A new Congressional Budget Office analysis of a nationwide minimum wage increase has stirred up lots of debate, since it predicts that an increase would have both good and bad effects.

The CBO examines three possible levels for the federal minimum wage – $10, $12, and $15 versus $7.25 an hour now (although some states have higher floors). For each level, it asked how much take-home pay would rise, and how much employment would fall. The most common economic theory – good old supply and demand – says that there’s a tradeoff between these two effects. According to this well-known theory, minimum wage forces pay above the level that a competitive market can bear, throwing some people out of work even as it raises incomes for those who keep their jobs.

The CBO analysis estimates that the most ambitious and popular proposal, a $15 federal minimum, would decrease employment by anywhere from 0 to 3.7 million. It placed the likeliest estimate at 1.3 million jobs lost. That would lower total U.S. employment by about 0.85 percent – a serious negative effect, though not catastrophic. Meanwhile, the CBO estimates that the $15 minimum wage would raise the incomes of poor families by a modest amount.

All in all, the CBO forecasts that even taking job losses into account, a total of about 1.3 million people would be lifted out of poverty.

Policy makers and pundits will no doubt disagree on whether this tradeoff is worth it. But we should remember that CBO analyses are, by their very nature, quite uncertain things. It’s the CBO’s thankless job to make official predictions about policies that are hard or impossible to anticipate, and there are many ways their analysis could be misleading.

First, to get their estimates, the CBO researchers relied on a large body of past research. They looked at a bunch of studies on the employment and income effects of minimum wage, and used their own judgment to extract what they felt were the most plausible numbers. But many of the studies the CBO relied on are old, using data sources and methods that are now out-of-date.

More recent studies, drawing on large numbers of minimum wage changes for evidence, and using sophisticated new statistical methods, tend to find very small job losses from raising pay floors. Perhaps the most authoritative study is a 2018 paper by economists Kevin Rinz and John Voorheis of the Census Bureau, who had access to the most detailed and comprehensive earnings data that exist. They found no evidence that minimum wage hikes caused any immediate job loss. After five years, they concluded, there might be a very small amount of job loss, but much smaller than what the CBO projects.

But even more fundamentally, the CBO might be implicitly relying on the wrong mental model of how labor markets work. The basic model of supply and demand in a competitive market generally does a poor job of explaining wages and employment, for a number of reasons. One big reason is that employers have monopsony power.

In this context, monopsony power means that workers can’t just quit their jobs and immediately find a new one if their wages are lower than they’d like. Switching jobs is expensive, risky and time-consuming. And some towns or neighborhoods may just not have that many alternative jobs that utilize a worker’s skills. Thus, employers are typically able to hold wages below what a competitive market would offer. Artificial wage suppression has the added negative effect of reducing the number of people who are willing to work. In recent years, some studies have suggested that monopsony power in local labor markets is widespread.

When employers have excessive power, minimum wages cause much fewer job losses, and modest pay increases can actually raise total employment by drawing marginal workers into the labor force. An interesting new paper by economists José Azar, Emiliano Huet-Vaughn, Ioana Marinescu, Bledi Taska, and Till Von Wachter found exactly this sort of effect. The authors concluded that when there are fewer general merchandise stores (think: Wal-Mart) in a particular area, minimum wages tend to cause less unemployment among store workers. In areas with only a very few stores, higher minimum wages actually raise the number of store employees – just as monopsony theory would predict.

This single study isn’t definitive, and needs to be replicated with alternative methodologies. But it suggests that minimum wages could work very differently from how the CBO predicts. The CBO’s analysis relies on the idea that the job loss from minimum wage is proportional to the amount that it would raise earnings in a given area – in places like rural Kansas, where wages are low, this implies that the impact of a federal $15 minimum wage could be ruinous. But since small towns are precisely the kinds of places that are likely to have only a few employers, the negative impact of a higher federal minimum wage might be more muted than the CBO expects.

In other words, though the CBO report gives policy makers reason to be cautious about embracing a $15 federal minimum wage, it’s also important to be cautious about the CBO report itself. Labor markets are not well-understood, and the more economists learn about them, the more they seem to find that raising minimum wages isn’t as scary as conventional wisdom suggests.

Noah Smith is a Bloomberg Opinion columnist. He was an assistant professor of finance at Stony Brook University, and he blogs at Noahpinion.