Network Agency: An Agent-based Model of Forced Migration from Ukraine

Zakaria Mehrab, Logan Stundal, Samarth Swarup, Srinivasan Venkatramanan, Bryan Lewis, Henning S. Mortveit, Christopher L. Barrett, Abhishek Pandey, Chad R. Wells, Alison P. Galvani, Burton H. Singer, David Leblang, Rita R. Colwell, Madhav V. Marathe

Abstract: Individuals in social systems are embedded in collective decision-making hierarchies, such as households, neighborhoods, communities, organizations, etc. The locus of agency in such systems is dispersed across the system, and can variously be viewed as individual, distributed, and shared agency. Here we propose a general notion of network agency that subsumes these descriptions and also allows for integrating related notions, such as peer influence. In our view, the social system can be seen as a multi-layer network, where each layer corresponds to different aggregations of the underlying units, representing different kinds of perception and decision-making. We illustrate this general framework with an agent-based model of the ongoing forced migration from Ukraine. In our model, individuals perceive hazards (conflict events), but decisions to migrate are taken at the household level, where peer influence from other households in the neighborhood is also taken into account. We present this model in detail to elucidate our concept of network agency. We also calibrate the model with data on daily refugee flows and show that our model is able to estimate the scale of the daily refugee flow from Ukraine for the first two months with a Root Mean Squared Percentage Error (RMSPE) of 0.24, outperforming state-of-the-art, which had an RMSPE of 0.77. Moreover, our model also captures the daily trend of outflow with a Pearson Correlation Coefficient (PCC) of 0.98. We also perform sensitivity analysis of the model and analyze the significant parameters of the model, which in turn tells us how different agencies are significant in different contexts.