ZEIT-ONLINE: "Matching for Refugees"

The Berlin Governance Platform project Re:Match was presented in a new ZEIT ONLINE article. This is a copy of the article called "Matching for Refugees", first published on 08.12.2022, the text was written by Anja Holtschneider and all rights belong to zeit.de.

TIME-ONLINE, 08.12.22

How can refugees in Germany be better distributed? Algorithms will soon help: people say what they need and the municipalities what they offer.

Do you prefer to live in the countryside or in the middle of the city? Is there a school for the children? Are there jobs and training opportunities? These are questions that will be answered in the future by an algorithm for refugees to find a suitable place for them to live. Hannes Schammann, professor of migration policy at the University of Hildesheim, is working with a team of researchers at the University of Erlangen-Nuremberg to develop Match'In, a digital matching tool for municipalities and refugees. It will soon be tested in practice.

The need is there: 190,000 asylum seekers arrived in Germany in 2021, and by the end of October more than 180,000 people had already applied for asylum this year. In addition, there are more than one million refugees from Ukraine. A place to live must be found for each of these people. Not only do they have to find a job, but they also have to build a social network. In the future, algorithms could help distribute the refugees in Germany. And in the process, they could also promote integration if more attention is paid to the needs of the refugees.

The distribution of refugees among the federal states has so far been carried out in Germany according to the Königstein Key: based on tax revenue and the number of inhabitants, each federal state must fulfill an admission quota. North Rhine-Westphalia takes in 21 percent of all refugees, Saxony just under five percent. Each federal state has its own regulations for further distribution to the municipalities, often staggered according to the municipalities' population or economic performance. Match'In starts with this distribution in the federal states.

The new solution sounds simple: Refugees enter in an app what they would like to see in their new place of residence, and the municipalities tell them what they could offer. An algorithm compares the two and suggests the best place for the refugees based on the data. Schammann's team has been working on this for a year and a half to ensure that it works in practice. They have already defined an extensive catalog of questions that includes topics such as career, family and leisure activities and collects data about the communities, such as the schools on offer and medical care.

German states join algorithm project

Match'In has a good chance of catching on in Germany because it has the crucial partners: the federal states. Lower Saxony, North Rhine-Westphalia, Hesse and Rhineland-Palatinate are each participating in the pilot project with five to six municipalities. The aim is to test whether the algorithm can make distribution easier for the municipalities and states. "In the best case, our algorithm should change the distribution across the board," says Schammann. That's why the algorithm will be constantly refined during the pilot phase. The first distributions in the federal states will start next spring.

Elias Bender considers the project ambitious but promising. He is head of department at the Rhineland-Palatinate Ministry of Integration and is responsible, among other things, for the distribution of refugees. At the transfer center, employees look to see which municipality currently has an admission freeze, which refugees need medical care and whether relatives are already living in Rhineland-Palatinate. They currently distribute an average of 600 refugees per month.

The Match'In algorithm not only gives refugees more of a voice, but also gives staff more information about the people they need to house. Bender wants to use the project to find out if the decades-old process of distribution can be improved.

However, it is already clear that the algorithm will not distribute the refugees in the countries on its own. "The employees are still responsible for the decisions," says Bender. The algorithm is merely a decision-making aid.

But implementation is not as simple as it sounds in theory. Algorithms can discriminate - depending on how they are programmed. To minimize the risk of individual refugees being disadvantaged in the matching process due to their nationality or lack of qualifications, Schammann's research team has developed a large catalog of questions. Around 50 questions cover everything from the refugees' marital status, occupation, leisure activities and health. These include questions about whether barrier-free housing is required or whether medical care is only desired by female doctors.

Digital solutions for migration policy

It takes one and a half to two hours to answer all the questions. Schammann knows that this takes a long time. But the detailed survey is necessary in order to be able to take integration-relevant criteria into account. For this reason, the profiles of the municipalities are also extensive. This avoids cherry-picking the best specialists.

Prioritizing answers makes this even more difficult, because it makes the answers in the questionnaire less predictable. Each refugee can determine for himself or herself which aspects are particularly important to him or her. This could be an LGBTQIA+ community, for example, or a rural place of residence. The refugees thus also get a say in the process.

Starting next year, refugees in the participating federal states can voluntarily participate in Match'In. At the beginning, the questions are filled out together with a staff member at the initial reception center and entered into a software program. "We are first concerned with the content and not with a nice user interface," says Schammann. But the goal is for refugees to be able to answer all questions on their own in an app in the future, says the migration researcher. That way, they might also be more confident about answering questions truthfully. That's important, for example, so that women or LGBTQIA people get the protection they need.

An algorithm that predicts integration opportunities

Germany is not the only country searching for digital solutions to migration policy. Big data and algorithm projects are underway around the world. For example, scientists at Stanford University and Zurich University are working on an algorithm that predicts integration opportunities. GeoMatch is trained with existing data from refugees in the respective country. Using this data, the algorithm calculates the best place in the country for a newly arrived refugee to find a job. In the U.S., it would be twice as likely to get a job if the algorithm had picked the place of residence beforehand, a calculation by the universities found. So far, however, GeoMatch has not been used in practice.

The Annie MOORE project, which was developed by the universities of Oxford and Lund and has been used by a resettlement agency in the USA since 2018, is already further along. The algorithm uses the personal profile of each refugee to calculate where he or she would have the greatest chance of finding a suitable job. The universities calculate that up to 20 percent more refugees can find a job in the first three months after their arrival.

An algorithm solution for the EU

In Germany, another algorithm project has been launched, Re:Match, which aims to distribute refugees starting in March. Re:Match is based on an idea by political scientist Gesine Schwan. With the help of an algorithm, refugees are to be distributed within the EU. The refugees will have more say in the distribution and, in the best case, will be able to choose their place of residence from the algorithm's suggestions. The Berlin Governance Platform, co-founded by Schwan, has been testing the idea since September. Within a year, the NGO wants to distribute 100 to 150 Ukrainian women who have fled to Poland among ten German municipalities using an algorithm. Both Ukrainians and municipalities are participating voluntarily. Cities like Kiel are already participating, and the NGO is still looking for more municipalities.

"The goal is to create a solution contribution for the distribution of Ukrainian women in Europe," says Giulia Fellin, project coordinator of Re:Match. Actually, the EU's mass influx directive provides for an intra-European distribution of refugees. It was first activated for Ukrainian war refugees. But distribution is still unequal across Europe. Re:Match could help change that.

Re:Match uses the experience of the Canadian company Pairity for its algorithm. This company has already matched refugees with volunteer groups in Canada and the Netherlands - with positive results. An initial interim evaluation of the Dutch project shows that a third of the refugees can now speak Dutch better. For 86 percent of the refugees, matching with Dutch people has had a positive impact on their lives.

The algorithm for the Re:Match project is still being adapted, but the preferences of the refugees are to take precedence, since the data of the municipalities is static. School places are offered by every municipality. Whether a refugee prefers to live in Kiel or Rottenburg am Neckar then makes a bigger difference. "The algorithm can mitigate bias by humans in the matching process," says Craig Damian Smith, founder of Pairity.

The social network is the most important

The Canadians don't focus on the job market as the most important criterion in matching. "We focus on access to social networks," Smith says. The idea is that social networks form the basis for cohesion with host communities. This sets Pairity apart from other global algorithm projects that tend to focus on job placement.

Hannes Schammann's Match'In project also does not rely solely on work as the most important criterion in matching. More space is given to social factors such as leisure interests. A study of refugees in rural communities by Schammann and several other migration researchers found that, in addition to housing and work, a social network was a decisive factor for refugees to stay in a region. The 2021 report of the Expert Commission on Integration Capability concluded that distribution systems such as the Königstein Key were not sufficiently suitable for the integration of refugees. Too often, refugees are distributed to regions with high unemployment, where it is then more difficult for them to find work and connections.

Match'In and Re:Match address precisely this problem. Will the algorithm projects work in Germany? The chances are good.

Correction note: Rhineland-Palatinate currently distributes an average of 600 refugees per month, the previously mentioned figure was incorrect.

The Berlin Governance Platform project Re:Match was presented in this ZEIT ONLINE article. This is a copy of the article called "Matching für Flüchtlinge", first published on 08.12.2022, the text was written by Anja Holtschneider and all rights belong to zeit.de.

Screenshot of a ZEIT ONLINE article "Matching für Flüchtlinge" Re:Match - Relocation via Matching The project Re:Match was presented in a new ZEIT ONLINE article: How can refugees in Germany be better distributed? Algorithms will soon help: people say what they need and municipalities what they offer.