Automating integration? How algorithm-based refugee allocation can reinforce social inequality

The allocation of refugees to a resettlement location within the country of arrival is a crucial decision that can critically affect a refugee’s integration chances. For instance, taking into account asylum seekers’ characteristics could improve the chances of integration into the local labor markets. Thus, several research teams have attempted to use algorithmic matching technologies to assist in allocation decisions, with the goal to optimize the overall integration chances of arriving refugees. These technologies usually draw on characteristics of the refugees. However, optimizing only for overall integration chances, while basing decisions on refugee characteristics, is likely prone to perpetuate inequalities in integration chances between refugee subgroups. In our project, we use agent-based models to investigate potential inequalities in economic and social integration outcomes that may result from using predictions and algorithms in refugee allocation. To this end, we simulate algorithmic versus (current) random allocation procedures of refugees arriving in Germany over several years. The agent-based model draws on a predictive model based on refugee characteristics’ data from the IAB-BAMF-SOEP Survey of Refugees (from 2016 to 2021) and uses an algorithmic matching procedure to assign individuals to federal states based on these predictions. Using an agent-based model also allows us to simulate how changes in the local demographic populations affect the predicted integration chances of refugees arriving to Germany in later time steps of the simulation. The results show that using an algorithmic procedure may indeed increase inequalities not only between refugee subgroups, but also between federal states. For instance, women appear to become more likely to be disadvantaged with respect to labor market integration, compared to men. More generally, the findings highlight some of the potential social impacts of using algorithmic technologies for refugee allocation. This project also showcases how agent-based models can be used to study unintended consequences of using algorithmic technologies in sensitive social contexts.