Sezibra/conflict-abm-simulation

Agent-Based Modeling of Conflict Dynamics in Ethiopia (2020-2022) # Agent-Based Modeling of Conflict Dynamics ## Research Question How do local retaliation dynamics between armed groups and state forces produce the spatial clustering of violence against civilians observed in the Ethiopian conflict (2020-2022)? ## Navigation | Section | Description | |---------|-------------| | Motivation | Why agent-based modeling matters for conflict research | | Key Findings | Main results across parameter sweeps, sensitivity analysis, and empirical comparison | | Data | Simulation-generated data and UCDP empirical benchmarks | | Model Design (ODD Protocol) | Agents, environment, behavioral rules, and scheduling | | Results in Detail | Figures and interpretation for each analytical step | | Limitations | Honest assessment of constraints | | Notebooks | Three-notebook analytical progression | | How to Reproduce | Setup and replication instructions | | References | Academic sources | ## Motivation Statistical models and machine learning identify patterns in observed conflict data. They work backward from outcomes to correlations. Agent-based modeling (ABM) works in the opposite direction: it specifies rules about how individual actors behave, runs a simulation, and observes what macro-level patterns emerge. This makes ABM the right tool for testing whether a proposed micro-level mechanism is sufficient to produce an observed macro-level pattern. In Conflict Event Data Analysis and Geospatial Visualization, I analyzed UCDP Georeferenced Event Dataset (GED) v25.1 data on the Ethiopian conflict and found 9 DBSCAN spatial clusters of violence against civilians, concentrated in Tigray, Amhara, and Oromiya. This project tests a specific mechanism: does local retaliation (where an attack by one side increases aggression among nearby allies of the other side) produce spatial clustering of violence similar to what the UCDP data shows? ABM has a long history in conflict research. Schelling (1971) showed that simple individual-level rules produce com …