Studying abroad is becoming increasingly common in many countries – with almost 3 million  students educated each year at the tertiary level in a country other than their own. For developing countries in particular, studying abroad offers many of the promises and fears of brain drain (both of which I think  are overblown). But understanding the causal impact is hard, because people self-select into whether or not to study abroad, and there are no lotteries or other experiments we can turn to for easy answers. Three recent non-experimental papers succeed to varying degrees in providing some convincing causal evidence.
The most convincing of the three studies is a recent paper  by Matthias Parey and Fabian Waldinger which has just appeared in the Economic Journal. They consider the impact of studying abroad due to the European Erasmus student exchange program on whether German students live abroad in the first 5 years after graduating. They find studying abroad for a year during undergraduate studies (after which they return to finish their studies) increases the likelihood of working abroad early in the career by 15 percentage points, and provide some suggestive evidence that one of the channels for this might be through meeting a foreign partner, in addition to the more work-related channels.
The Erasmus study uses instrumental variables for identification. They rely on the fact that the Erasmus program was rolled out slowly through German universities and departments within universities. Controlling for a student’s entry cohort, subject, and university, they argue that the fact that, for example, there were scholarships for political science at University X but not for economics, whereas for University Y there were scholarships for economics but not political science, was due to idiosyncratic reasons such as particular faculty connections. What is very nice about the paper is that they take threats to the exclusions restrictions very seriously, and have more than 2 pages carefully discussing possible threats to identification, checks they can do to rule these threats out, and a whole lot of sensitivity analysis. They also note that while IV allows them to only estimate a local average treatment effect (LATE), this LATE is precisely the parameter of policy interest- the effects of studying abroad for those people who only study abroad due to the Erasmus program.
A second approach is used by Oosterbeek and Webbink in a paper  just out in Economica. They consider Dutch students who apply to a scholarship program to study for year abroad of graduate study. The selection committee ranks all students, and only those whose rank is above a certain cut-off get a scholarship. This naturally leads to a regression discontinuity approach, which compares outcomes for students just above and just below this threshold. The downside is that the scholarship is pretty exclusive, so even pooling together multiple years of entrants still only gives 25 students just below the cutoff and 51 just above. They find for this group that studying abroad increases the likelihood of living outside of the Netherlands early in their career by 30 percentage points.
The identification idea is sound in this paper, but the small sample size makes it more difficult to do a number of the best practice smoothness checks around the discontinuity with any precision. Moreover, as is well-known, regression discontinuity designs only identify the treatment effect in the neighborhood of the discontinuity. In this case the sample is pretty specialized – talented Dutch students who apply for this particular scholarship, in a context where to apply for the scholarship they already have to have a definite plan of where they will study abroad, that it may be more difficult to generalize these findings.
A further challenge both these studies face is a common one in migration work – of actually being able to track migrants. Both surveys only look at people relatively soon after graduation, and tracking rates seem to still be only 51% in the Dutch survey and 25% in the German surveys. This points to the need for better systems of tracking migrants.
The third, and least convincing paper, is also the one that is likely of most direct interest to developing countries. In a paper appearing in a recent NBER volume, Kahn and MacGarvie  try and examine the impacts of the U.S. Foreign Fulbright program on knowledge creation in sciences and engineering. The Fulbright program provides scholarships to enable foreign students to come to U.S. graduate schools, but then requires that these students return to their home countries for 2 years after graduation. [Fun fact: apparently working for the World Bank or other international organizations is a loophole]. The question then is whether and how forcing people to go back to their home countries after graduate study impacts on their careers. The authors find Fulbright recipients in sciences and engineering have significantly fewer high-impact publications and overall citations, with this result strongest for people from the poorest countries – i.e. being made to go back to a poor country is a career killer.
The result seems intuitive enough, but the identification is not very convincing. The authors employ a matching approach, attempting to match each Fulbright recipient to another foreign student in the U.S. on a few basic characteristics such as ranking of PhD institution, field of study, year of Ph.D., gender, and log GDP of the home country. This is not convincing for several reasons. First, it assumes that people who got Fulbrights would have been able to study abroad if they didn’t get them – whereas a regression discontinuity based on comparing those who just miss out on the Fulbright to those who get it might be more compelling. Second, matching is on the basis of variables which themselves might be outcomes of getting the Fulbright, not ex ante determinants. Third, matching is more convincing when there is a rich set of variables to match on, which definitely doesn’t describe this case. And finally, this is a case where I would find it hard to find matching convincing – given how important this Fulbright requirement to return is, I would expect to find people self-selecting into whether they apply or not (and whether they take it up or not) depending on their desire to return home.
The return requirements of the Fulbright and other scholarship programs certainly warrant further study. John Gibson and I have studied emigration from Papua New Guinea, and find many  high-skilled individuals there who appear to have returned to PNG after studying in Australia because of a 2-year return requirement, and that few of these then seem to have subsequently left again. So I believe that these requirements may have large effects – but don’t think we know much about what the cost in terms of career prospects are of such requirements.
The impacts of policies to spur or hinder international student mobility are important to learn about, so it is great to see some papers starting to look at these issues – and to see plenty of scope for further work which builds on this. To get a broader view of new research in migration, take a look at the program  for the 4th Migration and Development Conference which was held a week ago at Harvard: lots of interesting new studies were presented.