Using ERGMs to Disaggregate Displacement Cascades*

Abstract How do civilians select internal displacement destinations during conflict? Existing research emphasizes the value of cascades as a guide to making these difficult decisions. Cascades may involve civilians following people in their social networks (community cascades), people with similar characteristics (co-ethnic cascades), or the crowd in general (herd cascades). Analyses relying upon interview or regression-based methodological approaches face substantial challenges in identifying the prevalence of, and relationship between, each type of cascade. While interview-based approaches can incorporate location characteristics and movement patterns, they struggle with assessing aggregate trends. Meanwhile, regression-based approaches can assess aggregate trends, but they struggle with incorporating location characteristics and movement patterns. Exponential Random Graph Models (ERGMs) that conceive of locations as nodes in a network and movements between those locations as ties can overcome these challenges and assess aggregate trends while incorporating location characteristics and movement patterns. This paper demonstrates the utility of this approach using data from UNHCR on internal displacement in Somalia from 2007-2013. Results reveal that herd cascades only form at high displacement levels, co-ethnic cascades form at medium and high displacement levels, and community cascades form at all displacement levels. Therefore, cascades provide stronger guides for displacement-related decisions as civilians switch from following the crowd in general to following those with similar characteristics to following social ties.