersey result has been replicated dozens of times across three decades. Dube, Lester, and Reich (2010) ran the same logic across hundreds of contiguous US county pairs that straddle a state border, and found the same near-zero employment effect. Cengiz, Dube, Lindner, and Zipperer (2019) built a bunching estimator over 138 state-level minimum-wage events and showed that the jobs paying just below the old minimum reappear just above the new one — the wage floor moves workers up the wage distribution without destroying the jobs. This is the lineage’s climax: the apparatus this thread has traced from Clark reached the point where its founding competitive prediction could be cleanly tested, and the prediction failed.
Now the honest caveat, reported at full strength. There is a genuine disagreement here, and it deserves a hearing. David Neumark and William Wascher have argued for years that across the full body of evidence the preponderance still shows disemployment for the least-skilled. Jardim and co-authors found that Seattle’s move toward a $15 minimum cut hours even where headcount held — the effect showing up on the intensive margin rather than in the count of jobs. But notice precisely what this disagreement is. It is not a quarrel about the frame: both camps work inside the same frictions-and-monopsony-augmented apparatus, which can in principle produce zero, positive, or negative effects depending on how far the floor sits above the local market. It is a disagreement about the magnitude and sign at observed hike levels, and which estimator to trust — a parameter-magnitude split inside a shared frame, not a frame-level split. The full employment-effect-magnitude debate is owned, at depth, by the walkthrough Do minimum wages cause unemployment?; the thread reports it at lineage resolution and defers the magnitude fight rather than re-running it.
One boundary worth marking. Card-Krueger is the labor field’s climax, but it is also one instance of a cross-field revolution in method — the same natural-experiment logic remade development economics, education research, and the study of crime and health. How that causal-identification toolkit accumulated across fields is a separate thread (the econometric-methodology thread, forthcoming); here we trace only how it landed in labor and broke labor’s founding prediction.
A century later: where the thread lands
The competitive-marginal-productivity prediction that descends from Clark failed its strongest empirical test. The modern labor apparatus is frictions-and-monopsony-augmented and answerable to evidence in a way Clark’s never was — but it retains marginal productivity as the firm’s labor-demand foundation and human capital as the workhorse of wage variation. Both survive as partial frameworks that describe important cases. What did not survive is the universality of the naive competitive prediction.
Keep the two layers of that verdict distinct. The apparatus succession — Clark to human capital to frictions and monopsony to credible causal evidence — is consensus; it is the discipline’s own account of its history. The minimum-wage magnitude is the parameter-magnitude split, where Card-Krueger has largely survived replication for moderate hikes (up to roughly sixty percent of the local median) while the disemployment evidence at large hikes and on the hours margin is real and acknowledged. The mainstream’s honest answer is not that Clark was naive. It is that his assumptions do not hold everywhere, and the credibility revolution showed precisely where they break.
The thread, whole
No era-organized book hands you this thread in one piece. The marginalist chapter has Clark, the labor chapter has the Mincer equation and the matching function, and a separate debate has Card-Krueger — and you are left to reconstruct the line that runs through all of them. Here it is, in four rungs, each forced into existence by an anomaly the rung before it could not absorb:
- The competitive baseline. Clark’s marginal-productivity equilibrium — wage equals product, a binding floor destroys jobs. Genuine, clean, and silent on why similar workers earn different wages.
- Human capital. Schultz, Becker, and Mincer made skill an investment — the Mincer equation explained the wage profile Clark could not, while keeping his equilibrium discipline. It still inherited a frictionless market.
- Frictions and monopsony. DMP made unemployment an equilibrium; Robinson and Manning made employer wage-setting power the default. The apparatus now predicted its own founding prediction could fail.
- Credible evidence. Card-Krueger and Angrist-Krueger ran the experiment the theory could not. The competitive sign was wrong; the field became answerable to data.
What survives: marginal productivity, as the firm’s labor-demand foundation, and human capital, as the workhorse of wage variation — both as partial frameworks. What was overturned: the universality of the naive competitive prediction that a wage floor must destroy jobs. The man was not naive; the assumption was not universal.
Three threads run on from here. Card-Krueger is the labor field’s climax but one instance of a cross-field method revolution — how that causal-identification toolkit accumulated across the whole discipline is a forthcoming econometric-methodology thread, and what the revolution did to economic knowledge more broadly is a forthcoming synthesis on the credibility revolution. And the labor apparatus has a live frontier this thread only points at: the task framework and automation, where the question is no longer how labor markets set wages but whether machines will set them — the debate running in Will AI replace jobs? The root question all of this answers — how are wages set and employment allocated? — is the perennial Labor question this thread walked from one end to the other.