When President Franklin D. Roosevelt first proposed a federal minimum wage in 1937, he declared that America should be able to provide its working men and women “a fair day’s pay for a fair day’s work.” Congress went along, passing the 1938 Fair Labor Standards Act, which established a minimum wage of 25 cents an hour (along with the 40-hour week and prohibition of child labor). However, while the public certainly agreed that very low wages determined by an unforgiving Depression-era labor market were unacceptable, some economists worried that the imposition of a minimum wage would come at a cost.

The costs figuring most prominently in debates over minimum wages relate not to the impact on profits, but to the unintended consequence for the workers themselves: the possibility that employers would react by cutting hiring, leaving fewer jobs for workers at the bottom of the pecking order.

Actually, some economists are still concerned. Partly in response to frustrations over income inequality and a stagnant federal minimum wage, a large number of cities and states have sharply increased their minimum wages in the last few years. So once again, the debate over the trade-off between the minimum wage and job opportunities is front and center.

Here, I survey the long-running dispute. Cutting to the chase, I find we have some way to go before raising the minimum could hurt the people it’s intended to help.

Econ 101 Revisited

The potential risk to jobs posed by an increase in the minimum wage comes from considerations of supply and demand. Simply put, it seems like common sense that when labor becomes more expensive, employers will buy less of it. Thus in the standard model drilled into college freshmen, a government-mandated wage above the market-determined wage will cost jobs.

However, as you’ve no doubt guessed, this simple model is not the last word on the subject. In the standard model, employers can hire as many workers as they want at the going wage – but no workers at all for a penny-an-hour less. There is reason to believe, though, that real-world frictions in the labor market give employers some degree of wage-setting power (in econspeak: monopsony power) that muddies the waters.

There are good jobs and bad jobs at the low end of the market, and movements between these lead to vacancies and turnover. While higher costs may dissuade some employers from creating new positions, it is a price worth paying for others needing to recruit and retain workers. So moderate increases in the minimum wage can reduce vacancies and turnover instead of killing jobs, especially when there is substantial employer-side power. The bottom line: the actual employment impact of minimum wages is an open question. Theory is no substitute for real-world evidence.

A Hard Look at the Numbers

The overall weight of recent research strongly supports the view that the minimum wage increases of the magnitude we have seen in the United States in recent years generate only modest employment effects. Still, there is room here for honest disagreement.

I see two major factors driving this disagreement. First, research has generally focused on narrow segments of workers (notably, teenagers) where the evidence is less robust than for the low-wage workforce as a whole. Second, there is some “publication bias” – the tendency in social sciences to highlight clear-cut results because nobody wants to read inconclusive conclusions. Thus the literature on the minimum wage tends to focus on one-off studies that seem to find particularly large job losses. However, as evidence has accumulated, the influence of such studies on the views of professionals has diminished – and will continue to do so as we log more “natural experiments” of changes in minimum wages.

The Kids Aren’t Alright

As noted above, researchers can now identify lots of cases in which the minimum wage has been raised, giving them before-and-after data on which to estimate the job impact. But results still vary, at least in part due to both the segments of the labor force studied and the various ways comparison groups are constructed.

A piece of the challenge, then, is to identify groups for which the minimum wage is actually binding – that is, where it raises wages of a lot of the workers in the group. If the group is too heterogeneous and only a few get a raise, any impact is lost in the background noise. And if we cannot find a wage effect, we cannot really interpret any employment effect meaningfully, because we cannot be sure the higher minimum made a difference for many workers.

That’s why economists have tended to focus on groups in which a large portion are making close to the minimum wage. Until recently, this has often meant studying teens or a handful of industry subgroups like fast food and restaurant workers. Summarizing the studies on employment effect of minimum wages from the past 15 years, Dale Belman (Michigan State) and Paul J. Wolfson (Dartmouth) noted that 54 percent of them report estimates for teens, and another 35 percent for eating and drinking establishments.

Nowadays, teens constitute only around a fifth of all minimum wage workers, but close to half of teens in the workforce earn near the minimum wage. It is thus easier to detect an effect (whatever it might be) for this group than by, say, studying all workers, or less finely grained subgroups of workers like those without college degrees. Nonetheless, there is disagreement on the actual effects. It turns out that while teens don’t earn much, they are still a difficult group from which to draw conclusions because teen employment rates have been plummeting (especially since 2000), albeit with a lot of differences across states.

Here is a way to view the problem. While some states have seen around an 8 percentage point reduction in the teen employment rate over the past 40 years, others have experienced as much as a 25 percent reduction. And when there are such differences, estimates are likely to be especially sensitive to methods for teasing out the impact of background trends that just happen to correlate with changes in the minimum wage.

The scope for bias is particularly great when we compare distant points of time (like 1980 versus 2015), over which employment rates have diverged a lot. The key thing to recognize here is that the enormous reductions in teen employment rates experienced across 35 years have been primarily driven by fundamental changes in the labor market (notably, an increasing focus on reducing school drop-out rates and expanding access to college) that would have taken place regardless of what happened to minimum wages. So as liberal states like Massachusetts have pushed for teens to stay in school even as legislatures raised the minimum wage, statistical analysis might produce a spurious negative correlation between teen employment and the minimum wage.

While this problem is well understood, the solution is less clear. One strategy is to hope that while there are a lot of differences in background trends, they are not systematically different across states raising minimum wages (the “treatment group”) versus those that are not (the “control group”). This is the assumption that the treatment and control groups have parallel trends, at least on average.

A second strategy is to explicitly control for such trends using statistical procedures. A third strategy is to seek out control groups that are more likely to have experienced similar trends – for example, comparing minimum wages and employment in contiguous counties that are alike in most ways but are subject to the labor laws of separate states.

As it turns out, in comparisons across long periods, results for teen workers are quite sensitive to how one accounts for background trends. In particular, if you simply ignore the background (strategy 1), the estimates tend to suggest substantial job losses – typically, that a 10 percent increase in the minimum wage reduces teen employment by 1-2 percent. However, the picture is considerably brighter when we explicitly control for background trends using various statistical techniques. If we compare geographically close areas likely to share background trends, we typically find very small employment effects.

Generally, strategies 2 and 3 suggest that a 10 percent minimum wage increase changes employment hardly at all – usually by a percentage that is statistically indistinguishable from zero. A fourth strategy is to use an “event study” design and track employment changes in treated and control groups for, say, 3 to 5 years. Unlike strategy 1, this approach avoids comparing data points that are decades away. It, too, suggests minimal employment effects.

I think strategies 2, 3 and 4 are much more compelling than strategy 1, both because comparing similar places makes more sense, and because strategy 1 often yields paradoxical findings – such as employment changes that precede minimum wage hikes, sometimes by many years. Even restricting attention to the past 20 years of data yields very little evidence that minimum wage hikes reduced teen employment, regardless of which analytic strategy you use.

But a much more important reality here is that focusing on teens misses the big picture. Recent advances in research techniques have allowed the capture of the effect on most low-wage workers or jobs. And newer studies have found a much more consistent picture – namely that minimum wages have a very limited effect on jobs.

Numbers, Please

In a 2017 paper that I co-authored with Doruk Cengiz, Attila Lindner (University College London) and Ben Zipperer (Economic Policy Institute), we provide what is arguably the most complete picture to date of how minimum wages impact low-wage jobs. It’s based on a novel method, but the basic idea is quite simple.

Imagine the minimum wage rises from $9 to $10 an hour in state X. We compare the increase in jobs paying $10 an hour or a bit above to the reduction in the number of jobs paying below $10. This difference tells us the change in the number of low-wage jobs in state X. However, it’s possible that some of this change would have happened independently of the minimum wage policy. To account for that, we compare the changes in low-wage jobs in X to similar changes in other states that did not raise the minimum wage. This “difference in differences” estimate tells us the total number of jobs lost due to the minimum hike. We then pool the findings across 138 incidents of minimum wage changes across states between 1979 and 2014.

As the figure above shows, minimum wage increases led to a clear reduction in the jobs that had previously paid below the new minimum wage, confirming that the minimum wages we study are binding. However, the reduction in jobs paying below the minimum was balanced by a sharp increase in the number of jobs paying the new minimum, along with additional increases in jobs paying up to $4 above the new minimum. Overall, we estimate that low-wage workers saw a wage gain of 7 percent after a minimum wage increase, but little change in employment in the five years following the policy change.

The Big Picture

The passionate back and forth among researchers, as well as the technical nature of some of their disagreements, can make it difficult for nonspecialists to reach independent conclusions about what minimum wages do to employment. This is where it is useful to consider research that doesn’t analyze data directly, but takes stock in a systematic way of what other researchers have found.

An influential review by Charles Brown (University of Michigan) in 1999 for the Handbook of Labor Economics concluded that employment effects of minimum wages were likely to be small, though the results varied depending on the methods. An up-to-date survey of the more recent evidence is offered by Wolfson and Belman in their 2014 book What Does the Minimum Wage Do? They corroborate Brown’s findings, concluding that it was unlikely that the minimum wage increases under study led to job losses.

A similar conclusion was reached by other economists doing formal meta-analysis, a well-defined statistical approach of pooling the results from a large number of separate analyses. A meta-analysis conducted by Hristos Doucouliagos (Deakin University) and T. D. Stanley (Hendrix College), along with one released in 2015 by Belman and Wolfson, also concludes that the overall impact of minimum wages on employment is small.

This evidence has shifted the tide of opinion among economists. While 20 or 30 years ago most economists believed that minimum wage increases invariably caused some job loss – the big issue then was whether the trade-off was worth it – the profession has updated its views.

In 2013, the IGM Forum, a panel of 41 leading economists organized by the Booth School of Business at the University of Chicago, was asked their opinion about the desirability of raising the minimum wage to $9 an hour as proposed by President Obama, and then indexing it to inflation. Only one-third of the panel agreed that the minimum wage hike “would make it noticeably harder for low-skilled workers to find employment.” Equally striking, economists on the panel supported the Obama initiative by a 3 to 1 margin.

Then, in 2015, the IGM panel was asked the same question, but for a $15 an hour federal minimum wage. This time the share agreeing with the proposition about job loss was even lower (26 percent).

To be sure, some prominent studies have suggested very sizable job losses. But I think in part these studies have gained prominence because they stand out in a sea of less noticeable ones. A really interesting finding in the meta-analyses of minimum wage studies is that there is a clear publication bias toward big, negative effects. There are numerous papers showing this, but, most recently, Harvard University economists Isiah Andrews and Maximilian Kasy used cutting-edge techniques to show that “statistically significant” negative effects are more likely to be written up and published than studies suggesting no effects or modest positive effects of the minimum wage.

This need not imply anything nefarious on the authors’ or journal editors’ parts. If the true effect is close to zero, just by chance some research will find a large positive effect while others will find a large negative one. However, scholars may be more skeptical of a large positive effect (that a higher minimum created a lot of jobs) than a large negative one based on their prior expectations. And the former studies will be more likely to be relegated to file drawers or sent into limbo when they are not published in a journal.

At the same time, some of the more negative findings have not held up to greater scrutiny. Consider a much-noted 2017 study by Ekaterina Jardim and colleagues from the University of Washington analyzing the impact of Seattle’s move to raise its minimum wage to $13 per hour. They found a substantial drop in the total number of low-wage jobs paying $19 per hour or lower, which they interpret as evidence of widespread job loss.

But other economists, including Ben Zipperer and John Schmitt from the Economic Policy Institute, pointed out that Jardim et al. may have failed to account for a critical factor. Seattle’s wage growth was extremely high over the studied period, which led to fewer low-paid jobs because many employers were giving raises – indeed, there is a large increase in the number of jobs paying more than $19 hour in Seattle. There is thus a very good chance that Jardim et al. mistook wage growth for job losses.

The exchange over the Seattle evidence also highlights the limits of drawing conclusions from single case studies. Quirks are much more likely to affect individual cases, making it hard to separate signal from noise. This is why we put more faith in our own study averaging over large numbers of minimum wage increases while using a method quite similar to the one used by Jardim et al.

How Much is too Much?

While I think the overall evidence strongly suggests relatively small employment effects, the impact might be quite different when the new minimums are set at a high level or are closer to median wages to begin with. We are currently witnessing minimum wage increases that go beyond our experience of the past few decades. California, New Jersey and New York, for example, are on a path to increase their minimums to $15 an hour in the coming years. The point at which the employment effects begin to bite hard enough to worry remains an open question.

My study with Cengiz, Lindner and Zipperer does offer clues, though. For the cases we examined, the minimum wage ranged between around 37 and 59 percent of the median wage. At some point, of course, the minimum would be felt in terms of job opportunities. Encouragingly however, we have found that going as high as 59 percent of the median wage generated little indication of job losses.

Two recent studies – one by Martha Bailey from the University of Michigan and co-authors, another by Ellora Derenoncourt from Harvard University and Claire Montialoux from Crest – provide evidence from the late 1960s, when the federal minimum wage reached its historical peak. (The 1968 minimum equaled $11.50 in today’s dollars.) And while there were some differences in findings among subgroups, strikingly, both studies found that while the high minimum had a strong impact on wages, it had a very small overall impact on employment. Zooming back to the present, initial findings from some of the recent citywide minimum wages also provide encouraging results.

These findings suggest there is likely room to experiment with minimum wages that are considerably higher than most states have set in the past several decades. And that is a very important finding in light of the political popularity of the policy – which typically draws support from across the partisan divide.

To be sure, a minimum wage increase is not the only arrow in our quiver for raising the incomes of less-skilled workers. For example, there are good reasons to combine expansions in both the minimum wage and the earned income tax credits, as they tend to complement each other. However, the weight of evidence suggests that minimum wages are a promising route to raising living standards for the working poor. Raising the minimum wage would be a tough call if it were likely to price marginally productive labor out of the market. But as I hope this essay makes clear, the rewards from realizing FDR’s dream of assuring a fair day’s pay for a fair day’s work seem to be much greater than the risks.