Non-experimental evidence

Angus Deaton recently said that all the attention that natural (and actual) experiments are getting is over blown. He claims experimental data has no special status in a hierarchy of evidence. I agree to the extent that I don’t think we should favor one form of evidence to the exclusion of other types of evidence ((I tenured member of the Cult of Identification told me once that she wouldn’t write a paper about a topic unless there was a clear source of exogenous variation. She proudly told me that she hadn’t used an instrumental variable in years.)). Evidence is evidence.

A readily available form of evidence about the relationship between native employment opportunities and immigration is cross-section data ((In applied micro seminars, you often hear Cult members hiss something to the effect, “But those estimates are from cross-sectional data”. With grimaces around the table at the mention of the taint.)). These data describe various geographical regions or worker skill groups. For each region or skill group, the analysts assigns average wages (or other employment outcome) and the percentage of the group that is immigrants. Then the analyst checks to see if there’s a correlation among the groups between wages and the number of immigrants.

As you can imagine, there’s a lot for the interested analyst to play with. Every country has its own data sources. You can change the definition of skill group. You can look at larger geographic regions like states or smaller ones like cities. And, like always, you can choose from the palette of statistical techniques to calculate your estimated correlation and effect size. Longhi, Nijkamp and Poot did a meta-analysis of 18 papers that reported 348 estimates of this correlation.

As a quick demonstration of what these papers look like, I’ve downloaded some Census 2000 data from IPUMS USA. For each state, I calculated the percent of workers that are foreign born and the average wage for native workers. Here’s the plot:
I’ve drawn the regression line. Surprisingly, the line has an upward slope, suggesting a positive correlation. The slope of the line is about 1.5.

One thing that’s wrong with this plot, besides the fact that I haven’t controlled for a bunch of obvious things, is that this simple correlation conflates the impact of immigration on native wages with the shared economic incentives of natives and immigrants to move to states that have positive wage growth. Both immigrants and natives will want to move to states that have good wage prospects; they select themselves, to use the jargon. We really only care about the first thing, the impact of immigrants on natives, and so we’d like to wash this correlation to get the stain of “selection” out.

A neat regularity among immigrants is that they tend to move to regions where previous immigrants had already called home. We’ll leave it to sociologists to tell us why this might be the case and for the moment just exploit this fact for our statistical purposes. We can predict the percentage of immigrants in a state in the year 2000 by looking at the percentage of immigrants in that state several years before. Here’s a plot:

The red line is the regression line and the black line is the 45 degree line. As you can see, the percentage of immigrants has uniformly increased in those 40 years, but the red line is positively sloped and the dots cluster pretty well around the regression line. The immigrate ratios in 2000 are predicted pretty well by their ratios in 1960(!).

So what? Well, suppose the percentage of immigrants in a state does not have an impact on the relative wage prospects in that state 40 years later. The prediction of the year 2000 immigration ratios using the red line, then, should be unrelated to the wage prospects for immigrants (and natives) in that year. This prediction is just the detergent we needed to get rid of the stain of selection. Basically, we’re taking the variation of immigrant ratios due to selection out and only looking at the variation due to immigrant clustering. Here’s a plot of native wages versus predicted year 2000 immigrant ratios:
The slope on the regression line is 1.8. That this slope is close to the slope of the one where I didn’t correct for selection suggests that selection isn’t that big of a deal.

While its size is a bit big and so makes me think I did something wrong, the sign of the slope I’ve estimated isn’t surprising. Longhi, Nijkamp and Poot found that almost as many estimates of the effect of immigration on native wages are positive as negative. Here’s their figure 1 which shows the distribution of estimates across analyses:
The estimates seem to cluster around zero. My estimates are 1.5 standard deviations away from the mean; not too bad for a quick and dirty analysis!

So even the non-experimental evidence suggests immigrants have little impact on native wages.

Even more zeros

There’s more experimental-like evidence that immigrants have little or no negative effect on native employment opportunities, if you need it.

In 1962, 900,000 pied-noirs repatriated to France after fleeing Algeria following their loss in the country’s war of independence. These repatriates settled in the warmer departments of France with climates more similar to their former home country. And as evidence that this was an exogenous shock to the labor force, they tended to settle in those departments with the higher unemployment and lower wages.

This exogenous shock to the French labor force that increased the number of workers by 1.7% was studied by Jennifer Hunt in her 1992 paper. She found a small but significantly negative effects on native unemployment six years after the mass repatriation. However, she found no effect on native participation rates. Her data on wages is a bit of a mess, but with what she had she found at most a small negative effect ((She worries that given her data limitations, she’s not able to properly control for the fact that the repatriates happened to have chosen to migrate to departments that were having bad economic outcomes unrelated to the mass migration. Interestingly, the migrants didn’t seem to have an effect on internal migration patterns, i.e. their presence didn’t discourage the native French from migrating to departments heavily populated by the new immigrants.))

Another crumbling empire provides us with another immigration experiment: the returnados from Angola and Mozambique in the mid-1970s increased Portugal’s work force by 10%. In a study design similar to Card’s Mariel paper, Carrington and Delima find the Europe-wide recession of those years swamps out any effect the the immigrants might have had.

А русских евреев принять на нашу работу?

Нет! ((Thanks Google Translate!))

An easy criticism of “experimental” data, like that from the Mariel boatlift, is it only applies to the particular situation under which the experiment was run. Immigrants may have had no effect on employment outcomes in the very special circumstances of Miami in the early 1980’s, but in other places and times immigration would have an effect. Card, himself, says that one of the reasons the Mariels had no effect on the Miami labor market was because the city had had ample experience in the preceding decades absorbing Hispanic and, in particular, Cuban immigrants. Maybe Miami is just an extremely efficient melting pot.

The only way around this criticism of experiments is to do as laboratory scientists do, replicate them. Of course, in the economic study of immigration, this is not something that can be done in the lab. Economists studying immigration can not run their own experiments; they can not produce their own data. They have to wait for nature run experiments for them. Lucky for economists, nature provided such an experiment in the collapse of the Soviet Union.

Between 1990 and 1994, Israel’s population grew by 12% due to immigration. Most of these immigrants were Jews from the becoming-former Soviet Union. Like the Mariel boatlift, this mass migration was almost entirely due to circumstances in the sending country and at least at first, it was largely unanticipated by the native population. The instability in the Soviet Union at that time encouraged many people to leave. Israel was chosen as the destination country simply because of shared religion and open immigration policies.

When something external to the system causes a change to that system, like these particular mass migrations, economists call it an “exogenous shock”. Laboratory scientists use random assignments into control and treatment groups, for example, as their exogenous shock. The great thing about exogenous shocks is that they remove the mystery from what causes what. If conditions in the system under study do not affect the timing or magnitude of the shock, then any changes to the system that come after the shock must be due to the shock itself. Because unlike laboratory scientists they can not induce exogenous shocks to the systems they study, the challenge for economists is to look for data that was generated from exogenous shocks or to use various, and sometimes complicated, techniques to analyze data in a way that makes it look like it was generated by exogenous shocks.

The mass migration from the Soviet Union to Israel, then, was an exogenous shock to the Israel labor force. However, the occupations chosen by Russian immigrants once they entered Israel are not exogenous. Because Rachel Friedberg wanted to study the effect of the mass migration on wages within occupations — her question was “Did Russian immigrants entering an occupation cause wages in that occupation to decline?” — she was worried about the non-random choice of occupation. Specifically, she was worried that the immigrants had chosen occupations that were experiencing above average growth in wages. In this case, if immigrants were depressing wages by increasing labor supply and so competing with natives for jobs, the raw correlation would show no effect of immigration. In the raw correlation the bad effects of immigrants on wages would be cancelled out by the fact that immigrants were disproportionately choosing occupations with higher wage growth. In a hypothetical world without immigrants, then, native wages would have been higher.

To test to see if Russian migrants were choosing occupations with higher wage growth and so creating an artificial zero correlation between immigration and Israeli native wages, she came up with a clever way to see what would have happened if those immigrants did not get to chose their occupations in Israel. What she did was to assign each immigrant to the occupation they had back in Russia. She reasoned that because of training and skills accumulated over the career, immigrants would prefer to have occupations in Israel that were similar to their old ones in Russia. Because some occupations get paid more than others, that the old occupation is similar to the new one means the old occupation is correlated with immigrant wages in Israel. The old occupations, however, do not depend on the growth in wages in Israeli occupations. So assigning the immigrant to their old occupation instead of their chosen one essentially removes the problem of occupation choice that was screwing up the raw correlation in the previous paragraph. Getting rid of the choice problem means we can see the real effect of immigration on wages.

The results are surprising. The immigrants that choose occupations different from their old occupations actually had lower wages than if they had stayed in their old occupation. Maybe “choose” is the wrong word here. It may be that immigrants, with their poor language skills or lacking social networks or because of discrimination, were forced into occupations with lower wages. In fact, when you control for this downward mobility of immigrants, the actual correlation between immigration and native wages is positive! Occupations that saw a disproportionate amount of immigration had higher wage growth in the period of the study.

This potentially throws the simple model of the affect of immigration on native employment opportunities on its head. Not only can we not confirm its implications, but its implications seem to be backwards. How is it possible that immigrants could increase wages of native workers!?

¿Los cubanos del Mariel toman nuestros trabajos?


David Card’s famous paper studied the impact of the sudden arrival of 120,000 refugees to Miami in 1980. He estimates that the total labor supply in Miami suddenly increased by 7% and the Cuban work force increased by 20%.

Because labor is one undifferentiated mass doing the same tasks, with the same skills, creating the same nondescript widget with a fixed number of machines, this increase in supply induces a decrease in wages and an increase in unemployment among native-born workers. At least this is the mental model most observers at the time carried around in their heads: people conjectured that unemployment spiked to 7.1% in the summer of 1980 because of the arrival of the refugees. It was presumed that the influx of workers created other systematic problems. For example, the boatlift was said to be partially to blame for riots in black neighborhoods that killed 13 people. Also crime, and specifically the homicide rate, spiked in 1980. The Mariel’s themselves participated in a number of crimes; they were responsible for about 10% of homicides. But presumably through their bad affects on the labor market, they caused higher crime rates even among the native population.

To a certain group of economists, to which Card is a high priest, the most important question to ask of correlations like the one found between unemployment and the Mariel boatlift is the following: is there some third factor that is correlated with bad market outcomes that is also, but independently, correlated with the influx of Cubans? Because the stories being told at the time suggested the boatlift caused bad labor opportunities for native-born Americans and then other social ills, this question is particularly important. If a third factor explains the worsening labor conditions in Miami around the time of the boatlift, then that would be the cause of so many social ills and not the boatlift itself. The obvious candidate for this third factor is the deepening national and international recession in 1980 ((The recession that year was later dated by the NBER to have started in January 1980, several months before the boatlift. Its also interesting to note that a primary reason for the boatlift was a bad economy in Cuba.)).

By comparing Miami’s experience with other similar cities that were not affected by the boatlift but were affected by the general economic slowdown, Card was able to show that wages and unemployment in Miami were not affected by the boatlift. He, in effect, subtracted out the effects of the recession on the Miami labor market and got zero. This suggests that the real culprit in Miami in 1980 was the recession. The boatlift just happened to have happened at the same time the economy was contracting.

Another group of economists cares more about the mental model used to understand the Miami labor market. In the case of the Mariel boatlift, for them the most important question to ask is: what’s wrong with the simple mental model that its obvious implication — more workers leads to lower wages and higher native unemployment — broke down? Here’s a number of possibilities:

  • Workers are not an undifferentiated mass; they have different and complimentary tasks or they have different skills
  • Workers are not all producing the same good
  • The number and quality of machines is not fixed

To the degree that any of those three things are true, the implications of the simple model break down. If Cuban refugee workers have skills that are complimentary to native-born workers or if Cuban workers are making products that natives don’t make, then those two groups will not compete with each other in the labor market. Just like an increase in the number of dentists would not have an effect on the wages of construction workers, refugees with different skills from natives would not have an effect on native wages.

But even if workers are an undifferentiated mass, if the machines and production processes they use can be quickly installed or upgraded to accommodate new workers, then the refugee workers will increase the total amount produced by the Miami economy and they will not affect wages or employment of native workers.

The unemployed aren’t the only ones seeking jobs

Work with me here. In normal times, say 2004 through 2007, suppose 10% of the working population are looking for jobs while still employed (do you know of a better estimate?). This means about 13 to 14 million employed workers are “job seekers” in normal times.

“Quits” are voluntary separations from jobs and folks do that because they’re leaving the work force (e.g. retiring) or because they found another job. Now look at quit rates over the last couple of years:

Quits have declined by about 40% compared to normal times. From here, about 50% of quits are retirements. If the retirement rate has stayed the same, then quits due to job changes went from 1.1% to about 0.5%. This suggests the number of “job seekers” among the employed has gone down by at least half.

How many job seekers are there right now? Supposing all unemployed workers are “job seekers” then the total number is about 20 million people. In normal times, that number is about… 21 or 22 million people (unemployed plus 10% of the working population). By this measure, the job market now is less congested than usual!

Making the assumptions that I made above, I constructed a “job seekers” per job opening time series:

If you assume none of the currently employed workers are “job seekers” then the graph above looks like the one put up by EPI. If you assume 20% of the currently employed are “job seekers” then even the up-tick seen in the later part of the time series goes away.

Why a good “policy stance” might matter

I haven’t read John Taylor’s newest working paper yet, but its about the Fed’s start-stop policy stance under the ancien régime. The story he’s riffing on in that paper is pretty standard. There was high and variable inflation under the old monetary policy regime because the Fed didn’t fully accept its role in determining inflation. It would have moments of commitment to fighting unemployment or the output gap and so it would loosen policy. When inflation came, they’d commit to lowering inflation. They’d tighten policy long enough to induce a contraction in output, but not long enough to re-anchor inflation expectations at lower rates. Rinse and repeat. Stagflation and uncertainty.

Then Volker came and saved the day!

The Great Inflation is the time before 1985 and the Great Moderation is the time after. I’ve labeled the regime average and standard deviations.

I’m not interested in debating the reasons for the Great Moderation starting in the mid-80s, but at least one reason for the decrease in volatility in both output and inflation is that the policy stance of the Fed changed. It gained credibility as an inflation stabilizer. It was going to set an (implicit) inflation target and do what it needed to hit it.

Compare the 25 years before 1985 to the 25 years after. The old regime saw 5 recession, the new regime only three (and two of those were pretty mild). As you can see in the graph (which includes the most recent recession), volatility of output decreased. Similarly, volatility of inflation and consumption also decreased. As Bernanke pointed out in his Great Moderation speech, these trends are evident in other countries as well.

The thing that changed was the adoption of a confident and credible policy stance by Central Banks around the world. They woke up to their power to control inflation. And Bernanke was trying to wake up the Bank of Japan to its power as an effective manipulator of inflation; to build their confidence. Central Banks also learned the power of credibility. And this is why Bernanke responded to Delong’s question the way he did.

Japanese unemployment


In November, Japanese unemployment was at 5.2%.

Many folks are comparing the Fed’s action today to those of the Bank of Japan in the “lost decade” (e.g.). I can’t quite peg down the comparison being made by Sumner, but Yglesias articulated it most clearly:

[Bernanke] knows that unemployment is a problem now and he believes that he could fight it, but that fighting it more aggressively would elevate the risk of inflation in the future and he thinks that reducing the possibility of future inflation is more important than reducing the reality of current unemployment. I think that’s nuts. But it’s an attitude the Bank of Japan has consistently maintained since the 1990s.

Here’s the employment to population ratio:
Its safe to say employment wasn’t a problem in the “lost decade” and there wasn’t too much weight put on inflation relative to unemployment.