Predicting civil war interventions

I am currently working on a project to build predictive models of civil war intervention. There are numerous studies seeking to explain third-party interventions into civil wars, but most focus on context-dependent factors and proximate causes. Furthermore, they do not tell us much about which factors are more or less important. I therefore set out to build predictive model(s) of intervention in order to aid in theory development and establish general empirical patterns of the internationalization of civil wars.

In short, I am building several models predicting who intervenes and on which side of a civil war in a global sample of civil wars from 1975 to 2017, with a total of 34,552 observations. I take an iterative approach to model training, tuning, and optimization based on Box’s loop, and I use several diagnostic tools to evaluate performance (F1-score), eliminate unnecessary features (recursive feature elimination and elastic net), and include new ones (Colaresi and Mahmood's diagnostic plot procedure). Because this is the first predictive model of intervention that I am aware of, I train and compare models using several different algorithms (random forest, SVM, XGBoost DART) to find the one that performs best. So far, the best performing model uses XGBoost DART with around 70 features, and it performs about 61% better than the benchmark model using multinomial logit.

While this project is a work in progress, all relevant files needed to produce the work so far, as well as various graphs, can be found on GitHub.

Cyber security and conflict

For years, I have been passionate about the strategic and political dimensions of cyber security, particularly as it pertains to international conflict. Before coming to UT Austin, I spent several years working on the theoretical, strategic, and policy dimensions of cyber security, resulting in various publications. My current interests revolve around questions about power, and how the information revolution has affected states' ability to coerce other states.

"Stability and international crises behavior: the effect of cyber operations" (under review, working paper available upon request)
In a project co-authored with Benjamin Jensen, Patrick James, and Brandon Valeriano, I examine the relationship between cyber operations and international crises, in onset and post-crises outcome, between disputing states. We hypothesize that the use of cyber operations should not increase the risk of crisis onset, and may even improve relations between states after a crisis has ended. We combine the cyber incident data (DCID) with the Interstate Crisis Behavior (ICB) dataset and estimate a series of statistical models of crisis onset and post-crisis outcome. We find no association between the use of cyber operations and the risk of interstate crises, and we find some evidence that operations conducted during crises are associated with better interstate relations post-crisis.
Replication material available on GitHub

"Articulating cyber regionalism" (under review, working paper available upon request)
In a paper co-authored with Miguel Gomez and Brandon Valeriano, we argue that the academic field of cyber security has not articulated the proper level of analysis in order to understand it as an international security challenge. Most studies look at cyber threats from a nation-state or global perspective, with disproportionate attention paid to the United States as an actor. Instead, we suggest that a regional approach to cyber security is the proper framework for analyzing threats. Going regional is the solution for scholars seeking to analyze vulnerability, capacity, and interests. We evaluate cyber regionalism in the Scandinavia and Southeast Asia regions.

"Mutually Assured Vulnerability: An Ecological Approach to the Study of Coercion and Power in Cyberspace" (working paper, PDF)
Despite years of debate and a growing body of scholarship, the future of cyber conflict remains unclear. The discussion has hinged on the arrival (or lack thereof) of some kind of cyber war, but the almost exclusive focus on a very specific phenomenon has stymied efforts to understand the actual object of analysis: cyberspace. This paper proposes a new study to cyber security where cyberspace is conceived of as an ecological system where the actors and the structure interact. By identifying the defining characteristics of cyberspace itself we can better understand the sources of cyber power, and thus how actors can achieve their political goals. Because cyberspace is a malleable system, the implication is that actors will attempt to change the structure itself in order to achieve strategic advantages. Coupled with asymmetric information and offensive advantage, this causes severe collective action problems between states and the deterioration of the viability of cyberspace as open system. (For the NUPI working papers series version, see here. For the most recent version, shoot me an email.)

"Cyber Security Capacity Building: Security and Freedom." NUPI Report, 2016.
The threats associated with ICT are multifaceted. The present report posits that cyber capacity building (CCB) should not be considered simply a risk management endeavor. The potential for malware, cyber attacks, and cyber crime are not the only challenges associated with the rapid spread of ICT. Policymakers must also consider the intersection of technology and politics, particularly in developing countries still transitioning into democracies.

"Den Akademiske Debatten om Cybersikkerhet" (in English, "The Academic Debate Regarding Cyber Security"). Internasjonal Politikk 71, no. 2 (May 2013): 229-240.

"Competing Academic Approaches to Cyber Security." In Conflict in Cyber Space, edited by Karsten Friis and Jens Ringsmose. London: Routledge, 2016. (NUPI working paper version)
Much has been written on the subject of cyber security over the past two decades, but there remain significant shortcomings in the literature. Cyber security is a field filled with fundamental conceptual, theoretical and empirical disagreements, making any systematic analysis of cyberspace difficult. This paper seeks to categorize the various approaches, here referred to as schools of thought, to cyber security and identify the sources of their disagreements. Much of the academic and policy debate has revolved around the “revolutionist” and “traditionalist” schools of thought, with debates over cyberwar and the efficacy of strategic information warfare. However, none of the schools offer a systematic approach to understanding the strategic utility of cyberspace. This paper identifies a third, less known approach that is best described as “environmentalist.” The “environmentalist” school’s approach to cyber power and analysis of cyberspace as a distinct environment or system offers the best way forward for the field.

War and conflict

Picture courtesy of Erwin Franzen.

Intervention, war expansion, and the international sources of civil war
My dissertation project develops an international theory of civil war and examines when and why civil wars become international conflicts. It resulted in one published paper, as well as a new dataset and a related working paper.

"Intervention, war expansion, and the international sources of civil war" (OnlineFirst in Conflict Management and Peace Science, PDF, appendix)
Why do some civil wars turn into interstate wars? I analyze an asymmetric information model of civil war onset, rebel-sided intervention, and interstate retaliation with endogenous stakes. Interstate war occurs when rebels believe the threat of intervention will compel the government to acquiesce, the third party believes the government will tolerate an intervention, but they both underestimate the government's resolve. The model also has implications for civil wars. Retaliation can deter intervention and rebellion, but intervention can either deter rebellion or compel the government into giving up power, depending on whether the rebels are susceptible to foreign influence.

"Intervention at your own peril: patterns of war expansion and retaliation" (PDF)
Most civil wars attract external intervention, with rebels often receiving external support from other states. However, interventions are not without risks. Domestic governments can and do retaliate against third-party rebel supporters, sometimes expanding the civil war into interstate war. In this paper, I present the Civil Wars Expansion Dataset (CWED), which captures various types of retaliation by domestic governments against external rebel supporters. It is the first global dataset of retaliation, and covers all civil wars with rebel-sided intervention in the period 1975-2009. Using this data, I estimate a statistical model of civil war intervention and retaliation, using urban population as a proxy for escalation costs from war expansion. My statistical analysis shows that countries with higher urban population are more likely to attract intervention and less likely to retaliate. Similarly, third parties with larger urban populations are less likely to intervene than stay out of a civil war.

Data collection: Civil War Expansion Dataset (CWED)
As part of the dissertation project, I am also collecting data on the expansion of civil wars. CWED measures whether and what type(s) of retaliation civil war governments conduct against external rebel supporters. The Department of Government and the Clements Center on National Security and Strategy at the University of Texas at Austin have provided generous support for research assistance.

Oil discovery, oil production, and coups d’état
In a project co-authored with Curtis M. Bell and Scott Wolford (International Interactions, PDF), I explore how the discovery and production of oil affect the risk of coups d’état. We analyze a model of bargaining between government and military in which oil rents (a) increase the value of capturing the state but also (b) allow leaders to coup-proof their governments and/or appease potential plotters. These mechanisms offset each other once oil wealth is realized; incentives to topple the government are countered by the government’s capacity to thwart or discourage coups. But when oil is newly discovered and rents have not yet been realized, plotters may launch a coup before power shifts decisively against them. Coup attempts are uniquely likely in such windows of opportunity, but those same coup attempts are exceptionally likely to fail. We uncover these relationships in an analysis of coup attempts and outcomes, oil production, and oil discovery in a global sample of states from 1980-2010.