Explaining and predicting the adoption, timing and location of terrorism in civil conflicts

Abstract

This thesis examines the adoption, timing and location of terrorism in civil conflicts. It analyzes the effect of territorial advances on the timing and location of state-based and one-sided violence, the effect of battlefield dynamics on the timing and location of terrorism, and the predictive power of commonly used independent variables explaining rebel use of terrorism. The thesis consists of three papers.

The first paper addresses the timing and location of state-based and one-sided violence in civil conflicts. Territorial control is central to civil conflict dynamics because it is a precondition for successful insurgency and determines how rebels interact with civilians and the government, their recruitment and funding structures, and ultimately how successful rebels are in challenging a government. While strategic locations such as natural resources or infrastructure nodes have been argued to increase conflict intensity, it is less their mere presence but more the active struggle over controlling these territories that increase conflict intensity. Using recent advances in estimating local territorial control in civil conflicts with spatial and temporal variation on the monthly level, I argue that momentum in the form of territorial advances determines the timing and location of both clashes between a government and rebels and violence against civilians. I use spatial regression models and fine-grained grid cell data for Nigeria from 2009 until 2017 and find evidence for the hypothesized relationship.

The second paper addresses the timing and location of terrorism in civil conflicts. Previous studies identified battlefield dynamics, and relative rebel losses in particular, to determine the timing of terrorism and which first-level administrative regions are more likely to experience terrorism. By changing the unit of analysis to the grid cell-month, I provide a localized approach to explain and predict both the timing and location of terrorism in civil conflicts using data on civil conflicts in Africa between 1989 and 2013. I argue that battlefield dynamics in the form of number of clashes and their intensity determine the timing and location of terrorism in civil conflicts. I further distinguish between three distinct types of terrorism by target choice which indicate the intended shock level. Using generalized linear mixed effects models, I find evidence for the hypothesized relationship with regards to the number of clashes while there is evidence only for a distinct terrorism target choice with regards to rebel and government casualties.

The third paper introduces the idea of data-driven theory building by testing the predictive power of previously identified explanatory variables. Previous studies used null hypothesis significance testing and predicted probabilities to determine explanatory power but this approach can deliver misleading results. One of many advantages of prediction is to test how well theories perform on unseen data. After all, theories should be able to both explain and predict well. The aim of this paper is to provide the foundation for a unified theory of rebel use of terrorism in civil conflicts by determining the relative importance of individual variables, illustrating their substantive effects, and exploring interaction effects present in the data as a complement to more traditional theory building. I use random forests, out-of-sample cross-validation, and interpretable machine learning techniques to analyze available data for rebel group use of terrorism in civil conflicts in Africa, the Middle East and Asia. Results suggest that some variables seem to be of crucial importance while other variables seem to have limited predictive power.