Analytics for Refugee Resettlement

Each year, tens of thousands of refugees are resettled in dozens of host countries around the world, including the United States. Growing evidence shows that the initial placement of refugees profoundly impacts their lifetime outcomes such as employment, housing, education, and health. Therefore, ensuring the best initial refugee placement is critical for social, economic, and humanitarian perspectives.

Narges Ahani, Foisie Business School alumna (MS OAM 2018) and present Data Science PhD candidate, together with her advisor, Professor of Operations Andrew Trapp, joined an international team of researchers to develop a computational tool to help humanitarian aid organizations improve refugees’ chances of successfully integrating into a new community. Known as Annie™ MOORE (Matching and Outcome Optimization for Refugee Empowerment), named after Annie Moore, the first recorded immigrant entering the U.S. through New York’s Ellis Island in the late 1800s, the tool integrates machine learning and optimization to generate data-driven, optimized recommendations on refugee placements. Annie™ was developed in close collaboration with U.S. resettlement agency HIAS (founded as the Hebrew Immigrant Aid Society) and has been deployed there since May 2018.

Formerly, matching refugees was largely a manual process with multiple sources of inefficiencies, motivating the development of Annie™. For one, estimating welfare outcomes for refugees across communities is a challenging endeavor. There is also complexity in keeping in mind all of the factors involved in the resettlement process and manually matching refugees into the communities while respecting refugee needs and community capacities.

Societal integration is paramount for positive refugee outcomes. Although many indicators have been proposed to evaluate successful integration, the only widespread integration indicator available in the U.S. is the (binary) refugee employment status, as measured 90 days after arrival. Annie™ derives signals from past refugees’ placement and outcome data to learn predictive models for estimating the employment probability of arriving refugees throughout the HIAS network of communities. These estimates are then used as refugee community–match quality scores in an optimization model to find the optimal refugee placements by maximizing the total expected number of employed refugees.

Algorithmically driven match recommendations are useful only to the degree to which they can be carefully evaluated by decision makers. With such high-stakes decisions involving vulnerable people, decision makers ought to be empowered in the matching process to have control over tool-generated recommendations. This is the purpose of Annie™: an interactive decision support tool designed for resettlement staff to interact with all aspects of the problem context.

Ahani, Trapp, and colleagues co-authored “Placement Optimization in Refugee Resettlement,” which recently appeared Operations Research[1], in which they reported that backtesting indicates that Annie™ can improve short-run employment outcomes by 22% to 38%.

Annie™ is a successful demonstration of how technology can deliver significant value to the economic and societal outcomes of vulnerable populations, such as refugees. While promising, Annie™ is but one of many opportunities that Trapp and PhD students are tackling to bring advanced analytics and fact-based, data-driven insights to bear upon improving outcomes for complex and challenging decision problems in our society.

Narges Ahani, WPI Foisie Business School alumna (MS OAM 2018), is a PhD candidate in Data Science. Her advisor is Foisie Business School Professor of Operations Andrew Trapp.

[1] https://pubsonline.informs.org/doi/10.1287/opre.2020.2093 , also https://project.nek.lu.se/publications/workpap/papers/wp18_23.pdf