Just a brief interruption to announce some breaking news:
A mere 17 years after releasing Where is the Love?, Fergie has finally decided to put her money where her mouth is. She’s going to take her own question seriously and has enrolled herself for a Ph.D. in Migration Studies.
Because she knows that there is really only one hump worth talking about:
The migration hump, of course.
The ‘migration hump’ is a well known and very influential theory describing the relationship between economic development and migration. It is based on research that indicates that when poor countries get richer, migration from those countries increases. This may seem counterintuitive at first, but is likely to be the case because migrating costs money and so as economic development increases, more people are unlocking the funds necessary to finally make the difficult decision to leave home in search of better opportunities.
I learned about the ‘migration hump’ — and its apparent refutal — by reading this insightful, articulate article on the Dutch ‘non-breaking’ news website De Correspondent. (They’ve recently also launched an English-language, US-based version at http://thecorrespondent.com/ — definitely worth checking out!)
Hoe ik in een wetenschappelijke fittie over migratiecijfers belandde en wat ik daarvan leerde
Ik moet jullie vertellen hoe de ontkrachting van een belangrijke theorie over migratie zelf weer ontkracht werd. Dat je…
Or not to hump…
The article describes how a recently published research paper by the Mercator Dialogue on Asylum and Migration [MEDAM] had apparently refuted the ‘migration hump’ theory. In layman’s terms, they had shown the hump for what it really was: a lump of synthetic silicone, overly inflated and ultimately disappointing.
They had arrived at this conclusion because, rather than analyzing the difference in migration between rich and poor countries (measured by GDP), they had compared the countries to themselves, over time. In this way, the researchers claimed, they avoided the interference of other variables that could be clouding the results; differences between the countries that would affect both their GDP and their migration patterns (such as political inclinations or their geographic location).
Hope for the hump remains…
The article then links to a long Twitter thread by Michael Clemens of the Center for Global Development [CDR]. Clemens is not so easily swayed. He claims that the trends suggested by the MEDAM paper simply do not match the real-world migration numbers.
Clemens claims there are two main problems with the research: the data and the statistical models.
The problem with the data is relatively easy to understand. The study measured migration by counting the number of residency permits given in OESO-countries, which were then grouped by country of origin. This of course misses the considerable amount of migrants who either don’t or don’t yet have their residency permits. Instead, Clemens suggests, one should measure migration using the measures of international migration stock, publicly available through the IMF/World Bank.
The second problem is more intricate, and one I hope to also better come to grips with myself by writing this post. It’s a problem with the statistics.
The trouble with the statistics, Clemens writes, is that the study regresses two non-stationary variables against each other, leading to a so-called spurious regression. Or in other words, it’s trying to measure the relationship between two variables that are themselves changing over time. Or in other other words, it’s hard to say anything objective about how big a hump is when you’re standing on another hump yourself…and both of those humps are moving through time and space.
The normal trick for regressing two non-stationary variables against each other is to remove the time trend from the variables. This means you’re not just looking at the changes in the variables themselves but in relation to their trend, i.e. you’re looking for changes in the variable that are larger or smaller than expected based on their trend over time. If both non-stationary variables are showing unexpected behavior at the same time, then maybe, just maybe, they might be related.
The problem with this is that it removes the long-term perspective out of your analysis. All you’re doing is looking at short-term timeframes in which the change in the variables differs from the general trend. This is especially a problem, Clemens writes, because long-term, positive change (in GDP per capita, to be specific) is the very definition of development. To remove the long-term perspective out of your analysis means you are no longer measuring development (generally measured over multiple decades) but studying the effects of economic shocks (spanning at most a few years).
Which hump are you measuring, anyway?
The authors of the original article came back with a defense of their research. It’s long and I don’t understand all of it, but the basic premise of their defense is that they are, in fact, most interested in shorter-term (5–10 years) impact of GDP increase on migration. In that sense, their claim that their results refute the ‘migration hump’ theory is misleading. Their research is sound and their conclusions even valuable…but not to the question they claimed to be asking. A poignant lesson in the power of framing: of course, by writing that their results debunked a major theory, they got a lot of people talking and reading their work…but it also opened them up to some serious critique because their research wasn’t actually answering the question they were claiming it was answering.