Do minimum-wage increases destroy jobs?

Your city votes to raise its minimum wage, and the raise is coming to your paycheck. The first feeling is relief. The second, if you have ever heard the standard warning, is a colder one. You close the restaurant four nights a week, the place runs on thin margins, and someone has told you, plausibly, that when an hour of your work costs the owner more, the owner buys fewer hours of it. So the question arrives fast and personal, well before any argument about policy: when the floor goes up, do you keep your shifts, or does the promised raise turn up attached to a thinner schedule, or to none at all?

The warning has a clean logic and a century of textbook authority behind it, and the case for it belongs first, before the test. It also happens to be among the most heavily examined propositions in labour economics, because the last three decades turned it from a blackboard exercise into a question answered with real data from real wage floors. The short form of what that data says is this: at the levels minimum wages have actually been set in the United States and comparable countries, your shifts most likely survive, and the higher wage bill tends to surface as slightly higher prices, marginally fewer hours, and lower staff turnover rather than as lost jobs. The longer form keeps one caution in permanent view. That finding holds at the levels that have been tried, and nobody knows how high a floor would have to climb before the textbook warning comes true. This chapter is about both halves at once, because leaving out either one turns the answer into a slogan.

The prediction that makes the warning sound obvious

Begin with the case for the warning, which is the standard model, not a straw man. Treat an hour of labour like any other thing that is bought and sold. Employers will hire more hours when labour is cheap and fewer when it is dear, so the demand for labour slopes downward in its price. Workers will offer more hours when the wage is high and fewer when it is low, so supply slopes upward. Left alone, the two settle at a wage where the hours employers want to buy equal the hours workers want to sell. Now impose a floor above that settling point. At the higher forced wage, employers want fewer hours than before, more workers want to supply them, and the gap between the two is exactly the people who would have worked at the market wage and now cannot find the hours. The model even predicts who they are: the least experienced and least productive, the ones whose output is worth least to an employer, priced out first. If you make something more expensive, less of it gets bought, and there is no obvious reason labour should be the exception.

The test that put the prediction on trial

For most of the twentieth century the argument stayed mostly an argument, because the data could not settle it. Every time a minimum wage rose, a hundred other things changed with it, and picking the wage floor’s effect out of the noise was guesswork. Then in 1992 New Jersey raised its minimum wage and neighbouring Pennsylvania did not, and two economists, David Card and Alan Krueger, treated the coincidence as an experiment. They surveyed fast-food restaurants on both sides of the line before and after the increase, the kind of employer the textbook says should cut staff first. They found no drop in employment in New Jersey relative to Pennsylvania, and if anything a small rise. The result was startling enough to draw years of challenge, and the challenge forced everyone to be explicit about method.

What grew out of that fight was a way of testing the question that no longer depended on a single pair of states. Because state lines frequently cut through a single labour market, one can compare every county that sits against a state border with the county just across from it, one side having raised its minimum wage and the other not. The two share the same regional economy, so a difference in low-wage employment between them is hard to attribute to anything but the wage floor. Applied across hundreds of such county pairs, this design generally found the same thing Card and Krueger had: employment on the higher-minimum side tracked employment on the lower-minimum side closely, with no clear gap opening up. A related approach counted jobs directly by wage level, and found that after an increase the jobs paying just below the new floor did not vanish from the economy; they reappeared just above it, so that the number of jobs paying at or above the new minimum rose by about as much as the number below it fell, leaving total low-wage employment roughly unchanged. The workers were still working. They were simply working at the higher wage.

Where the estimates land

One study is an anecdote and a handful is a debate, so the question is what the whole body of credible estimates looks like laid out together. The trouble with any such picture is that whoever chooses which studies to include can produce nearly any answer, and both sides of this argument have accused the other of exactly that. The way through is to use a collection whose membership rule is written down where anyone can check it, rather than one assembled to make a point. The figure below plots one such collection: a public repository, maintained by the economists Arindrajit Dube and Ben Zipperer, that records one representative estimate from every minimum-wage study meeting a stated set of conditions.

Read the figure for its shape. The central estimates crowd around zero, the median close to a seventh of a point of employment lost for each proportional unit of wage gained, small enough that most of the studies with a measurable margin of error cannot rule out no effect at all. This is what economists mean when they say the modern evidence finds minimum-wage employment effects near zero at observed levels. A review commissioned by the United Kingdom government reached the same reading of the international evidence, and the pattern holds across the United States and across countries.

The objection, kept in view

The lean is real, and so is the dissent, and the dissent has a name. For three decades David Neumark and William Wascher have argued that the near-zero finding is partly an artifact of the methods that produce it. Their case has two edges. The first is substantive: they contend that the comparisons which find no effect, the neighbouring counties and matched regions, may quietly absorb the very job losses they are meant to detect, because a place that raises its minimum and a place next door are not always as alike as the design assumes, and that a wider look at the evidence, especially for teenagers and the least-skilled, still turns up negative effects. The second edge cuts at the figure above. The repository plotted there admits a study only if it first finds a clear, statistically significant effect of the policy on wages. That rule sounds innocuous, and there is a reason for it, since a study that cannot detect the policy moving wages is poorly placed to measure the policy moving jobs. But the same rule can tilt the sample: it keeps the cases where the minimum wage demonstrably bit, and it screens out designs that find neither a wage effect nor an employment effect, so some of the clustering near zero may be built in by the entry criterion. That caution is this chapter’s own reading of the rule rather than a published objection to this repository; the nearest published complaints from the Neumark camp are narrower, that own-wage elasticities are hard to compute cleanly and that collections restricted to studies which can supply them cover a different universe from the literature at large. This is a live methodological contest, not a settled question dressed as one; the figure’s rule is stated in the open, where it can be argued with, and the reader is entitled to know that the clustering depends in part on a choice about who gets counted.

What the dispute does not do is erase the finding. Even critics of the county-pair method work with estimates that, for the wage floors actually enacted, are far smaller than the older time-series studies once suggested, and the disagreement is now about whether the true effect is near zero or modestly negative, not about whether raising the minimum at observed levels throws large numbers of people out of work. That older, larger claim is the one the evidence has moved away from.

Where the adjustment goes, and the level nobody has tested

If the jobs mostly survive, the higher wage bill has to go somewhere, and tracing where is what keeps the finding from sounding like a free lunch. Some of it passes into prices, a few cents on a fast-food meal, spread across all the customers rather than concentrated on the displaced worker. Some of it comes out of turnover: a workplace that pays more loses fewer people, and every worker who stays is one the employer does not have to find, hire, and train again, so part of the raise funds itself. Some surfaces as marginally fewer scheduled hours or slower new hiring at the edges, which is a real cost even when the headcount holds. And some comes out of profit, which is possible without job loss when an employer had the power to set wages below what a competitive market would pay in the first place, so that a floor pushes pay back toward that level rather than above it. That last mechanism is the leverage of the opening chapter seen from the other side (how pay is set by bargaining position rather than by the value of the work): where employers hold most of the cards, a legislated floor can raise the wage without costing the job, because the wage was being held below the worker’s value, not above it.

All of which carries a boundary that the chapter’s thesis refuses to hide. Every estimate in that figure comes from a minimum wage set within the range that has actually been tried, roughly up to half or a little more of a region’s typical wage. None of it can speak to a floor set far above that range, because no such floor has been imposed and measured. The textbook mechanism is not wrong; it is waiting. Push the wage high enough above what workers produce and employers will buy fewer hours, exactly as the model says. The evidence establishes that this point lies above the levels reached so far, not that it does not exist. When official forecasters have estimated the effect of a national floor well beyond the tested range, they have projected real job losses, precisely because they are extrapolating past the evidence into the region where the textbook prediction is expected to reassert itself. Those forecasts are projections for an untested level, not measurements of an observed one, and they belong on the other side of the line the evidence can actually draw. The floor is close to costless in jobs at the heights tried so far. It is not costless at any height, and no one yet knows where the turn comes.

Why this one is worth getting right

The minimum wage is a case where the economics you can reason out from a chair and the economics you can measure in the field come apart, and where the measuring turned out to matter. The intuitive model is not junk; it names a real force and correctly predicts what a high enough floor would do. It simply turned out to be a poor guide to what the floors people actually pass have done, and it took a generation of careful comparison to establish that, one state line at a time. Holding the two findings together, that jobs survive at tried levels and that the textbook waits at untested ones, is harder than repeating either the warning or the reassurance alone, and it is the position the evidence supports.

The next of these flashpoints turns from the floor under everyone’s pay to a gap that opens between two groups of workers doing the same economy’s work. The claim that women earn a fixed fraction of what men do for the same work is repeated as often as the claim that the minimum wage kills jobs, and it dissolves and re-forms in much the same way once the measured pieces are pulled apart. What actually explains the gap between men’s and women’s pay is where the volume goes next.