Terrorism and Insurgency

What causes terrorist and insurgent groups to form? How do they evolve over time? What factors predict conflict escalation?

Introducing the Armed Group Dataset, 1970-2012

Status: Under Review

Abstract: Data limitations on the campaign histories and organizational characteristics of smaller armed groups often leads scholars to omit key information about these actors in the study of terrorism and insurgency. This risks introducing selection bias into our empirical understanding of political violence. This article addresses these gaps by introducing an original dataset on 1,202 armed groups that operated in 124 countries between 1970 and 2012. It outlines the dataset’s construction and highlights some of the new group-level variation within it using a principal component analysis. It then showcases its comparabilities to existing datasets along with a replication analysis of state sponsorship and armed group duration.  The dataset creates substantial opportunities for developing and testing new theories about terrorism, insurgency, and civil war.

Economic Shocks and Militant Formation

Status: Under Review

Abstract: Do poor economic conditions cause militant campaigns? Conventional wisdom suggests negative economic shocks should increase the likelihood of rebel campaigns and civil conflict, but existing research finds little to no support for this claim. This paper suggests these results arise for two reasons. First, scholars conflate when campaigns form and when campaigns escalate to war. Second, scholars tend to ignore militant campaigns that never intensify into civil conflicts. I argue negative economic shocks increase the probability militant campaigns initially form, but these effects tend to dissipate before a campaign ever transitions to civil war. Using original data on the timing of 944 militant campaigns between 1970 and 2007, I estimate the effect of export commodity price shocks on the probability of formation. I test the underlying mechanisms by seeing how different shocks mobilize different social sectors to militancy and how these shocks affect campaign dynamics over time. The results show shocks increase the probability of formation due, in part, to hampering the state’s repressive capacity. However, the lag time between formation and civil conflict reduces the long-term effect of these shocks. These findings advance understanding about the causes of political violence and risks of economic shocks.

Timing of Militant Violence

Status: Working Paper

Abstract: When do armed groups launch violent militant campaigns? Armed groups organize on average for 2.5 years before launching violent militant campaigns, but when and why they decide to start fighting remains relatively unclear. This paper treats violence onset as a strategic decision on the part of armed groups. I argue an armed group initiates violence when it accumulates the minimum level of resources to credibly challenge the state. The speed at which an armed group grows into a credible threat depends on its latent capabilities, its operational environment, and the interaction of these two factors. Using an original dataset on 1,202 militant campaigns, I identify different observable characteristics which can facilitate this acquisition. I then test how these factors affect the risk of violence onset using a multilevel discrete hazard analysis. The results show armed groups tend to initiate violence sooner when they have previous combat experience. Environmental conditions only matter in the absence of prior combat experience. This paper advances understanding about the causes of political violence and timing of conflict.

Splintering, Extremism, and Militant Violence (with Katy Robinson)

Status: Under Review

Abstract: Within the terrorism and insurgency literature, splinter groups have a reputation for being incredibly violent due to their extremist preferences, but few empirical tests have assessed this claim. In this paper, we address three inter-related questions to test the conventional wisdom: Do splinter groups conduct more attacks than other types of armed groups? Under what conditions is splinter violence more prevalent? And, finally, why do splinter groups behave this way? Using cross-national organizational data on 1,202 armed groups, we show that splinter groups conduct more attacks than non-splinter groups, but only across countries. Within countries, splinter groups are no more violent. Additionally, splinter attacks are no more prevalent than non-splinter attacks around opportunities for spoiling new peace agreements or outbidding in fragmented conflict environments. To explain why splinter groups are only sometimes more violent than other armed groups, we develop and test two competing mechanisms. On the one hand, splinter groups may be more ideologically extremist, resulting in more violent campaigns to intimidate opponents. On the other hand, splinter groups may be organizationally stronger due to combat experience, resulting in lower organizational costs to fighting. Using genetic matching techniques, we test these two stories and show that organizational capacity, not extremism, drives splinter violence. These results advance understanding about political extremism and the consequences of splintering for armed conflict.

Conflict Contagion and Militant Mobilization (with Lindsay Hundley)

Status: Working Paper

Abstract: Do civil wars in neighboring countries increase the risk of civil conflict at home? Despite some evidence of contagion effects from the Arab Spring and the Color Revolutions, scholars still disagree over how and even whether militant violence spreads. We argue this debate exists, in part, because of a lack of fine-grained data about lower-level militant campaigns, which had the potential to escalate into national revolutions. This paper develops a new theory to explain both when and why political uprisings spillover by disaggregating along conflict intensity. We argue that contagion effects increase the likelihood that armed groups mobilize to challenge the state, but state reactions minimize the escalation of these conflicts. The paper derives a series of observable predictions about under what conditions contagion effects are most likely to emerge and test these hypotheses on an unprecedented, cross-national dataset of approximately 1,200 militant campaigns between 1970-2012.

Threat Detection and State Responses

Why do states prioritize some militant threats over others? How do states identify emerging militant threats? What factors predict state response?

Credit: Sgt. Ryan S. Scranton

Uncertainty and Civil War Onset

Status: Under Review

Abstract: Why do some armed group campaigns escalate to civil war, while others do not? Only 27% of campaigns between 1970 and 2012 ever became violent enough to surpass the threshold commonly used to demarcate “civil conflict.” I develop a theory that argues this variation occurs because of an information problem. States neutralize potential civil war threats on the basis of observable characteristics about an armed group’s prospective strength, but two scenarios make it harder to get this decision right, increasing the risk of civil war. I identify a set of group-level risk indicators for civil war and apply machine learning methods to test the predictive ability of these indicators. The results show observable information poorly predicts escalation to civil war in strong states, but not weak states. Further, less visible campaigns are more associated with civil war. These findings advance understanding about why civil wars begin and the effect of uncertainty on conflict. 

Agents of Subversion? Subnational Analysis of State-Sponsored Terrorism

Status: Working Paper

 Why do states provide external support to some armed groups, but not others? Between 1970 and 2012 29% of the armed groups operating around the world received external support from at least one sponsor state. Sponsor states have unusual discretion in choosing whom to support because there are often multiple militant groups operating in the same target state. However, incentives for a militant group to misrepresent its preferences in these multi-actor environments can create uncertainty about its suitability for external support. This paper builds on insights from principal-agent theory to explain how a sponsor state navigates this potential adverse selection problem and how it strategically decides whom to support. I test these predictions on an original dataset about external support for armed groups using a combination of fixed effect regressions and classification methods. I show two types of shared ideological preferences and historical support for ideologically-similar militants drive selection. This paper advances understanding about the strategic logic of external support in multi-actor environments.

Signal and the Noise: Threat Assessment for Terrorism and Insurgency (with Katherine Irajpanah)

Status: Working Paper

Abstract: How do policy-makers evaluate the threat of emerging militant groups? Existing explanations predict policy-makers rely on costly signals to guide threat assessments, but emerging militant groups often have incentives to misrepresent their strength, producing noisy signals instead. We develop an alternative argument that policy-makers assess the risk of emerging militants groups based on two factors: memory of comparable cases and relative bandwidth capacity. We identify different international and domestic events which shape these factors and measure their prevalence using an unsupervised text analysis of declassified intelligence estimates. We further process trace the mechanism through a case study of the 1979 Herat Rebellion in Afghanistan, where Soviet and American policy-makers reached opposing conclusions about the same militant group. Our findings advance understanding about the effect of uncertainty on threat perception.

Other Papers

Uncertainty Trade-Off: Reexamining Opportunity Costs and War (with William Spaniel)

Status: Published (International Studies Quarterly, 2019)

Abstract: Conventional wisdom about economic interdependence and international conflict predicts increasing opportunity costs make war less likely. But some wars occur after trade flows grow. Why? We develop a model that shows a nonmonotonic relationship exists between the costs and probability of war when there is uncertainty about resolve. Under these conditions, increasing the costs of an uninformed party’s opponent has a second-order effect of exacerbating informational asymmetries about that opponent’s willingness to maintain peace. We derive precise conditions under which war can occur more frequently and empirically showcase the model’s implications through a case study of Sino-Indian relations from 1949 to 2007. This finding challenges how scholars traditionally believe economic interdependence affects the probability of war—instruments like trade do not solely mediate incentives to fight through opportunity costs.

High Valuations, Uncertainty, and War (with William Spaniel)

Status: Published (Research and Politics, 2021)

Abstract: Many theories of war predict conflict becomes more likely as a state increasingly values the prize at stake. This article showcases an important limit. If — as in many cases — a state has uncertainty over its opponent’s material cost of fighting, then increasing the opponent’s valuation can decrease the probability of war. Why? Uncertainty condenses the various types’ reservation values, reducing the peace premium, and incentivizing a proposer to make safer offers. We also recover an analogous result under some conditions with uncertainty over power. The results indicate that higher valuations of the prize do not have a clear-cut relationship with the probability of war.

Recurrent Neural Networks for Conflict Forecasting

Status: Under Review

Abstract: Can history predict the escalation of future violence? This research note evaluates the use of a Recurrent Neural Network (RNN) for the Violence Early Warning System (ViEWS) Prediction Competition. Existing research on civil conflict shows violence is a persistent and recurring process, often shaping the direction of future conflicts. Building on this insight, I build a RNN model to examine how well historical patterns in conflict predict long-term trends. A RNN is a simple, but powerful machine learning tool for time series forecasting due to its capacity to learn long sequences of information. The results show that the model produces relatively accurate forecasts in weak and failing states, consistent with existing research on “conflict traps.” The model struggles to predict new civil conflicts, consistent with informational theories of conflict onset. The results provide important lessons for conflict forecasting and demonstrate opportunities for RNN applications in future political science research.