Date of Award

Fall 2019

Degree Type

Open Access Dissertation

Degree Name

Political Science, PhD


School of Social Science, Politics, and Evaluation

Advisor/Supervisor/Committee Chair

Mark Abdollahian

Dissertation or Thesis Committee Member

Jacek Kugler

Dissertation or Thesis Committee Member

Yi Feng

Terms of Use & License Information

Terms of Use for work posted in Scholarship@Claremont.

Rights Information

© 2019 Khaled Eid


Intrastate conflict is an ever-evolving problem – causes, explanation, and predictions are increasingly murky as traditional methods of analysis focus on structural issues as precursors of conflict. Often times these theories do not consider the underlying meso and micro dynamics that can provide vital insights into the phenomena. Tactical decision-makers are left using models that rely on highly aggregated, country level data to create proper courses of actions (COAs) to address or predict conflict. The shortcoming is that conflicts morph quite rapidly and structural variables can struggle capture such dynamic changes. To address this some tacticians are using big data and advances in machine learning techniques to provide highly accuracy predictions of conflict, however these models lack explanatory power. Decision makers need a solution that can provide accurate explanations and predictions at higher frequencies and geographic granularity – essentially theory informed machine learning models. To achieve this, relational data, constructed from event data and social network analysis (SNA) is used to provide more granular and higher frequency data. Using this data and SNA, structural factors of power, parity, and satisfaction specified in Benson & Kugler (1998) and Lemke (2008) are recreated. Initial testing provides evidence that new measures capture results found in theory. Next the theoretical model was expanded using a complex adaptive systems framework to incorporate meso and micro levels of analysis. Outcomes suggest that examination conflict from all three levels of analysis provides higher explanatory power when compared to just a structural approach. Taking insights gleaned from statistical analysis, both the theory and CAS models were used in creating a classification and regression tree as well as random forest model for prediction. Results suggest that a CAS random forest model provides highly accurate temporally frequent, geographically specific predictions of conflict. Further the parsimonious nature of the model and structure of data means that tactical decision makers can make month-to-month predictions of conflict and explain why onset occurred. Further, by leveraging data heterogeneity, predictions can be made province-by-province, extracting different drivers of onset unique to each region. Tactical decision makers can create more nuanced and specific COAs better tailored to specific areas, rather then general policy. This research provides evidence that an extended approach using social network analysis and complex adaptive systems framework can provide a more detailed explanation of conflict as well and provide highly accurate, geographically specific, month-to-month predictions. The goal is to develop a theory informed, enhanced, replicable and area agnostic framework for producing higher accuracy conflict forecasts, explanations of conflict as well as more granular temporal and geographic stability predictions – aiding the move from strategic to tactical decision making.