It wouldn't make any sense to send a French-speaking refugee to a German-speaking town in Switzerland.
But under Switzerland's current system of placing refugees, that's a situation that can easily happen. This problem isn't unique to Switzerland, and it's not the only kind of mismatch that might happen.
The solution, says a new study from Stanford University's Immigration Policy Lab, ETH Zurich and Dartmouth College, is the creation of an "algorithm" — in layman's terms, the set of rules given to a computer that will enable it to reach a specific goal. The algorithm described in the study, published online Thursday in the journal Science, uses data to predict where a refugee — or one person in a family of refugees — has the best chance of getting a job.
It's especially important to improve the placement process now, during the biggest refugee crisis since World War II, says Jens Hainmueller, a Stanford professor and one of the study's lead authors.
"There are big questions about how you can facilitate the integration of refugees into host countries, set them up for success and make sure they become productive contributors to the host country's economy and society," he says. "It's a significant challenge for governments that are facing these increasing numbers of refugees."
Using the algorithm in the U.S. would have improved the employment rates of about 900 refugees by an expected 40 percent, the authors found. Their sample of refugees were those who arrived to the U.S. in the third quarter of 2016 (the most recent data available) who were free to be assigned to any location. They also did a separate test using data from refugees in Switzerland, finding that it would have improved refugee employment rates there by about 70 percent.
To create the algorithm, researchers entered data about refugees who had already been resettled, including their country of origin, language skills, age, resettlement location and employment status. They used that data to create a model that can predict the place within the host country where a refugee (or one person in a family of refugees) awaiting resettlement has the best chance of getting a job. Using those insights, the algorithm then makes recommendations for refugee placements that take into account limitations such as the number of available spots at each location.
"What we focus on is the probability that at least one person in the family finds a job, which makes sense from a family self-sufficiency standpoint," Hainmueller says.
And the researchers say their inability to point to any one variable as the key to determining refugees' success in finding a job seems to show that the algorithm is taking advantage of sometimes subtle interactions between variables that humans might not be able to pinpoint.
"There are some places that are just better for refugees in general. They might have stronger labor markets that make it more likely for any refugees to find employment," he says. "We also found that certain places ended up being a better fit for certain types of refugees depending on their characteristics, things like their age, their gender, their language skills or the ethnic network," says Kirk Bansak, one of the study's lead authors. He's a doctoral candidate at Stanford and a data scientist at the Immigration Policy Lab.
The idea for the algorithm came from workshops the authors had with refugee resettlement agencies in the U.S. and the Department of State about potentially improving the process of deciding where refugees are placed. (They collaborated with one agency on the study but declined to name it.)
"We had heard about all these other potential interventions, like cash assistance or training programs, but our attention very much focused initially on these [resettlement] allocations because we figured out pretty quickly that where you send refugees is a really important driver of their potential integration success," Hainmueller says.
At the end of 2016, there were 22.5 million refugees around the world, according to the U.N.'s refugee agency. This year, the U.S. will resettle up to 45,000 refugees (in fiscal year 2018) — about half as many as it admitted in 2016.
The way the system works now is that placement officers consider factors such as medical conditions, the availability of interpreters and the location of other family members in the U.S. to help determine where a refugee will live in the U.S.
For refugees who don't have existing ties in the U.S., placement officers at the International Rescue Committee, one of nine resettlement agencies in the U.S., look at factors such as employment rates and public transportation systems within cities, explains Robin Dunn Marcos. She's the senior director of resettlement and processing at the International Rescue Committee.
Marcos sees this algorithm as a potential complement to the agency's placement process.
"Many of the variables that would feed into the algorithm are things that we've been using for placement decisions," she says. "The algorithm definitely seems like a valuable addition to our current approach."
And as new data is added to the algorithm, it adapts to changing conditions, the researchers say. For example, if an agency adds data that shows newly-resettled refugees aren't getting jobs in a certain city, the algorithm will be less likely to recommend they be placed there.
Cindy Huang, a senior policy fellow at the Center for Global Development who wasn't involved in the study, says this algorithm is an example of how innovation can help vulnerable people. (One of the study's co-authors, Jeremy Weinstein, is a non-resident fellow at CGD.) And it's an improvement on other ideas she's seen that involve attempts to use existing technology, like e-learning platforms, to help refugees — but that aren't cost-effective because they weren't designed with refugees in mind.
"What the study shows is that you can improve employment outcomes, which are critical to longer-term integration," she says. "More refugees should be resettled, but this is a way to do more with the number that have already been accepted into a country."
But since the findings from the algorithm are based on historical data, she cautions that it's still unproven in a practical setting.
"To validate the findings and see how it works in the messy world, the next step is a trial to see how it performs in the field," Huang says.
Bansak and his colleagues hope to create user-friendly software and data integration that would allow resettlement agencies to use the algorithm. They'll need about $100,000 to make that happen, Bansak says.
Marcos sees a potential wrinkle in putting this algorithm into practice in the U.S.: current policies on refugee resettlement.
"When they first started looking at this, it was in the last administration when we were bringing in a much higher number of refugees," she says. "Not only has the ceiling been slashed in half, but the additional bureaucratic steps that have been put in place have slowed everything down."
Courtney Columbus is a multimedia journalist based in the Washington, D.C. area. She covers science, global health and consumer health. Her past work has appeared in the Arizona Republic and on Arizona PBS. Contact her @cmcolumbus11.
Copyright 2021 NPR. To see more, visit https://www.npr.org.