Warning: Very geeky post.
Since I posted the final version of my Mariel paper earlier this week, I have heard from a number of people asking for all kinds of details about the paper. One of the nice things about having this blog is that I can quickly address these reactions/questions/doubts without having to resort to writing yet another paper. So here are some responses for those who are really into the minutiae of this stuff.
1. Which native group did the Marielitos really affect?
According to the 1980 census, 44 percent of native-born high school dropouts working in Miami were African-American, and those low-skill blacks made up almost 40 percent of the native-born black workforce. Let me rephrase it as clearly as I can: Low-skill blacks seem to be the most vulnerable native group.
2. Why is the measured wage depression smaller in the ORG than in the March CPS?
This graph shows the visual difference. Although the wage effect in the ORG is numerically smaller, it is still sizable and statistically significant.
The paper has a long footnote addressing this (note 29). It almost surely has to do with the fact that “working men” and “earnings” mean different things in the two samples. In the March CPS, I examine weekly earnings of men who worked at any time during a year. In the ORG, I examine the hourly wage rate of men who worked last week. If I manipulate the data in the March CPS to try to get close to what the ORG does, I can explain about half of the difference in estimates between the two data sets.
3. Why look at non-Hispanic men? Why not look at, say, non-Cuban men?
The switch from non-Hispanic to non-Cuban makes a difference and I spent a lot of time explaining why here. The graph below shows that switching from non-Hispanic to non-Cuban attenuates the measured wage depression. (For the true geeks, you can derive the graph on the right-hand-side by using the Figure_2_cps.do program that I posted online, and changing the “keep if hispan==0” command on line 31 to “drop if hispan==300”).
Some will prefer what the right-hand-side graph says. But before jumping to conclusions based on what a graph looks like, let’s think about what is going on. Ideally, we want to know what happened to the wage of native workers in Miami. But the CPS does not provide any information on country of birth during the relevant period, so we need to approximate “natives.” About 75% of immigrants in Miami at the time were Hispanic, so it made sense to me to use the non-Hispanic population to approximate the native population.
The key problem with adding (non-Cuban) Hispanics to the sample is that this population has grown rapidly since the 1970s. Adding Hispanics increases sample size, but many of the new observations are Hispanic immigrants who moved to the United States after 1980. The 1990 census tells us that if we were to add working Hispanics to the sample, 52 percent of the persons we would add are immigrants who arrived in the 1980s! Although that data manipulation leads to a graph that may be preferable to some, it is highly dubious. And, unless one is blindly wedded to a particular ideological position, it only takes a little common sense to see why: the quickly changing sample composition obviously contaminates the post-1980 wage trend.