Why Statistics Are Not “Damned Lies”. The effect of Mariel boatlift on Miami wages



Join 36.6K other subscribers

We all have heard the phrase: “There are three kinds of lies: lies, damned lies, and statistics” (for its origin, see the entry in Wikipedia). This phrase is a damned lie.

You cannot lie with statistics if they are properly done; that is, if you show both the data and the method you used to analyze the data. Only the innumerate can be lied to with statistics. “Innumeracy” is inability to deal with numbers and quantitative reasoning, just as “illiteracy” is the inability to read. Unfortunately, although only 30 percent of Americans are functionally illiterate, a much, much higher percent are functionally innumerate. However, for important issues, you will always find numerate people on both sides, and thus when one side attempts to lie with statistics, the other side will point out the problems and perform alternative analyses. Numerate public, by spending a little effort to follow the logic of both sides, will immediately see who is right and who is wrong. And that’s why statistics are very different from “damned lies.”

Here’s a good example of how this works in practice, with me a numerate bystander in one very important debate: does massive immigration depress wages of native workers?

The proper beginning of the story is 1962 when the young George Borjas arrived as a Cuban refugee in Miami. Eventually he became an economist and was hired by Harvard. His specialty is labor economics and he is one of the foremost American experts on the consequences of immigration for labor markets.

Then, in 1980 Fidel Castro allowed a mass exodus from Cuba, which became known as the Mariel boatlift. Within months more than 100,000 immigrants arrived in Miami.



The following graph taken from Borjas forthcoming paper illustrates the magnitude of this this labor supply “shock,” to use economics jargon:


What we see here is two initial waves during the 1960s, following the Cuban Revolution of 1959 (including Borjas as a “data point” for 1962). During the 1970s emigration from Cuba was shut down by the Castro regime. The huge spike in 1980 is the Mariel boatlift, after which emigration from Cuba was again shut down as a result of behind-the-scenes agreement between Cuba and the US. The smaller spike around 1995 is known as “Little Mariel.” More recently the increase in Cuban immigration is due to the wet-feet, dry-feet policy.

What we have here is a perfect natural experiment to find out how massive immigration influxes affect the wages of native workers. The Berkeley economist David Card saw the potential of this labor supply shock and used it in a paper that was published in 1990 in Industrial and Labor Relations Review.

cardThe 1990 Card study became a classic and had been used as one of the strongest arguments in supporting the view that immigration has no negative effects for us to worry about.

Now fast forward to 2015, when one summer morning George Borjas decided to revisit this analysis in light of what we have learned about immigration effects since 1990 (much of it due to Borjas own efforts). You can hear him tell the story in this video.

He did something very simple, and you can actually see what he did if you watch the 6-minute section of the video that starts at c.6:30. He used the same CPS data as David Card. However, he focused only on workers that were (1) non-Hispanic (as the best approximation to the native-born), (2) aged 25-59 (prime working age), (3) male, and (4) high-school dropouts. The last characteristics is key, because 60 percent of Marielitos did not complete high school. And even many of those of the rest 40 percent, who did, were looking for unskilled jobs due to their lack of linguistic and other skills. So Marielitos competed directly with high school dropouts, and if there is an effect on the wages, this is where we should look.

Borjas next compared the inflation-adjusted wages of Miami residents, who had these characteristics, to wages of the same segment of the American population in all other American metropolitan areas but Miami. And here’s what the data say:

resultThe vertical line at 1980 indicates the arrival of Marielitos. The blue curve for Miami begins diverging from the black curve (other metropolitan areas) after 1980 and the difference reaches its maximum around 1985. The reason that it takes time for the effect of labor oversupply to reach its maximum impact is that wages are, in economics jargon, “sticky”—it takes several years for them to adjust to new labor market conditions. I saw the same effect in my own analysis of the effects of labor oversupply on national wages in the US. Interestingly, I also estimated the lag effect at 5 years, although at the time I did not know of Borjas analysis (well, because I did mine two years before—in 2013).

Eventually other forces come into play and the wage gap shrinks. Another divergence occurs following the Little Mariel in 1995 and the gap is not closed by the end of the series, probably due to the constant, if at a lower level, immigration into Miami. The blue band and the black dotted line are the 95% confidence limits. What they tell us is that when there is no overlap, the difference between the two curves is statistically significant—highly unlikely to happen by chance alone. In other words, the Miami wages for native-born men without high school diplomas were indeed much lower than for similar workers in other US metropolitan areas during the 1980s and then again in the late 1990s, following the two spikes of Cubans migrating to Miami. During the 1980s Miami wages were 20 percent lower than elsewhere. A very substantial effect.

And here we are. What we have here is David Card and George Borjas starting with the same CPS data. Then they used clearly described procedures in analyzing it. We know precisely why their results are different. One of them is clearly wrong, and you can decide whose procedure is better (I known what my answer is). So where are “damned lies”?


Most Voted
Newest Oldest
Inline Feedbacks
View all comments
al loomis

if i understand the basic principles of capitalism, increasing supply usually drives down cost. more labor, lower wages. when you find this isn’t true, you should look again, unless you are getting paid not to look at all.


Since the point of this blogpost is transparency in statistical exercise, shouldn’t we note the Peri & Yasenov critique of the recent Borjas paper?

“We find that the main reason is the use of a small sub-sample within the group of the high school dropouts, obtained by eliminating from the sample women, non-Cuban Hispanics and selecting a short age range (25-59). All three of these restrictions are problematic and, in particular, the last two as they eliminate groups on which the effect of Mariel should have been particularly strong (Hispanic and young workers). We can replicate Borjas’ results when using this small sub-sample and the smaller March CPS, rather than the larger May-ORG CPS used by all other studies of the Boatlift. The drastic sample restrictions described above leave Borjas with only 17 to 25 observations per year to calculate average wage of high school dropouts in Miami. This increases the measurement error so substantially that not much can be learned from the data. We show that the measurement error for average log wages across metropolitan areas in the Borjas sample from the March-CPS has a standard deviation of 0.15 log points. Hence differences of 15-30% in average wages can easily arise between two cities only because of measurement error.”

Personally, I agree with the Borjas POV on these matters, but on a first principles basis. Not because of this particular pointless statistical exercise.

This comment should not be allowed to stand so easily:

Yes, isn’t it interesting that the law of demand and supply is one of the most fundamental principles in economics, yet when it comes time to apply it to the price of labor, mainstream economists become shrinking violets.”

Not really. “Price of labour goes down when there is excess supply of labour” is true ceteris paribas, and in partial equilibrium. But in general equilibrium, the long-run effect can be different, because there are other (dynamic) responses. That’s why even Borjas’s estimates of the aggregate effect of 1990-2010 immigration on wages are fairly modest. See Table 5.2 in Borjas’s textbook https://books.google.com/books?id=onhlAwAAQBAJ&pg=PA126&dq=table+5.2+simulated+wage+impact+of+1990-2010+immigrant+influx&hl=en&sa=X&ved=0ahUKEwjk3e-kovjNAhUEPiYKHY8bCoUQ6AEIHjAA#v=onepage&q=table%205.2%20simulated%20wage%20impact&f=false


Peri & Yasenov is here http://www.nber.org/papers/w21801


Oh I forgot: Borjas’s response to Peri & Yasenov’s critique of Borjas: https://www.hks.harvard.edu/fs/gborjas/publications/working%20papers/Mariel2015a.pdf


Yes. Which is why the Miami wage even in Borjas’s graph eventually goes up after the shock !


But I already agree that it is due, in the aggregate economy, to mass immigration in large part! But I agree because of Borjas’s own computable general equilibrium analysis in his textbook. You can’t do it with your time series curve fitting methods.


There is also Claudia Goldin’s analysis from 1994 about the impact of immigration on unskilled wages. Goldin is another economist from Harvard who isn’t a “shrinking violet” per your description. Then there’s the pair of non-shrinking violets Hatton & Williamson 1998 (Williamson is also from Harvard!): “these magnitudes suggest that the real wage would have been 4.7-5.9 percent higher in the absence of immigration after 1890, and 10.9-13.7 percent higher in the absence of immigration after 1870.”

Steve Sailer

I’d point out that the Mariel Boatlift of 5/80 coincided with the Great Miami Cocaine Boom of the early 1980s (c.f. “Scarface,” “Miami Vice”), so Miami’s economy from 1980-1984 was uniquely supercharged by one of the most legendary events in economic history:



But Steve, that’s the common argument of both minimum wage AND open borders advocates: the ceteris are so un-paribas that the negative effects, if real, are overwhelmed by other factors. Even in the Borjas graph above wages recover fairly rapidly after each immigration shock. Hell, even convergence with wages outside Miami is within the margin of error

Franz Ramsey

As an economist (not particularly working on labour though) I know since quite a while of this forthcoming article from Borjas. Yet I haven’t thought about it diligently. Thank you for picking it up.

What would be our best guess if we map this findings into a European context? Minimum wages (or minimum wage equivalents – e.g. collective bargaining – that are not referred to as minimum wage) exist in virtually every country and prevent wages from dropping. So there must necessarily be another channel through which these effects are diverted. Or am I wrong?

Gene Anderson

Actually looks like a fair case for “lies, damned lies, and statistics.” To resolve the difference I’d have to know more about what sample Card was looking at, what the non-HS-ediucated workforce in Miami is like relative to other cities, and above all what Florida’s notoriously awful and anti-worker government did during the time. There are lots of reasons why the wages of the less educated could have dropped at that point. Stats are always easily used to back up any position; just choose your initial data point and what other confounding variables to ignore. So I still don’t really know who’s right for what subpopulations.


Good point by Gene Anderson.

The assumptions should always be examined, and the comparison shouldn’t be with “all other metro areas”, but “all other metro areas with laws, culture, and an economic mix or going through an economic transition similar to Miami”.


Not the exact same, but there are certainly other metro areas in FL and other metro areas in states with weak labor laws, not much unionization,not much heavy industry, etc.

Wouldn’t a comparison with those metro areas be more legitimate than a comparison with all metro areas (including those with heavy unionization, strong labor laws, heavy industry, etc.)? I certainly think so. You can’t compare like with exact like, but comparing with some-what like is better than comparing with a heterogenous mix that includes metros that look nothing like Miami.


Reading the series of filters imposed on the data to get at the result reminded me of Andrew Gelman’s “garden of forking paths” paper


Warren Dew

Is the base for the log wage level 10? That is, the effect was a fall in weekly wages by a factor of 3?

  1. Home
  2. /
  3. Cliodynamica
  4. /
  5. Regular Posts
  6. /
  7. Why Statistics Are Not...

© Peter Turchin 2023 All rights reserved

Privacy Policy

%d bloggers like this: