The experiment that never settled.

What arrived in Miami that year arrived fast, and it was big enough to matter. The city's labor force grew by about 7%, and the increase was larger still in the kinds of jobs that need no formal qualification, because most of the people arriving had none. If a sudden jump in the number of people looking for work pushes down what employers have to pay, this is where you would see it.

It is a clean case for a reason that has nothing to do with economics. The wave was set off by Cuban politics. It had nothing to do with how Miami's economy happened to be doing that year, which is the whole thing that makes it useful: the supply of workers moved for its own reasons, and there were other American cities sitting there to compare Miami against.

That 7% is an estimate, and it is worth knowing where the estimate comes from, because knowing that is the whole point of this piece. Card built it on a settlement share he describes as widely assumed rather than measured: that about half of the arrivals stayed in Miami for good. Census data that did not exist in 1990 later put the share higher than that. Notice which way that cuts. It makes the shock bigger than 7%, not smaller, which is the opposite of what a caveat usually does for the side quoting it.

The first answer came back in 1990, and it was not the one anybody expected. Card looked at the wages and the unemployment rates of less-skilled workers in Miami and found virtually no effect. Not a small effect. Effectively none. And it held for the group with the most obvious reason to be hurt: Cubans who had immigrated earlier, who were competing most directly with the people getting off the boats.

That last one is worth one clause of care, because it is not a flat line in the data. Cuban unemployment in Miami rose after 1980 and Cuban wages fell against white wages. Card's answer is that the Marielitos had just entered the Cuban pool themselves and were pulling the average down with them, which is the same who-is-in-the-sample argument this whole piece is about, turning up on his side of the ledger for once. He calls the calculation far from conclusive in the paper.

It is the finding everything else in this piece is built on top of.

Then, decades later, Borjas reopened it. His objection was not that the arithmetic was wrong. It was about who belongs in the comparison. If most of the arrivals had not finished high school, then the workers actually competing with them are the workers who also had not finished high school, and averaging those people together with everybody else would hide whatever happened to them. Look at high-school dropouts in Miami on their own, he found, and the wage drop is not subtle. It is somewhere in the range of 10 to 30%.

That range wants three clauses, and Borjas supplies all three of them. It is a spread across which placebo and which data file you use, not an estimate with a confidence interval around it. It is men only: pool men and women together and it becomes 5 to 20%. And in the larger of the two files, the one Card himself worked from, it sits at the bottom of the range, hovering around 10%.

The instinct, when two studies disagree, is to work out which one is lying and throw the other away. Here that instinct will cost you the actual lesson, because the disagreement is the finding.

Card and Borjas are not quite answering the same question. One asks what happened to less-skilled workers in Miami. The other asks what happened to the narrowest slice of them. Both of those can be true at the same time. An average that barely moves can sit on top of a smaller group that got hit, and a group that got hit can be small enough that the average is still the number you want for most decisions. Which one you should care about depends entirely on what you are deciding.

The uncomfortable part is that slicing is also where a motivated person can go shopping. Narrow far enough and nearly any dataset will eventually hand you a dramatic number. That is not an accusation against anybody here, and the re-analyses above are arguments about method, not about motives. It is the reason the size and the stability of a sample is the first question to ask about a finding rather than the last.

The shape to notice is that the answer depends on who you count.

You will meet it the next time somebody quotes a study at you, about anything at all. Before you argue with the number, ask what is inside it. How many people, picked how, compared against whom, over what stretch of time. A surprising share of arguments about evidence turn out not to be arguments about the evidence. They are two people looking at different samples, each convinced the other one is being dishonest.