On Mariel

A couple of readers of early drafts of We Wanted Workers made some comments last spring that planted an idea in my head: perhaps it was time to revisit Mariel and see what we could learn from that supply shock with the hindsight of 25-years worth of additional research.

I resisted the idea for a while, as I thought it would be a complete waste of my time. But it kept nagging me. So one Sunday morning I wake up, go downstairs to my office, and start looking at the March Current Population Surveys (CPS) for the 1980s. Within an hour, my monitor was flashing a graph like this one:

Screen Shot 2016-01-23 at 11.03.50 AM

And I remember saying out loud “What the heck!” except I didn’t use those words. I then spent the entire summer working time-and-a-half on my Mariel paper. The paper went through several rounds. I got a lot of feedback from many friends who read early drafts. And I even did something that I had never done before: I hired someone to replicate the entire exercise from scratch just to make sure it was right!

The paper came out as an NBER working paper in September 2015. At least in my corner of the universe, it created a disturbance in the force reminiscent of the destruction of Alderaan, leading to a debate in the past few weeks (here’s the Peri-Yasenov criticism) and to my writing a follow-up paper showing that the critics are wrong. David Frum wrote the best description of the debate in the Atlantic Monthly, and came up with a terrific phrase that I’m going to borrow from now on whenever I see this kind of data manipulation: What the critics are doing, Frum wrote, is essentially “data dredging on an industrial scale.” And here is a popular piece I published in National Review that summarizes my take on what is going on.

The critics harp on the fact that my sample of prime-age, non-Hispanic working men is small (which it is, as I explicitly noted in my original paper). But they ignore that I report many statistical tests showing the post-1980 wage drop in Miami to be statistically significant, despite the small samples.

Even worse, the only way to make sure your lying eyes see the “right” wage trend is to enlarge the sample in ways that are, at best, questionable and, most likely, just plain wrong. They want to add women, which seems fine except for the fact that many women entered the labor market in the 1980s, so that the sample composition is changing in ways that need to be accounted for. They add Hispanics, which also seems fine until you realize that a big chunk of those Hispanics were immigrants who entered the country after 1980, again changing the sample composition. And, finally, they add “workers” aged 16-18, which means that all high school students with part-time or summer jobs are classified as high school dropouts because they lack a high school diploma. This is so obviously wrong that the less said the better.

After everything is said and done, it surely seems as if something happened to the low-skill labor market in Miami after 1980, and that something depressed low-skill wages for several years. This fact has a really interesting implication. Suppose that the Mariel natural experiment is giving us the correct estimate of the wage depression. We may then be severely understating the economic gains from immigration.