For the millions of people around the world threatened by persecution and violence in their home countries, acceptance as a refugee somewhere else is a lifeline. But what kind of life that will be largely depends on where they end up.
A new study set to be published in Science magazine shows that location plays a huge role in whether a refugee can get a job—and successfully integrate into her new home—or not. The analysis, done by political science researchers at Stanford University, Dartmouth College, and others, is based on an algorithm they developed to help resettlement agencies assign refugees to places where they are likely to thrive.
The algorithm takes into account a variety of factors that resettlement agencies usually don’t, from age to English proficiency to the size of the immigrant community from a refugee’s country in any given location. In the US, refugees are mainly sorted depending on the capacity of different places to receive them. Switzerland’s approach is to evenly distribute refugees throughout cantons. But by weighing in specific details about the refugees and the communities where they are headed, the algorithm can place them where their skills are the most marketable.
Though it still has to be deployed on the ground, trials based on cases of refugee families that were settled through traditional methods show the algorithm would have significantly increased the chances of at least one family member to find work. In the US, chances of finding work rose to 50% with the algorithm, for the median refugee. Without it, those chances are around 25%.
These are the analysis results by location in the US. (The researchers are not disclosing the names of the locations they studied to protect the refugees.)
In Switzerland, the algorithm would have increased refugees’ employment chances to 26% from 15%. Below, the results by canton.
Here’s roughly how the algorithm works: The process starts with a series of models trained to predict where a certain type of refugee is likely to fare best based on previous cases. On the other end, the researchers input constraints, such as capacity at resettlement agencies at each location. The algorithm sorts all possibilities to determine the best destination for all cases.
In other words, the locations and the number of refugees remain the same, they are just matched more precisely and systematically. The final decision would still be made by a resettlement officer, but informed by the data trove parsed by the algorithm. “There’s a lot of factors to consider,” says Jeremy Ferwerda, one of the study’s authors. “It’s really hard for a human to juggle all of those.”
Unlike other efforts to integrate refugees, such as language lessons or job training, the algorithm could be inserted in the resettlement process pretty seamlessly—it could eventually become a piece of software installed in the computers of the officers’ already making the decisions, says Ferwerda. And that could be done at a relatively low cost, a big concern given the surge in the number of global refugees.
Aside from improving individual lives, machine matching also has the potential of helping refugees overall. The more successful refugees are where they settle, the more welcoming locals are likely to be in the future.
The researchers’ team is in talks with resettlement agencies in both the US and Switzerland to test the algorithm in pilot programs with actual cases.