There’s been more back and forth this week in the argument over whether a giant influx of Cubans into Miami in 1980 lowered wages for low-education people already living there. A seminal 1990 paper by David Card said no. A 2015 reanalysis by immigration skeptic (and Cuban immigrant) George Borjas said yes. A 2015 blog post by me and a paper by Giovanni Peri and Vasil Yasenov said I don’t think so. And now Michael Clemens and Jennifer Hunt, both of whose work appears in my immigration evidence review, have announced the discovery of what they term a flaw in the Borjas analysis. It turns out that just as the Marielitos began arriving, the Census Bureau sharply increased its coverage of black Miamians in the surveys it conducts to monitor the pulse of the U.S. economy. Since black Miamians had especially low incomes, the racial shift had the power to generate the (apparent) wage decline that Borjas highlights. Borjas retorted on Tuesday, labeling the criticism “fake news.”
So, once more, academics are arguing. And concerned observers are confused by the dueling contentions and graphs. In an attempt to clarify, I’ll make a few points.
Disclosures and disclaimers: I used to work for the Center for Global Development, where I was a colleague of Michael Clemens. Now I work for the Open Philanthropy Project, which provides general support to CGD and specific support for Michael’s work on migration. This blog post represents my personal views and does not speak for the Open Philanthropy Project.
Four points:
1. There was no smoking gun to explain away. If you look squarely at Borjas’s Miami earnings data, as I tried to do in my 2015 post, you will find no sudden drop or trend break in 1980, the year so many immigrants arrived on Florida’s shores. Yes, wages for low-education workers do seem to have fallen over the course of a decade back then. But I think that evidence will change almost no one’s mind. If you come to it with a strong prior that immigration lowers wages of natives (or of earlier immigrations), your prior will be confirmed: look, wages went down. If you come to the data skeptical, you too will leave as you arrived: nothing unusual happened to wages in 1980. Findings that can change almost no one’s mind: that seems like a good definition of “weak evidence.”
To see what I mean, look at these new renderings of the first two graphs of my old post. Borjas uses two different Census Bureau data sets, so my two graphs plot average earnings in both. (The data sets are labeled “ASEC” for Annual Social and Economic Supplement and “ORG” for “Outgoing Rotation Group” and are explained in my old post.) Both copy Borjas in restricting to male non-Hispanics aged 25–59 without a high school education. The green dots show changes since the starting year, 1977. The orange dots adjust those numbers in exact analogy with Borjas’s 2015 analysis.1 The sparser blue dots, again imitating Borjas, take three-year averages. The leftmost dots are zero by definition, since the graph shows changes from the start. The remaining dots are set inside 95% confidence ranges, the equivalent of pollsters’ margins of error.
Do you see how wages suddenly plunged right after the Mariel boatlift of April 1980? No, neither do I. In fact, the Borjas data are compatible with the hypothesis that wages fell at a constant rate between 1977 and 1986, with no break from trend in 1980 or any other year in that span. A formal test of that hypothesis, for the adjusted averages (in orange), returns high p values: 0.82 for the first graph, 0.37 for the second. This means that any deviations from steady, long-term earnings decline in the years around the boatlift cannot be distinguished from the play of chance. (Surveyors talked to different people in different years, who had different lives.)
Moreover, the Borjas results look rather mined: they are attained by dropping Hispanics, dropping women, dropping people who completed high school, and looking at trends in wages rather than unemployment.
I would thus quibble with Clemens and Hunt in one respect. In my mind, the lack of any sudden wage drop at the time of the boatlift constitutes the main “flaw” in the Borjas analysis. And if the sudden drop is not there, it’s not begging for an explanation.
2. Nevertheless, Clemens and Hunt point up an interesting discovery. If that discovery can explain the long-term decline in wages in the Borjas sample—and they convince me that it can—it further weakens the case for the Mariel boatlift as a wage driver. As I mentioned, the Census Bureau began boosting the number of low-income blacks it interviewed in Miami for its monthly economic surveys—the surveys that generate Borjas’s data. Clemens and Hunt detail the historical reasons for this push, above all, the belief that blacks were underrepresented in Census surveys. The push coincided almost perfectly with the onset of the Mariel boatlift.
Now, the Census Bureau routinely provides weights with its data so that researchers can compensate for such unrepresentativeness by counting some data points more and others less. Ideally, these weights would correct for the under-coverage of black Miamians before 1980. Clemens and Hunt show that they don’t, and thus that the low incomes of these new interviewees had the power to create the (apparent) earnings decline that drew Borjas’s attention. This figure shows the share of blacks in the two Borjas data sets for Miami, by earnings year:
Borjas makes a subtle but interesting point in reply: Yes, coverage of low-income blacks surged circa 1980. But of the two Census Bureau surveys he uses, the ASEC exhibits this spike most sharply, and that survey asks people how much they earned last year. Thus we should expect the coverage jump in 1980 to create the appearance of an earnings fall between 1978 to 1979, before Mariel.
But if, as I contend, there is no strong evidence of a sudden wage drop in 1980, then it does not matter as much if the coverage of blacks increased out of step with the boatlift. The useful insight we can retain is that the long-term rise in coverage of low-income blacks in both data sets helps explain the long-term fall in wages in the Borjas samples. To the extent that Borjas is leaning on long-term trends, the new discovery weakens his case.
3. Borjas replies that if he drops blacks from his analysis in order to sidestep the problem discovered by Clemens and Hunt, his results persist. However, this change also exacerbates the concerns in my first point, since shrinking samples further reduces precision. The new graphs in the Borjas post include no confidence intervals and obscure the timing of any shifts circa 1980 by taking moving three-year averages. Here’s how my two graphs look if I copy Borjas in dropping blacks:
Some of those confidence intervals are really wide. For example, looking at the blue dots in the second graph, we can say with 95% confidence that the change in earnings for low-education, non-black, non-Hispanic men between 1977–79 and 1981–83 was between +10% and –39%.
This table, adapted from one in my 2015 post, shows why. It documents how small the Borjas samples get when blacks are dropped:
4. I appreciate Borjas’s transparency, if not his tone. It is gradually becoming more common for researchers to post their data and computer code. Borjas appears to do so routinely, sometimes even without waiting for acceptance by an academic journal. This has facilitated high-quality, public debate over his work.
Data and code and lots more graphs
- Regressing across Card’s five cities and controlling for year, age, and city effects. (back)