The WSJ Weighs In On Mariel

The WSJ weekend edition just published a long essay (here’s an ungated pdf version) on the academic debate sparked by my reappraisal of the Mariel evidence. Ben Leubsdorf, the WSJ reporter, has been working on this story for quite some time. He flew up to Boston back in March to have an extended conversation with me, so this is definitely not an off-the-cuff reaction to whatever happens to be the controversy de jour in this seemingly never-ending (and increasingly tiresome) tale. Ben obviously did his homework, digested all the relevant work, and talked to a lot of people. I think it’s a pretty good account of the state of the debate. It made me wonder yet again where things would be today if the question of whether wages respond to shifts in supply had not been so depressingly politicized.

One of my favorite solutions to this question comes from Paul Samuelson, the Nobel-Prize-winning economist whose Nobel citation noted that he had “done more than any other contemporary economist to raise the level of scientific analysis in economic theory.”

After 1965, laws were passed greatly liberalizing immigration. A flood of immigrants has been admitted since then . . . By keeping labor supply high, immigration policy tends to keep wages low.

………..I know it’s mid June, but April Fools! I tricked you by strategically changing a few words in what Samuelson said. Can one even imagine a world-renowned economist making such a statement in today’s political environment? What Samuelson actually wrote in his introductory economics textbook back in 1964 was:

After World War I, laws were passed severely limiting immigration. Only a trickle of immigrants has been admitted since then . . . By keeping labor supply down, immigration policy tends to keep wages high.

I’ve highlighted the words I changed in the quote. As Paul Samuelson noted long ago, and as the low-skill workforce in Miami learned back in 1980, the labor market is not immune to the laws of supply and demand.

Race And Mariel

I finally finished the paper that addresses the latest Mariel-related brouhaha–the claim that the large drop in the wage of high school dropouts in post-Mariel Miami was spuriously created by a change in the racial composition of the March CPS sample. As I documented in earlier blog posts here and here, not much happens to the results of my Mariel paper when one uses race-adjusted data to look at wage trends in Miami and comparison cities. The new technical paper summarizes much of this evidence, shows that the before-after wage drop remains even if we were to start the analysis in calendar year 1979 (after the unexplained change in the racial composition of the survey), compares what happened in Miami to what happened in over 123,000 alternative placebos, and adds even more data/discussion. Put simply, the claim that the post-Mariel drop in Miami’s low-skill wage was spuriously produced is fake news.

I realize that it is the type of fake news that will be accepted unquestioningly by those who are ideologically wedded to–or financially dependent on–the notion that a 20 percent increase in supply does not change prices (at least in the immigration context). But the paper lays out all the facts and even the most cursory look at the actual data demonstrates the inescapable conclusion that something indeed did happen to low-skill wages in post-Mariel Miami. (All the programs used in the preparation of the paper are here).

One part of the paper is worth discussing more fully now, as it seems to be the direction in which the debate is headed. The point is a bit on the geeky side, but definitely worth thinking about as it shows just how easy it is to torture the data into screaming “PLEASE! STOP! THERE IS NO WAGE EFFECT!” by making what seem to be innocuous assumptions.

In a recent response to my blog posts, Clemens finally estimated the statistical model that corresponds to my analysis and concluded that although he can replicate my regression showing the drop in the “race-adjusted wage,” the ultimate answer depends on just how the race-adjusted wage is calculated. In an important sense, the blog response subtly moves the “goalpost” of the Clemens-Hunt criticism. It is no longer that the change in the black share of the workforce induced a spurious correlation that led to lower wages in post-Mariel Miami; just look at the figures in my paper or the regression evidence and it’s obvious that this particular argument is just plain wrong. It is now instead that the measured wage impact of Mariel could be zero if we calculated the race-adjusted wage in a different way.

Let me explain what a race-adjusted wage is. It is the wage we would see a black worker earn if his employer suddenly became color-blind and saw him as just another white worker. The trend in the race-adjusted wage would then show what happened to Miami’s low-skill wage in a world where race was no longer relevant.

Obviously, the race-adjusted wage is not available in survey data. It needs to be calculated somehow, usually by estimating a regression model. And this is where all kinds of tricks can be played to get different answers. So I cooked up a trivial numerical example in my new paper to get the point across in the simplest way possible.

Tale of Two Cities.png

I’m going to tell a hypothetical tale of two cities, Miami and New York. In this tale, New York did not receive any immigrants, but Miami did. The table shows the average wage of black and white low-skill workers in the two cities before and after the supply shock. Panel A at the top gives the unadjusted wage data–the data that would be available in the CPS. By construction, immigration had a much larger impact on black workers in Miami, reducing their wage from $7 to $4, while the wage of white workers fell by only $1, from $10 to $9.

Panel B shows the race-adjusted wage in each city. As I said earlier, we need to calculate that wage, and to do so we are going to use all the low-skill wage data available across cities, across race groups, and over time. We would then look at the available data in the top panel of the table, see that there is a $3 racial wage gap among low-skill workers in Miami prior to the supply shock, and use that information to infer that the race-adjusted wage of a black worker in Miami in that period should be $10. After the supply shock, we would see a $5 racial wage gap, and use that information to infer that the race-adjusted wage of a black worker in Miami should be $9. (In fancy econometrics jargon, we just ran a fully interactive regression model, allowing wages to fully vary by city × education × race × year).

Suppose that half of Miami’s workforce is black. The average race-adjusted wage in Miami fell only from $10 to $9, or 10 percent. In fact, the average wage in Miami fell from $8.50 to $6.50, or nearly a 25 percent drop. The drop in Miami’s race-adjusted wage is not all that big for a simple reason: If the calculation of the race-adjusted wage ignores that the racial wage gap in Miami might have increased because of immigration we are going to greatly understate the impact of immigration.

Panel C at the bottom of the table shows what would happen if we used an alternative calculation of the race-adjusted wage that does not throw the baby out with the bathwater. Suppose that Miami is a very small city relative to New York. We are now going to use national data on how the racial wage gap for low-skill workers changed over time to calculate the race-adjusted wage. We would again look at the actual data in the top panel and see that the average black worker nationwide earns $3 less than the average white worker both before and after the supply shock. This would imply a race-adjusted post-migration wage for black workers in Miami of $6 (or $3 less than what whites get). If we use this approach, the average race-adjusted wage in Miami fell from $10 to $7.50, or 25 percent. (In econometrics jargon, we ran a regression that allows wages to vary by education × race × year).

In short, the mechanics of calculating the race-adjusted wage matter a lot. But is it proper to calculate the race-adjusted wage by netting out the change in the racial wage gap in Miami when that change could have been caused by immigration? It seems plausible that Mariel affected the wage of black and white workers in Miami differently. There were substantial differences in the jobs the two groups held, in the occupations they entered, and in the industries that employed them. The Marielitos obviously penetrated some sectors more than others, affecting the magnitude of the racial wage gap for a particular education group in Miami relative to other cities. A “race-adjusted wage” that nets out this differential impact removes much of the effect that immigration might have had on the local labor market. As a result, it would not be surprising if the measured impact of immigration became much smaller, perhaps near zero.

The two panels of the table below shows how the bias shows up in real-world data when I calculate the actual wage impact of the Marielitos using alternative calculations of the race-adjusted wage. The top panel uses the fully interactive model, netting out the fact that the racial wage gap for high school dropouts in Miami changed over time (perhaps because of Mariel). As in my cooked-example, the measured wage effects are small, though some are still statistically significant in the ORG.

Interaction Table.png

The bottom panel instead allows for the racial wage gap at a particular point in time to vary across age groups, across education groups, and across cities–but does not net out that the racial wage gap for high school dropouts in a particular city (like Miami) might have changed over time. Note that the wage effects of the Mariel supply shock are strongly negative and statistically significant.

So the question now becomes: do we know anything about whether immigration into a particular city affects the low-skill racial wage gap in that city? In other words, does immigration affect the wages of low-skill blacks and low-skill whites differently? Amazingly enough, only a handful of papers estimate the wage impact of immigration separately for black and white workers. And out of that handful, as far as I know, there is only one paper that estimates the impact for low-skill blacks and whites. Ironically, this happens to be the classic paper by Joe Altonji and David Card. This is the relevant page from the Altonji-Card study (click to enlarge, and the relevant numbers are the ones furthest to the right in the bottom row of each table):

Screen Shot 2017-06-09 at 8.28.18 AM

It sure seems as if the negative impact of immigration on the low-skill black wage is about twice as large as the impact on the low-skill white wage, making my numerical example quite relevant. In fact, this very large estimate of the impact of immigration on low-skill blacks was the one specifically cited in Table 5-2 of the recent National Academy of Sciences report.

I know that this geeky discussion may not be particularly gripping to those who just want to know the answer (especially if one is looking for a different answer). But the statistical exercise used to compute the race-adjusted wage in a city at a point in time should not follow blindly from a kitchen-sink approach to regressions. Careful thought must be given to why racial wage differences might arise, and how the time trend of those racial differences in a particular city might be affected by immigration. It is entirely possible (and much too easy for those tempted to do so) to hide away the wage impact of Mariel by using the wrong conceptual approach.

Still More On Mariel

After posting my reaction to the new critique of my Mariel paper yesterday, a few friends and many other people contacted me to ask if I had done any additional statistical work to back up my claim that the key message in the Clemens-Hunt paper was, as I put in the title of my blog post, “fake news.” I obviously had, but chose to write a post that summarized my take in a way that would be easiest to explain to a broad audience. Yesterday I presented some simple graphs showing that wages for low-skill workers fell after Mariel even when blacks are excluded from the sample. The point of the exercise is easy to grasp, but the samples, as I emphasized, are very small.

So several people asked me exactly the same question: What would happen to the regression results in the key Table 5 of my paper if one were to redo the entire analysis using the age- and race-adjusted wage of workers? This approach has the huge advantage that we do not need to cut down on sample size. I can still use my original sample (which was small to begin with) and simply include a variable in the wage regressions indicating whether a particular worker was black or white. We can then use the regression results to calculate the average age- and race-adjusted wage in a particular city in a particular year and see if there is any Mariel effect in those trends. Here is what those regression coefficients look like. (Here are the programs for those who want to replicate; all I’ve done is add a “black” indicator variable to the individual-level regressions):

Race Adjusted Table 5

In rough terms, the regression coefficients give the percent wage difference between Miami and the placebo cities at some time after 1980 relative to what the difference was prior to 1980. So for example, the -0.117 statistic in the last column of the first row tells us that the low-skill wage in Miami relative to all other cities fell by 11.7 percent between 1977-1979 and 1981-1983.

It’s pretty obvious that the regression coefficients–which account for the changing black share of the workforce in Miami and elsewhere–still show a significant wage drop in Miami relative to any placebo one cares to pick. And this is true both in the March CPS data as well as in the ORG. So the key inference from the key regression table in my paper is unchanged. Something happened to Miami’s low-skill wage after 1980.

Several people also asked me how I could be so sure that there is no relation between the very strange increase in the fraction of black workers in Miami and the wage drop exhibited by Miami’s low-skill workforce between 1980 and 1985. Because black workers tend to have lower wages (even after adjusting for education), a higher fraction of blacks in the sample would mechanically reduce the average wage of the population. So it is certainly possible that the wage drop could be attributed to the change in sample composition.

It is trivially easy to show that this cannot possibly be the explanation by simply looking at the year-by-year data in either the March CPS or ORG. Let’s look at the March CPS first. The figure below shows the trend in the age-adjusted wage used in my original paper–which includes blacks–and plots it alongside the trend in the black share of the workforce.

March Year to Year.png

To emphasize my point, I’ve shaded in the period 1979 through 1983. It is obvious that nothing whatsoever happened to the black share of the workforce (as measured by the March CPS) in this particular period. But it is also obvious that this is the period where the average wage of Miami’s low-skill workers fell most. In short, it is impossible to explain that steep wage drop in terms of a rising black share. And this leads to an obvious inference: the Clemens-Hunt argument is not consistent with the timing of the increase in the black share and the drop in the average low-skill wage.

(A geeky point about the March CPS graph. The March CPS data in a particular year gives earnings in the previous calendar year. So, for example, the 1980 wage data comes from the 1981 survey. To make sure everything is consistently timed, I’ve lined up the graph so that the 1980 data for both earnings and percent black come from the 1981 survey).

The ORG data in this next graph is equally striking. Again, the average low-skill wage in Miami (including blacks) fell dramatically between 1980 and 1984 while the black share rose slightly and then declined slightly over the period–ending up pretty much at the same place it started. So how could the change in the black share possibly account for the drop in the average low-skill wage? It can’t.

ORG Year to Year.png

Finally, several people wanted me to opine on where things stand and where we go from here. Well, let me give credit where credit is due. Clemens and Hunt discovered a really weird thing about the racial composition of Miami’s low-skill workforce as measured by the March CPS, with a somewhat similar trend in the ORG data. This is something that future work must take into account. I think we would all agree that the ideal exercise is to track the average person over time to see what happens as a result of the Mariel shock–and we definitely don’t want that “average” to change as a result of changes in sample composition.

It would not surprise me if the weird pattern in the black share of the low-skill workforce as measured by the March surveys is the result of a data glitch or imputation problem that lies undetected in the vaults of the BLS or IPUMS offices. But I also suspect that the less weird ORG pattern of a gradually increasing black share (although with ups and downs through 1987) is not something we should altogether dismiss. This trend may contain valuable information. Could it be that, for reasons maybe related to Mariel or maybe not, the Miami of the 1980s increasingly became a place that did not reward whatever it is that low-skill whites bring into the workplace? And that is something worth investigating.

 

 

More Fake News On Mariel

It seems that the tremors set off by my Mariel paper (which first circulated privately almost two years ago; here is the published version) are still reverberating. I’m quickly losing track of all the rebuttals. But those critiques– including an early reaction written about a month after the public release of my NBER working paper by David Roodman, the Peri-Yasenov paper that appeared three months after the NBER release, and a recent exercise by Alex Nowrasteh at Cato–have not been able to demolish my evidence.

As anyone involved in the immigration debate well knows, the narrative that immigration is good for everyone must live on. Each time one of these critical appraisals comes out, the reaction is the same. A lot of gloating from the usual suspects in the interwebs about my original paper being proved wrong, etc. But, somehow, the paper refuses to retire peacefully to that burial ground populated by tens of thousands of forgotten and useless academic studies, as additional rebuttals keep appearing to beat up what the gloaters have repeatedly declared to be a dead horse.

So it is not surprising that my inbox is again cluttered with messages about yet another paper that questions my results. And this time the paper comes along with the appearance of paid-for empirical research. This new exercise was funded by a Silicon Valley “philanthropic” organization, Good Ventures. It’s hard to make this stuff up, but Good Ventures, run by Facebook co-founder Dustin Moskovitz, actually lists “love” as its first value. And, as we all know, such organizations, just like pharmaceutical and energy companies, will never fund research that offers anything but a balanced and objective appraisal of their missions.

The main criticism that Michael Clemens and Jennifer Hunt make of my Mariel paper is succinctly stated in their abstract:

We show that conflicting findings on the effects of the Mariel Boatlift can be explained by a sudden change in the race composition of the Current Population Survey extracts in 1980, specific to Miami but unrelated to the Boatlift.

I have not had the time–and most definitely do not have the desire–to go line-by-line through their code. But I can very easily dismiss their entire criticism by simply looking at what happens if I excluded all blacks from my analysis, so that the post-1980 increase in the relative number of blacks could not possibly play any role in generating the wage drop in Miami. Curiously enough, the evidence resulting from this trivially simple exercise is not reported in the Clemens-Hunt paper.

One crucial caveat: By excluding blacks, the sample size in the March CPS becomes even smaller than it was in my original Mariel analysis. Nevertheless, the results from the larger ORG samples seem similar.

This exercise is extremely easy to do with the programs and data that I put online last year. You only need to add one line to the code–a line that drops blacks from the sample (and here are the new programs). To my surprise, and despite the very small sample sizes, not much happens. Just look at the graph of the three-year moving average of the wage of non-black, non-Hispanic high school dropouts in Miami and in all other cities.

March wage

And here’s the same graph with the larger ORG sample:

ORG wage

And for those interested in regressions, these are the regression coefficients and standard errors that go along with those reported in the last column of Table 5 in my paper. As in the original paper, the regression coefficients are smaller and less significant in the ORG, but I showed that some of that arises because the ORG sample excludes many people who happened not to work in the survey’s reference week.

Revised regression table

In short, using the increase in the relative size of Miami’s black workforce after 1980 to dismiss my Mariel evidence performs the job of obfuscating the debate further, but does little to clarify.

There is no doubt that the racial composition of the sampled low-skill workforce in Miami changed beginning in 1980 (at least in the March CPS). These are the trends in both the March and ORG samples. (As an aside, there seems to be a problem with Table 1 of the Clemens-Hunt paper that confuses the survey and earnings year for the ORG. You can tell their data are wrong because they show the sample size for the ORG increasing in 1980. That increase should have been observed in 1979, the first year of the ORG files).

Percent black

But the claim that the rising proportion of blacks explains the observed decline in Miami’s low-skill wage is more than just a little misleading–it is downright false. Most obviously, note that the fraction of blacks in Miami’s low-skill workforce is relatively constant between the 1980 and 1984 survey years (representing earnings between 1979 and 1983), which just happen to be the years when the wage of high school dropouts fell most in my original paper! Here’s the graph showing the original year-by-year wage trend in the March CPS including blacks, rather than the 3-year moving average. It is trivially easy to see that the timing is off. There’s no connection between the 1979-83 large drop in the low-skill wage and the black share of the workforce. (And here’s the comparable graph in the ORG for the curious geek).

Wage and percent black

From my perspective, the increased proportion of low-skill black workers beginning in the 1980 survey raises even more questions. Could it be more than just a sampling glitch or a coding problem with the original CPS data? Where did all the white low-skill workers go? Might there be a link between their gradual disappearance in the ORG and the post-Mariel labor market dislocations? What do we make of the very different trends in the racial composition of the workforce in the March and ORG surveys? What does it say about the sacred statistics derived from the CPS?

In the meantime, however, the narrative must live on. And if there are funds to ensure its survival (and there seem to be an awful lot of “charity” organizations out in Silicon Valley trying to reenact the Summer of Love), there will surely be a large supply of researchers with an incentive to use every trick in the book to throw noise into the discussion, and further confuse and obfuscate the issue “with a little help from their friends.” Hopefully, this new Summer of Love will not come crashing down in Altamont.

Update, 5-23-17. About an hour after the blog post went online, I discovered that the specification I used in the ORG regression was not identical to the one I used in the Mariel paper. I have updated the regression table using the same specification, and also updated the programs.

Update, 5-24-17. And here are some more results.

The Weekly Standard On We Wanted Workers

The Weekly Standard just published Peter Hansen’s careful review of We Wanted Workers. Hansen did a great job, neatly capturing the essence of what my book is about. My favorite part:

it’s hard to imagine a more suitable book if you’re genuinely seeking information about what may well be today’s most politically charged issue.

Hansen also grasped the significance of the “elementary error” I point out in David Card’s Mariel study. Believe it or not, Card actually used post-1980 (that is, post-Mariel) data to create the set of placebo cities that Miami should be compared to. (The quote from the Card article describing what he actually did appears at the end of this post). As I write in my book,

This elementary error is akin to a medical researcher choosing the placebo by looking for patients who were not injected with a harmful dosage of an experimental drug but somehow got sick anyway.

This is one of those things that is universally swept under the rug when describing Card’s work. Just imagine the reaction to a young economist (or medical researcher) today if he/she published a paper where the placebo group was deliberately chosen to resemble the post-treatment outcomes of the treated group!

This blog has not been active for a few weeks. It’s been very hectic, as I’ve been downsizing and moving. We sold our big old house in Lexington and moved to a condo in Cambridge within walking distance of my office. Valuable advice for the young ones. Throw out all that junk now before it starts to accumulate and overwhelm. It’s way too much work to take care of it when you are trying to downsize.

Continue reading “The Weekly Standard On We Wanted Workers”

Earnings of Undocumented Immigrants

I have a new paper that looks at how undocumented workers perform in the U.S. labor market. Here are some of the main findings:

First, the age-earnings profile of undocumented workers lies far below that of legal immigrants and of native workers, and is almost perfectly flat during the prime working years. Second, the unadjusted gap in the log hourly wage between undocumented workers and natives is very large (around 40 percent), but half of this gap disappears once the calculation adjusts for differences in observable socioeconomic characteristics, particularly educational attainment. Finally, the adjusted wage of undocumented workers rose rapidly in the past decade. As a result, there was a large decline in the wage penalty associated with undocumented status.

It is this last result that I find particularly intriguing. Define the “wage penalty” to undocumented status as the difference in wages between observationally equivalent legal and undocumented immigrants. Here’s the graph that illustrates just how noticeable the decline in the wage penalty has been:

Wage Penalty

As I conclude in the paper, as long as we take these trends at face value, it seems that “a regularization program may only have a modest impact on the wage of undocumented workers.”

This paper is a follow-up to my earlier work on the labor supply of undocumented immigrants, which showed that undocumented men have very high labor force participation rates and inelastic labor supply. That paper is now finished and forthcoming in Labour Economics. Click here if you’d like to access the code that I used to conduct the calculations reported in that paper, including the code that imputes undocumented status in the post-1994 Current Population Surveys. As soon as I have a little extra time (we’re in the midst of downsizing and moving), I’ll clean up and post the code for the new paper as well.

 

 

Immigration And Poverty: Updated Facts

Last year, I published this paper discussing the link between immigrant poverty and the 1996 welfare reform legislation. It showed that although the legislation cut the number of immigrants receiving assistance, their poverty rate actually fell. Many of the affected immigrants resorted to good, old-fashioned employment as a means of support.

There have been some rumblings recently about how those results translate in today’s immigration debate. Although the paper was published last year, I actually wrote it years ago, focusing on the period around welfare reform and examining data through 2001. But then I got sidetracked into various other things. (FYI: there are two or three other papers in my computer hard drive awaiting resurrection in the same way).

The current discussion sparked my curiosity: What do immigrant poverty rates look like now?

poverty-rate

So I took out the Current Population Survey (CPS) data from 1994 to 2016, the period over which this exercise can actually be done, and proceeded to calculate the poverty rate of immigrants aged 20+. I dropped the children to avoid the complication of whether the US-born children of immigrants should be counted as “immigrants” or “natives.” In any case, a few minutes later I ended up with the graph above.

It is easy to see the post-1996 decline in poverty rates that motivated my paper. But it is also easy to see the very adverse impact that the Great Recession seems to have had on immigrant poverty. Their poverty rates rose substantially, from 14.1 percent in 2007 to 18.7 percent in 2011. The increase was much smaller for natives, from 9.7 to 11.5 percent. It is also interesting that the economic recovery reduced poverty rates much more for immigrants than for natives. I am sure there’s a paper waiting to be written that explains the “excess sensitivity” of immigrant poverty rates to outside shocks.

A number of readers have indicated they like these types of fact-based blog posts, where I use available data to easily compute relevant statistics. So I’m creating a new tag for this type of post simply called “Factoids,” and I’ll try to go back in the archives and recategorize past posts. And for the true geeks, the STATA code that creates the graph is, as they used to say, below the fold.

Continue reading “Immigration And Poverty: Updated Facts”