Location Matching on Shaky Grounds: Re-Evaluating Algorithms for Refugee Allocation
MCML Authors
Christoph Kern
Prof. Dr.
Associate
Abstract
Christoph Kern
Prof. Dr.
Associate
Abstract
The initial location to which refugees are assigned upon arrival in a host country plays a key role in their integration. Several research groups have developed tools to optimize refugee-location matching, with the overall aim of improving refugees’ integration outcomes. Four primary tools are already being piloted across various countries: GeoMatch, Annie™ Moore, Match’In, and Re:Match. The first two tools combine supervised machine learning with optimal matching techniques, while the latter two rely on heuristic methods to match refugee preferences with suitable locations. These tools are used in a highly sensitive context and directly impact human lives. It is, therefore, not only desirable but critical to (re-)evaluate them through the lens of algorithmic fairness. We contribute in three key aspects: First, we provide a comprehensive overview and systematization of the tools aimed at the algorithmic fairness community. Second, we identify sources of biases along the tool design stages that can contribute to disparate impacts downstream. Finally, we simulate the application of the GeoMatch tool using German survey data to empirically illustrate the impact of target variable choice on matching outcomes. While GeoMatch optimizes economic integration, we demonstrate that the integration gains differ substantially when social integration is prioritized instead. With our use case, we highlight the susceptibility of algorithmic matching tools to design decisions such as the operationalization of the integration outcome and emphasize the need for more holistic evaluations of their social impacts.
inproceedings
ACM FAccT 2025
8th ACM Conference on Fairness, Accountability, and Transparency. Athens, Greece, Jun 23-26, 2025.Authors
C. Strasser Ceballos • C. KernLinks
DOIResearch Area
BibTeXKey: SK25a