Cities, Data, and the Needs of Ukrainian Refugees
Lessons Learned from the June 2022 CrisisReady Workshop in Budapest, Hungary
Refugees do not settle in countries. They move to, and live their lives within, cities, where cultural and community connections, economic opportunities, transportation networks, and high-density services offer the best chances for a new life. Nowhere is this more evident than in the European Union today, where millions of Ukrainians have fled the Russian invasion of their home country and sought refuge in cities from Warsaw and Brussels, to Berlin and Budapest. As we in the humanitarian and research communities struggle to use new data sources and methods to understand and respond to this crisis, our attention needs to be on how to bolster the informational and analytic needs of our colleagues in city administrations and non-profit networks which are preparing to support refugees throughout what promises to be an arduous winter.
The focus on cities and novel data sources in the context of the rapidly evolving Ukrainian refugee crisis was the organizing principle of a workshop on June 27th and 28th, co-hosted by the CrisisReady collaboration between Direct Relief and Harvard University, and the mayor’s office of Budapest. Participants included 43 representatives from city administrations in Budapest and Warsaw, Hungarian non-profits, UN agencies including IOM and UNHCR, World Bank GFDRR, IFRC, companies including Meta, Boston Consulting Group, Orange, Vodafone, and Pelion, as well as academic researchers.
Whereas Hungary has been more of a conduit into the rest of Europe than a resting place for the hundreds of thousands who have traversed its eastern border, given issues of language, community connections, and immigration policy, Budapest has been a crossroads and a potential site of long-term resettlement for tens of thousands. The questions which touched off this workshop from the city’s point of view were how many Ukrainians reside within the Budapest metropolitan area, where have they tended to reside, how those neighborhood level settlement patterns may be changing over time, and what are the most important patterns of need?
Refugee Mobility Dynamics and Data Limitations
Each question is subject to constraints based on data scarcity. While one might imagine that data collected nationally on refugee transit through the country’s rail system, part of a service designed to reduce refugee transport costs, might answer some of these population dynamics issues, political differences have prevented data sharing. In a related but causally different sense, more to do with private data agreements and public analytics capacity than politics, data from mobile network operators like Vodafone has not yet been utilized within Budapest, nor many other cities, to understand refugee movements and resettlement patterns. Collaboration presented at the workshop between Orange and the French Ministry of Interior, utilizing Orange’s Flux Vision to understand changing refugee densities across France, is an exception to that rule. Refugees meanwhile are participating in Facebook groups and online support services, but data from those sources has thus far been poorly structured and cumbersome to analyze at scale.
Where mobility data has made more of a difference in the insights available to public agencies is in the uses made of data from the Data for Good program at Meta. CrisisReady began producing maps with this data showing movement anomalies associated with large-scale refugee flows at the start of the conflict, and shared them with response agencies and public authorities. Research from the European Commission’s Knowledge Centre on Migration and Demography has since affirmed the value of rapid migration analysis using these types of social media-based sources and methods.
Figure 1: Maps by CrisisReady based on mobility data from Data for Good at Meta help to visualize and understand broad shifts in refugee movement.
As refugees have diffused throughout the EU, social media-based signals on population movement anomalies, measured in terms of population density changes in the Facebook app user base relative to a pre-war baseline, have become diminishingly specific to refugees alone. Given sample size constraints, in part based on the need to remove data counts which do not meet privacy thresholds, increasing spatial diffusion makes it more difficult to use this type of mobility data for refugee movement analysis at the smaller scale of cities and neighborhoods.
Figure 2: Neighborhood-scale mobility analysis of Budapest based on data from Data for Good at Meta.
Meta’s own data science team has produced high-quality estimates of Ukrainian refugee flows across the continent, and at city level, based upon the long-term displacement cohort methodology. By measuring displacement and return of a singular cohort given the IP address of their imputed home location within a crisis-relevant bounding box, Meta’s approach produced an estimate of refugees within Budapest, as of August 15th, just over 43,000 people, an increase of more than 12,000 since the end of May. This data is available only to a select group of partner organizations who are able to proxy that information as needed to public sector partners, making establishment of analysis support agreements increasingly important.
In the long run, the answer to many of these population data limitations for humanitarian response and support for displaced people is to expand safe access to mobile network operator data for public and non-profit agencies, and their research partners. This type of data covers in general a larger sample of the population than app-based data alone. In the case of the Ukrainian refugees, there is also a possibility to subset the population specifically for refugees given that free sim card programs cover a large portion of the refugee population and allow for more accurate understanding of refugees, apart from the general population of phone users.
In order to make mobile network operator data safe and useful for humanitarian aid and public services, however, protections must be put in place specific to the appropriate temporal and spatial scale of analysis, along with standardized metrics which do not put individual privacy or community vulnerability at risk. More than a decade of research and practice, including differential privacy and federated learning, has already gone into developing the elements needed to assure that data access for specific public health and social service related questions does not compromise individual or community privacy and safety. The next step is to get companies which produce and manage this data to agree to systemic conditions for emergency release of mobility data for the sake of advancing the common good.
Data for Effective Social Service Delivery to Refugees
Beyond the questions of how many refugees have been moving, and where they may be resettling, the key areas of interest for the representatives assembled in Budapest focused on types of social service delivery required to ensure decent living standards for refugee populations.
Employment was foremost in mind at the workshop, but considered especially challenging given that most Ukrainians do not speak the local, in this case Hungarian, language, and often come to the host country with a skills mismatch between their own relatively higher level of education and training and the lower-skill occupations which are usually on offer to refugees. Nevertheless, even basic analytic services such as compiling better refugee-relevant job boards, expanding surveys to determine employer needs and capacities, and analyzing origin points in Ukraine relative to the prevalent mix of occupations to make predictions about the probable mix of skills present in the city, could all make a difference.
In Poland, one of the key health data projects presented was the Health4Ukraine program, funded by Direct Relief and administered by the Polish healthcare company Pelion. Ukrainians who have completed refugee registration in Poland qualify for a cash card to be used at any private pharmacy for prescription or over-the-counter (OTC) drug purchases. The data from this type of program informs not only understanding of the relative density and distribution of refugees based on transactional frequency, but also the types of health commodities being purchased. Given that type of data it is possible to forecast specific health needs and trends which may be reflected by purchases, and which in turn can inform humanitarian public health programming. Although a corollary to the Health4Ukraine program does not yet exist in Hungary, the examples being set in Poland are promising for future efforts across the region.
Figure 3: Data from Health4Ukraine in Poland tracks anonymized health commodity purchases by Vovoidship to inform understanding of health needs exhibited by refugee purchasing decisions, including higher rates of pharmaceutical purchasing in cities relative to more rural areas.
Over the coming months, strong data partnerships across the EU will be more essential than ever to ensure that public agencies, non-profit service providers, and UN agencies are well equipped to anticipate situations before acute crises develop, and allocate resources effectively and equitably. The private sector remains a crucial partner, and the source of much of the data needed for modeling and prediction of refugee needs. More still needs to be done to inform understanding of these use cases, methodologies, and conditions of data privacy and protection, across public, non-profit, and private organizations, in order to build on the examples which already exist for translating across interests and viewpoints, so that we don’t spend precious time debating data access policy when we could be delivering resources.
About the author:
Andrew Schroeder is the Vice President of Research and Analysis for Direct Relief. He is the co-founder and co-director of CrisisReady, a global research and digital health emergency response platform based at Harvard University and Direct Relief. He is also the co-founder of WeRobotics and the global Flying Labs Network, a non-profit organization and network of experts across 36 countries, dedicated to creating localized applications of robotics and data science for aid, health, and environmental protection.
