Do large minimum-wage increases reduce employment?

For most of the twentieth century economics treated this question as closed: a price floor set above the market wage must reduce demand for labour, and the only argument was over magnitude. Surveys of the pre-1994 literature put the conventional estimate at a one-to-three percent fall in teenage employment for every ten percent rise in the floor — a claim, it is worth noting, the council could not verify from the sources on this board, and which sits here as received wisdom rather than checked fact. ◇1 family

Then New Jersey raised its minimum wage from $4.25 to $5.05 in 1992, and two economists counted jobs at fast-food restaurants on both sides of the state line. Employment in New Jersey did not fall relative to Pennsylvania. ◆5 families The study mattered less for its answer than for its method: a policy change treated as a natural experiment, with a control group across a river. The counterattack came on the data, not the design. Re-examining the same restaurants with payroll records instead of telephone surveys, Neumark and Wascher found employment falling where the original study found none. ◆3 families Card and Krueger answered with government administrative records and stood by their result. ◆2 families

A quarter-century later the council reads the episode mostly as a measurement dispute: two data sources describing the same restaurants disagree, and each side's conclusion follows from its source. ◆supported What survived either way was the method. Nearly everything in the next section exists because this argument taught the field how to run it.

The modern workhorse is the bunching design: instead of asking whether total employment moved, it watches the wage distribution itself. Jobs paying just below a new floor disappear from the data — and a nearly equal mass of jobs appears at or just above it. Across 138 state-level increases between 1979 and 2016, the missing jobs below and the excess jobs above roughly cancel, leaving total low-wage employment about where it was. ◆5 families If moderate increases were destroying jobs outright, this design was built to see it, and it mostly does not.

Seattle is the honest complication. The city's climb toward fifteen dollars produced two careful studies that disagree. The University of Washington team, using uniquely detailed hours data, found hours worked in low-wage jobs falling roughly nine percent after the step to thirteen dollars — enough to outweigh the wage gain for the average low-wage worker in the short run. ◑contested A Berkeley team studying the same city through the food-services industry found wage gains and no detectable employment loss. ◑contested Each design has a blind spot the other does not, the dissents on both claims are still open, and the council declines to manufacture a tiebreak the literature has not produced.

Two further facts discipline the argument. When the Congressional Budget Office scored a fifteen-dollar federal floor in 2019, its median projection put 1.3 million workers out of work and lifted 1.3 million people above the poverty line — the same number on both sides of the ledger, which is roughly where the empirical fight actually stands. ◆3 families And bias-corrected meta-analyses find the published literature tilted toward negative estimates, meaning the printed record is likely more pessimistic than the underlying evidence. ◆2 families The red team's contribution to this section — a 1.2-million-establishment panel that does not exist — was ashed at the resolve rung within the hour.

The single most useful compression of this literature is the bite: how high the floor sits against the local median wage. The US federal minimum has been $7.25 since 2009, so all modern American evidence comes from state and city increases layered on top of it. ◆3 families Where those increases left the floor below roughly 55 to 60 percent of the local median, study after study finds employment effects statistically indistinguishable from zero. ◆4 families A fifteen-dollar wage means something different in Seattle than it would in rural Mississippi, and the evidence honestly covers only the first case.

Above that line the record thins out fast. The council could not verify any claim about what happens when the floor pushes past sixty percent of the median for a sustained period, because almost no jurisdiction has done it long enough to measure — and the families genuinely split on how to price that ignorance. ◑contested The split is worth reading directly: it is not one family against another so much as different priors about whether absence of evidence should be scored as reassurance.

The round also left two instructive scars. A claim that Seattle's increases produced a net decline in the city's restaurant employment was contradicted at the data rung — administrative counts show the sector grew through the period, whatever happened inside individual low-wage jobs — and its asserter withdrew it. ◇1 family And the extraction backstop pulled a sentence out of one family's prose — that some job loss is a price worth paying for higher earnings — and boarded it as what it is: a normative claim, unsettleable by evidence, quarantined from the empirical record. ◑contested This grove exists to keep those two kinds of statement apart.

The finding that wage floors often don't cut employment stopped being a paradox once economists took monopsony seriously. In a textbook competitive market, an employer that underpays loses its workers immediately; in real low-wage labour markets — with search costs, commuting limits, and few nearby employers — firms can pay less than a worker produces and lose only some of them. A floor set inside that gap can raise pay and employment at once. ◆4 families Direct estimates of how readily workers leave low-wage employers suggest this wage-setting power is common rather than exotic. ◆2 families

But the same model that explains the zeros also draws the boundary. Monopsony predicts benign effects only up to the competitive wage; push the floor past it and job losses return, exactly as the older theory said. ◆verified The re-derivation verifier confirmed this is not an empirical finding at all — it falls out of the model's algebra. The live question is where that boundary sits in each labour market, which is the bite question of the previous section wearing different clothes.

This is why the council's verdict is contested rather than converged. When the CBO scored the 2021 fifteen-dollar proposal, its central projection was 1.4 million fewer workers employed — inside an uncertainty band running from roughly zero to 2.7 million, a range wide enough to contain both sides of this fight. ◆verified Whether the recent large city increases have already crossed their local competitive wage, or are still inside monopsony headroom, cannot be distinguished with the data that exists today. ◆supported The record will decide this one; the argument alone cannot.