Which Way to Peace? Estimating Directionality in Conflict Event Data

• In Event: Comparative Politics Applications of Latent Variable Models

Abstract

Today civil conflict is characterized by multiple warring parties with different motivations, capabilities, and patterns of behavior. Current research on civil conflict assumes that government actors and non-state armed groups fight because governments are uncertain about rebels' willingness and capacity to continue the fight. We contend that battle initiation is a key signaling mechanism that represents any given armed group's willingness to fight. Further, we argue that a precise measurement of fighting capacity is required to investigate which actors drive changes in conflict intensity over time. Yet currently available event data does not record information about the initiating actor, it only records which two (or more) actors engaged in a battle. Most event data is based on news sources, a format that can render determining the directionality of an event very difficult. We present a novel latent variable approach that parses directionality from event data via a network decomposition approach. We apply this approach to the Nigerian Civil conflict. To validate our approach, we hand-code a portion of the events for the Nigerian case and show that our unsupervised classifier is able to parse directionality from event-based sources with a high degree of accuracy. We then run our model across all conflict based events recorded in Nigeria from 2001 to 2019 and evaluate how the intensity of conflict (in terms of the number of battle deaths and civilian victimization) changes as a function of whether or not the government is on the defensive or offensive.

Authors

• Cassy Dorff, Vanderbilt University

• Shahryar Minhas, Michigan State University