Hidden Markov Models for Frozen Conflicts
Title: Hidden Markov Models for Frozen Conflicts
DNr: NAISS 2024/22-670
Project Type: NAISS Small Compute
Principal Investigator: Joshua Krusell <joshua.krusell@gu.se>
Affiliation: Göteborgs universitet
Duration: 2024-05-06 – 2025-06-01
Classification: 50601


Frozen conflicts are conflicts which undergo an initial period of high intensity violence terminated by an unstable peace whereby the underlying issue remains unresolved and there is an ever-present threat of violent re-escalation. Previous research has attempted to identify frozen conflicts, and their escalatory and de-escalatory transitions, through manual identification, often setting a threshold to the observed number of battle related fatalities as a classification cutoff for what may be considered an active conflict. This project takes a different approach, and instead treats a sequence of conflict-years as a markov process whereby "active conflict" and "frozen conflict" are considered latent states to be inferred from observable conflict dynamics, such as battle related fatalities or the number of conflict events within a year. This approach avoids the arbitrariness of previous coding schemes and allows for the quantification of uncertainty over whether a conflict-year can be considered frozen while also being amenable to analyses of how certain time-varying factors affect the transition probabilities between states. As an example of the last point, this project analyses the effect of arms transfers to state actors on the probability for civil conflicts to transition between active and frozen. This is done using a series of hierarchical, bayesian hidden markov models in order to contribute not only to the measurement of frozen conflicts, but also the understanding of how international factors influence the dynamics of civil conflicts.