Does a higher minimum wage cost jobs? Image: RamenJunkie, public domain

Does a higher minimum wage cost jobs?

On 1 April 1992 the minimum wage in New Jersey rose from 4.25 to 5.05 US dollars an hour, an increase of 19 per cent. Across the Delaware River in Pennsylvania it stayed at 4.25. David Card and Alan Krueger, then both at Princeton, saw the chance for a comparison that the textbook model made a clear prediction about, and telephoned 410 fast-food restaurants in the two states before and after the rise to ask how many people they employed.

What the textbook predicted

In the competitive model of the labour market, which Unit 12 of the masterclass sets out, a firm hires workers up to the point where the value of what the last worker produces equals the wage. A minimum wage above the market wage raises the cost of the last worker above their product, so firms let some workers go, and employment falls along the demand curve. The size of the fall depends on the elasticity of labour demand, and the prediction of a fall is unambiguous. Most economists in 1992 believed it, and the few studies then available, which compared teenage employment with the minimum wage over time, broadly agreed.

What Card and Krueger found

Their 1994 paper in the American Economic Review reported that employment in the New Jersey restaurants had not fallen relative to the Pennsylvania ones. Measured in full-time equivalents per restaurant, the New Jersey stores had gained a fraction of a worker on average while the Pennsylvania stores had lost about two, so the difference in differences was positive, about 2.7 workers per store, with a standard error large enough that the estimate was consistent with no effect and inconsistent with the sizeable fall the competitive model predicted. Prices of a meal rose in New Jersey by about 4 per cent relative to Pennsylvania, which suggested that part of the cost had been passed to customers.

The method mattered as much as the result. Comparing New Jersey before and after would have confused the wage rise with everything else that happened in 1992, and comparing New Jersey with Pennsylvania at one date would have confused the wage with every other difference between the states. Taking the difference of the two differences removes both, provided the two states would have moved in parallel without the policy, and that is the assumption a reader has to judge. Unit 1 of the masterclass uses this study as its one worked example of how economists tell cause from correlation, and the research spotlight of that unit, by Joshua Angrist and Jörn-Steffen Pischke, explains why designs like this changed the field. Card shared the 2021 Nobel prize in economic sciences for this and related work; Krueger had died in 2019.

The argument since

The paper was attacked at once. David Neumark and William Wascher obtained payroll records from a subset of the same restaurants and found a fall in employment; Card and Krueger replied with official employment data for the whole state that showed none, and the exchange was published in the same journal in 2000. Reasonable people read the two sets of data differently, and the episode taught the field that telephone surveys and payroll records can disagree.

The research that followed used far larger data. Arindrajit Dube, William Lester, and Michael Reich compared every pair of neighbouring counties on either side of a state border with different minimum wages between 1990 and 2006 and found no detectable loss of jobs in restaurants. Doruk Cengiz, Dube, Attila Lindner, and Ben Zipperer, in a 2019 paper in the Quarterly Journal of Economics, looked at 138 state-level minimum wage increases from 1979 to 2016 with a new method: they counted the jobs paying just below the new minimum that disappeared and the jobs paying at or just above it that appeared. The jobs that disappeared from below were almost exactly matched by the jobs that appeared above, which is what one sees if the increase raised pay without reducing the number of jobs. The study of Seattle's rise towards 15 dollars in 2015 to 2017 produced two teams with two answers from the same city, one finding reduced hours for low-wage workers and the other finding no loss of jobs in restaurants, and the difference came down to which workers and which firms each team could observe.

Why the answer could be small

There is a respectable theory behind a small or zero effect, and it is older than Card and Krueger. If an employer has some power over the wage it pays, because workers cannot costlessly move to another employer, the firm pays less than the value of the last worker's product and hires fewer people than a competitive market would. Economists call such a firm a monopsonist, by analogy with a monopolist on the selling side. A minimum wage set between the monopsony wage and the competitive wage then raises both pay and employment, because the firm can now hire more workers at the fixed wage without raising the wage of everyone it already employs. Push the minimum above the competitive wage and jobs are lost as the textbook says. The theory therefore predicts a range: small increases from a low base do little harm and may help; large increases do harm; and the point at which one becomes the other depends on how much power employers have, which differs between a city with a hundred employers and a town with one.

A worked number

The research literature summarizes its findings as an employment elasticity with respect to the minimum wage: the percentage change in employment of the affected group divided by the percentage change in the minimum. Estimates from the better studies of the last twenty years cluster between about 0 and −0.1 for the groups most exposed, with the older time-series studies nearer −0.1 to −0.3 and a few studies of very large increases finding more. Take the New Jersey rise of 19 per cent and an elasticity of −0.1. Employment of minimum wage workers falls by about 1.9 per cent, so for every hundred such workers, two lose their job and ninety-eight get a rise of 19 per cent. The total wage bill of the group rises by about 17 per cent. At an elasticity of −0.3 the figures are six jobs lost per hundred and a wage bill up 12 per cent. That is the trade-off, and the choice between the two sides of it is a political judgement that economics can inform and cannot make.

The honest answer

Moderate increases in the minimum wage, up to perhaps half or 60 per cent of the median wage, have small effects on the number of jobs according to most of the evidence from the United States and Europe, and the effects grow as the minimum approaches the median. Employers respond in other ways too, through prices, hours, turnover, and the pace of automation, and these take longer to measure. The debate about the 2025 and 2026 increases in several countries will be settled by the same kind of evidence, with the same care about what is being compared with what. Unit 12 of the masterclass sets out the competitive and monopsony models side by side, works through Card and Krueger in detail, and places the bunching evidence next to it as the current state of knowledge. Its research spotlights, on Claudia Goldin's and Raj Chetty's work, concern two other questions about who earns what and why.

Study the economy properly

The Economics Masterclass: weekly units, quizzes, a final exam, and a certificate.
About the program