weapon of the market-dominant: a market theory of...

52
1 Weapon of the Market-Dominant: A Market Theory of Terrorism in Civil War Aslihan Saygili, Renanah Miles, and Laura Resnick-Samotin Columbia University Abstract How does the distribution of capabilities among rebel groups affect their use of terrorist tactics? We argue that the amount of civil wartime terrorism perpetrated by a rebel group depends on its “market share”—the group’s military capabilities assessed vis-à-vis other rebel groups in the same conflict. Market-dominant rebel groups have both the material capabilities to make proliferate and deadly use of terrorist attacks and the strategic incentives to do so as a means of maintaining their monopolistic position in the conflict. By highlighting the links between relative strength, competition, and tactics, we add nuance to the conventional wisdom that conflicts involving multiple rebel groups are inherently more prone to terrorism. Using data on 342 non-state actors involved in internal armed conflicts between 1970 and 2011, we find that rebel groups who militarily dominate the conflict market use terrorism more extensively. Controlling for rebel strength vis-à-vis the government, terrorism appears to be a “weapon of the market-dominant” among non-state actors involved in civil conflict. Our findings demonstrate the explanatory power of inter-rebel distribution of capabilities for predicting terrorism in civil war and make an important modification to outbidding studies in the terrorism literature.

Upload: others

Post on 05-Oct-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

1

Weapon of the Market-Dominant: A Market Theory of Terrorism in Civil War

Aslihan Saygili, Renanah Miles, and Laura Resnick-Samotin Columbia University

Abstract

How does the distribution of capabilities among rebel groups affect their use of terrorist tactics? We argue that the amount of civil wartime terrorism perpetrated by a rebel group depends on its “market share”—the group’s military capabilities assessed vis-à-vis other rebel groups in the same conflict. Market-dominant rebel groups have both the material capabilities to make proliferate and deadly use of terrorist attacks and the strategic incentives to do so as a means of maintaining their monopolistic position in the conflict. By highlighting the links between relative strength, competition, and tactics, we add nuance to the conventional wisdom that conflicts involving multiple rebel groups are inherently more prone to terrorism. Using data on 342 non-state actors involved in internal armed conflicts between 1970 and 2011, we find that rebel groups who militarily dominate the conflict market use terrorism more extensively. Controlling for rebel strength vis-à-vis the government, terrorism appears to be a “weapon of the market-dominant” among non-state actors involved in civil conflict. Our findings demonstrate the explanatory power of inter-rebel distribution of capabilities for predicting terrorism in civil war and make an important modification to outbidding studies in the terrorism literature.

Page 2: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

2

1 Introduction

How does rebel strength relative to other rebel groups influence the use of terrorism in

civil war? One of the key points of consensus among terrorism scholars is that terrorist tactics are

primarily employed by “militarily inferior groups who lack other means of coercing the target

government” (Crenshaw 1981, 387). There are two important puzzles, however, that the

conventional wisdom about rebel strength and terrorism fail to explain. First, the vast majority of

armed non-state actors are militarily weaker than the governments they fight, yet we see

significant variation in the extent to which rebel groups rely on indiscriminate attacks against

civilian populations.1 Second, there are many examples of militarily advanced rebel groups who

are notorious for their extreme use of terrorism alongside conventional or guerilla tactics. In East

Africa, for instance, Somalia and Kenya have suffered scores of high-profile terrorist attacks

perpetrated by al-Shabaab, a powerful rebel organization that controls large swaths of territory in

central and southern Somalia and even managed to capture parts of the capital city of Mogadishu

during its peak in 2011. While examples like al-Shabaab do not falsify the common notion of

terrorism as a “weapon of the weak,” they indicate important gaps in our understanding of the

relationship between rebel military capabilities and terrorism.

This study suggests that a closer examination of the ways that the distribution of rebel

capabilities in civil conflict influence rebel tactics will fill in some of these gaps. Although a

growing literature explores the strategic logic of terrorism in civil war (Fazal 2017; Findley and

Young 2012a; Fortna 2015; Stanton 2013; Thomas 2014), the relationship between inter-rebel

distribution of capabilities and terrorism remains poorly theorized. Many empirical studies of

civil wartime terrorism either neglect military capabilities altogether or conceptualize rebel 1 In this study, we use Fortna’s (2015, 522) definition of terrorism as intentionally “indiscriminate violence against public civilian targets to influence a wider audience.”

Page 3: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

3

strength in strictly dyadic terms relative to the government. This gap in the literature is puzzling

given that “outbidding” competitors is one of the key strategic objectives that terrorism is

considered to serve (Kydd and Walter 2006). Surprisingly, the empirical literature on the

relationship between rebel competition and terrorism also makes minimal reference to relative

capabilities, failing to acknowledge that how rebel groups perceive and respond to potential

competition depends largely on their military strength compared to other actors.

In this article, we posit a more complex relationship between rebel strength, competition,

and terrorism than previously assumed. Our main theoretical proposition is that the amount of

terrorism perpetrated by rebel groups varies their “rebel market share,” defined as the share of

available military resources that they command relative to other groups. These resources

represent the troops, weapons, and materiel that enable rebels to fight. We argue that, as a rebel

group gains a larger share of the resources available in the conflict system and approaches

market domination, both its capacity and incentives for profligate and deadly use of terrorism

increase. Market domination introduces the burden of maintaining a monopolistic position and

preventing potential competitors from arising, which incentivizes the use of terror as a means of

exerting authority. We refer to this dynamic as the “monopoly effect.” Monopoly effects also

create structural conditions that may exacerbate non-strategic use of terrorism; for example, by

reducing a group’s ability to gauge public support for its choice of tactics. At the same time,

additional military capabilities may increase the ability to produce terrorism as well as the

willingness to accept the risks inherent in its use. We call this dynamic the “wealth effect.” The

outcome is larger production of terrorism by militarily superior rebel groups who dominate the

conflict market.

We test our hypothesis using new data on terrorist attacks perpetrated by all rebel groups

Page 4: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

4

that were operational between 1970 and 2011, which we compile by matching actors from the

Non-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan 2009; 2013) with those

from the Global Terrorism Database (GTD) (LaFree and Dugan 2007). Our dataset introduces an

original measure of rebel market share that we operationalize by calculating the proportion of

troops that each rebel group controls relative to all rebel troops active in the conflict year. Using

zero-inflated negative binomial estimations, we find strong support for our hypothesis: rebel

groups with a larger market share carry out a significantly higher number of attacks and are

responsible for significantly more deaths from terrorism. The finding is robust to alternative

specifications, demonstrating the explanatory power of inter-rebel power differentials for

predicting terrorism in the context of civil war.

Our study makes several important contributions to the scholarship on terrorism in civil

war. First, we show that the relationship between military strength and terrorist tactics in the

context of civil war is more complex than traditional theories can account for, with terrorism

appearing to be the weapon of the stronger in a given conflict market. Our empirical findings

suggest that, in a given distribution of power among rebel groups fighting the same government,

it is the market dominators and monopolists who make profligate use of terrorist attacks. These

powerful rebel actors not only have the weaponry, manpower, and logistical access to urban

centers that facilitate their execution of deadly attacks, they also have incentives to use terrorism

to deter competition from rival groups or internal factions that emerge from within.

Second, by highlighting the link between relative capabilities and competition, we add

missing nuance to the conventional wisdom that civil wars that feature multiple rebel groups are

naturally prone to increased terrorism. Multiparty conflicts characterized by a roughly even

distribution of capabilities among rebel groups differ significantly from those where one group

Page 5: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

5

militarily dominates the conflict landscape. While the former may engender competition and

incentives to use terrorism, the latter is a doubly dangerous context where dominant groups

possess both the material capabilities and motives to employ terror extensively.

2 Literature Review

The relationship between rebel capabilities, competition, and choice of coercive tactics

has received widespread attention across literatures on intra-state violence. Studies of terrorism

and civilian targeting offer insights about how militarily disadvantaged parties use attacks

against civilians to compensate for their comparative weakness (Crenshaw 1981; Downes 2008).

Hultman (2007, 206) argues that killing civilians serves as “a militarily cheap and easy strategy

to raise the government’s costs for standing firm” when battlefield losses weaken rebels’

coercive capacity. Others suggest that military losses erode popular support for rebels and

increase defection rates, creating incentives to intimidate civilians through selective targeting

(Kalyvas 2006) or indiscriminate violence (Wood and Kathman 2015).

While some studies treat terrorism and other forms of civilian targeting interchangeably

(e.g., Eck and Hultman 2007; Wood 2010), we differentiate between them, following Stanton’s

(2013, 1009) argument that “different forms of violence are associated with different strategic

objectives.” Terrorist attacks target civilians randomly to coerce a different audience, usually the

government (Fortna 2015, 4). Other civilian targeting, in contrast, often seeks to coerce the

civilians themselves, whether to induce collusion, deter defection, or extract resources.

For its part, the terrorism literature takes the relationship between military weakness and

terrorist violence almost as a truism. Widely characterized as a “weapon of the weak” (Crozier

1960), terrorism is seen as an attractive strategy for non-state actors who seek political

concessions yet lack the physical capacity to inflict material costs large enough to induce a shift

Page 6: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

6

in government behavior. Polo and Gleditsch (2016), for instance, argue that rebel groups who are

sufficiently equipped to rely on a conventional military strategy have few incentives to employ

terrorism, given its low military impact against the target and potentially counterproductive

effects such as legitimacy costs. For insurgents in a position of strength, terrorism is thus

considered “unnecessary—and even wasteful” (Thornton 1964, 89).

While the “weapon of the weak” argument has found empirical support, it fails to fully

capture the interplay between rebel military capabilities and strategic incentives for terrorism.

Because of the widespread assumption that capabilities only matter insomuch as their inferiority

makes terrorism a necessity, most empirical studies of civil wartime terrorism conceptualize and

measure rebel strength vis-à-vis the government (e.g., Fortna 2015; Polo and Gleditsch 2016;

Stanton 2013). This dyadic measure fails to capture meaningful variation in military capabilities

across rebel groups, as most groups are significantly weaker than the government forces they

oppose. Around 94% of rebel group-year observations in our dataset on civil wars that took place

between 1970 and 2011 are coded as “weaker” or “much weaker” than the government.2 As

such, these studies fail to account for the variation in tactics across the bulk of rebel actors who

are outnumbered by government forces yet sufficiently equipped to fight a civil war.

The literature on inter-rebel competition and terrorist outbidding highlights another

problem with treating government forces as the sole basis of comparison when examining the

link between rebel capabilities and tactics. Not all terrorist attacks are meant to coerce the

government: in multiparty conflicts, rebel leaders may see terrorizing civilians as an effective

instrument to eliminate competition. When rival organizations threaten their access to material

support from the local population, rebel groups may use terrorist attacks to signal their

2 We use the NSA dataset measure of relative strength. Of the 342 rebel groups in our dataset, only six surpass the government forces in strength at any point in the conflict.

Page 7: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

7

commitment to the cause and attract support away from the competition (Kydd and Walter

2006). Outbidding theories predict an increase in high-profile terrorist attacks on the

government’s civilian constituency when armed groups compete for popular support and recruits

(Bloom 2005; Oots 1989). Terrorist outbidding is hypothesized to help groups secure local

support through attraction rather than intimidation; the assumption is that aggrieved people want

to be represented by “zealots” (Kydd and Walter 2006, 76-78). As Bloom (2005, 78) puts it,

terrorism becomes “the litmus test of militancy” among rival groups vying to represent the

population.

In environments where outbidding may occur, we expect relative strength to be a key

determinant of how each group perceives and responds to threats posed by its rivals. However,

most empirical studies operationalize rebel incentives for outbidding with the number of non-

state actors party to the same conflict, treating competition as an inherent characteristic of

multiparty civil wars (e.g., Findley and Young 2012b; Polo and Gleditsch 2016; Stanton 2013).3

This overlooks the fact that each rebel group is likely to perceive threats posed by potential

competitors differently, depending on where it stands in a given distribution of power. Yet we

cannot accurately assess the strength of a group’s incentives to outbid competitors—or the

conditions under which they translate into increased use of terrorism— without accounting for

the group’s relative conflict standing.

In sum, extant theories of terrorism in civil war fail to fully elucidate how rebels’ military

capabilities influence their use of extreme tactics in civil war. The weapon of the weak thesis

invokes a narrower definition of rebel strength vis-à-vis the government, while studies of inter-

3 An exception is Nemeth (2014, 355), who finds that competition may reduce terrorism, but that the relationship is contingent on the state’s acceptance of violence and the group’s ideology. However, his measure of market share is based on ex post outputs (i.e., number of attacks) rather than the ex ante distribution of capabilities that we measure here.

Page 8: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

8

rebel competition pay insufficient attention to the ways that rebel strength vis-à-vis other groups

shapes their tactical preferences. Next, we present a novel account of civil wartime terrorism that

seeks to bridge these theoretical gaps.

3 Theory: Rebel Market Share and Terrorism

We argue that rebels’ use of terrorism is affected by the distribution of capabilities among

all rebel groups active in an ongoing civil war. We conceptualize these groups as operating

within a rebel conflict “market.” The market consists of available military resources, including

the troops, materiel, and arms that generate fighting power. Market share thus captures the inter-

rebel distribution of material power: the group with the largest share is the most powerful rebel

actor in a given conflict. This conceptualization draws on economic models of civil war that

characterize conflict as “a kind of ‘industry’ in which different ‘firms’ compete by attempting to

disable opponents” (Hirshleifer 2001, 331). Unlike economic markets, however, conflict markets

are zero-sum and value is redistributed by force (Hirshleifer 2001, 331-333).4

Our main proposition is that in the context of civil war, terrorism is employed more

intensively by rebel organizations that command a larger portion of the military resources

available in the conflict market. Increasing market share affects the use of terrorism in several

ways. As groups amass market share, their incentives for using terror as a means of exerting

authority and deterring potential challengers increase through a “monopoly effect,” which may

also increase nonstrategic use of terrorism. At the same time, their capacity to commit attacks—

and willingness to incur the costs of doing so—increases through a “wealth effect.”

4 Formal contest models of conflict adopt this terminology to examine the effects of combatant strength on war outcomes. In these models, armed groups produce violence, allocate resources to defeat their adversaries and, if they win, appropriate the defeated parties’ resources. For an overview in the context of civil wars, see Blattman and Miguel (2010).

Page 9: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

9

3.1 Monopoly Effects

Monopoly effects occur when a rebel group controls much or all of a conflict market.

Much like a monopolistic firm in an economic market, rebel groups who attain market

dominance face pressure to maintain their monopolistic position vis-à-vis potential competitors.

This pressure creates strategic incentives to use extreme tactics like terrorism.5 Market-dominant

groups are also more likely to miscalculate in their use of terrorism, much like monopolies

produce inefficiencies in economic markets.6 First, we briefly consider how monopoly effects

may induce non-strategic use of terrorism; next, we discuss how they shape strategic incentives.

In the absence of competition, rebel leaders face a problem of asymmetric information

that limits their ability to assess the level of local support for terrorist tactics. As Nemeth (2014,

337) puts it, “The terrorism literature is replete with accounts of groups miscalculating the

amount of violence a society was willing to endure.” An example is Iraq in 2007, when al-Qaeda

in Iraq’s excessive brutality cost it the support of the local Sunni population. By reducing

groups’ ability to gauge public support—and by reducing their accountability to the public for

support—market-dominant rebel groups become less constrained by societal preferences against

terrorism. Monopolistic rebels can ignore the legitimacy costs of terrorism more easily, whether

because they cannot hear public rebuke or because they feel emboldened not to listen.

Monopoly effects also give rise to strategic incentives to use extreme tactics. When faced

with challenges to their power by rising competitors or internal factions, monopolistic and

5 Importantly, we only examine strategic reasons why groups might choose terrorism—not whether or not terrorism is effective. Some scholars find that terrorism does not “pay” in terms of concessions (Abrahms 2012) or victory (Fortna 2015), but others find that extreme tactics succeed in driving governments to the bargaining table (Thomas, 2014). We simply note that, effective or not, groups continue to use it as a tactic, thus meriting scholarly attention. 6 Vickers (1996) provides a useful discussion of the different ways that monopolies and collusion introduce economic and social inefficiencies in markets. In particular, Vickers explains ways that monopolistic markets can cause “private and social interests” to diverge (13).

Page 10: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

10

market-dominant groups may perceive terrorism as an effective means to re-exert authority and

raise barriers to market entry by other groups.7 Market-dominant rebel groups enjoy strong

military capabilities and control a large share of all available resources, which facilitates the

execution of terrorist attacks. At the same time, the threat of losing that share in the future raises

concern over preventing incipient challenges. Possession of a dominant market share coupled

with the threat of future monopoly loss thus creates a doubly dangerous situation. Importantly, a

group does not have to be the only non-state actor in a conflict for monopoly effects to occur; the

logic also applies to rebel groups that command a large market share. We expect, however, that

these dynamics will become more pronounced as a group’s market share grows.

One way that market dominators use terrorism strategically to eliminate competition is

through a combination of provocation and outbidding. As theorized in the literature, terrorism is

sometimes used to provoke government backlash, under the expectation that harsh and

indiscriminate counterterrorism responses will radicalize the local population and drive recruits

and supporters into the rebels’ camp (Kydd and Walter, 2006). This strategy also serves

monopolistic and market-dominant rebel groups’ efforts to prevent relatively weaker competitors

from accumulating power. A government backlash so brutal that only the strongest can survive

may be seen as a risky but effective way to eliminate smaller groups who would grab larger

market share if allowed to grow.

The Liberation Tigers of Tamil Eelam (LTTE) in Sri Lanka illustrates how powerful

rebel groups can outbid competitors through terrorist provocation. The LTTE emerged in 1974 as

7 Of course, terrorism is not the only tactic that market-dominant groups use to deter or eliminate competition. Rebel groups often strive to become the “only game in town” by engaging their rivals militarily, using terrorism in conjunction with conventional attacks. Another tactic, widely used by the LTTE, is to physically eliminate the leadership of rival organizations. In contrast to outbidding and provocation, which target competitors through indirect mechanisms, these tactics aim directly to destroy rivals’ organizational and fighting capabilities.

Page 11: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

11

one of many Tamil separatist groups purporting to fight for Tamil independence. In the 1970s

and 80s, relatively weaker Tamil rebel groups competed with the LTTE for recruits. The LTTE’s

early “archrival,” the People’s Liberation Organization of Tamil Eelam (PLOTE) refrained from

attacking civilians, which a U.S. intelligence assessment concluded had “given it a broader base

among Tamils than LTTE has” (Central Intelligence Agency 1986, 6). The LTTE, however,

believed that terrorism would provoke harsh government reprisals that would target Tamil

civilians, increasing support for the LTTE and wresting it away from their rivals. The strategy

worked: the LTTE’s use of terrorism provoked a massive counter-terrorism response from the

government that helped, along with direct attacks on its rivals, to wipe out the weaker groups. As

such, use of terrorism enabled the Tamil Tigers to consolidate a conflict market monopoly.8

3.2 Wealth Effects

While monopoly effects shape group motives for using terrorism, wealth effects influence

the capacity and risk acceptance needed to execute attacks. Our observation on capacity is

straightforward: for groups already inclined to use extreme tactics like terrorism, any increase in

capabilities will increase their ability to execute attacks. Powerful actors have more weapons,

soldiers, and access to urban centers, all of which facilitate the execution of terrorist attacks.

Scholars find that group size associates positively with terrorism (Asal and Rethemeyer 2008a;

2008b; Boyns and Ballard 2004). Even if smaller groups want to use terrorism, they often lack

the requisite human and material capital, essentially creating a “labor constraint” (Clauset and

Gleditsch 2012, 2) on the production of violence that diminishes as groups grow and gain

experience (Asal and Rethemeyer 2008b). Indeed, the three groups with the highest number of

terrorist attacks in a given year for which we have troop data are Shining Path (Sendero

8 At its peak, the LTTE was one of the best-equipped rebel organizations in the world and the only rebel group to boast its own army, navy, air force, and merchant marine (Mampilly 2011).

Page 12: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

12

Luminoso; SL hereafter) in Peru (1984), Communist Party of India (CPI-Maoist) (2010), and the

Taliban in Pakistan (2009). Although these groups vary in terms of geography, ideology, and

goals, they share a common feature: disproportionately large shares of the overall capabilities

across rebel groups active in the war.9

Our second point is more nuanced: as groups are able to do more, they become willing to

do more. The economics literature has a familiar term for this dynamic: wealth effect. The

wealth effect posits that as individuals’ assets increase in value, they feel more secure about their

wealth and thus are willing to spend more. In the context of civil war, this means that willingness

to produce terrorism may grow with the capacity to do so. The common wisdom is that as

capabilities increase, terrorism will decrease due to countervailing incentives. Scholars argue that

because the non-material costs of terrorism outweigh its material benefits, rational rebels will

apply all their resources toward conventional military strategies when they can (Polo and

Gleditsch 2016, 818; Wood 2014). While we agree that rebels who can hit government forces

will do so, we question whether use of terrorism is zero-sum or mutually exclusive with other

tactics.

The logic of wealth effects predicts that more capabilities represent extra capacity and

additional expendable assets, encouraging risk-acceptant behavior. Risk acceptance in this

context refers to the costs that terrorism incurs, namely legitimacy and reputation costs.

Wintrobe (2006, 186) similarly argues that an increase in the capacity of rebel organizations

affects their preferences over tactics: “[A]s wealth increases, the leader will switch to a relatively

more risky portfolio, that is, make relatively more use of extremist methods.” In a nutshell,

groups may view control of resources as a license to use all available tactics—not as a reason to

9 At their peaks of violence, CPI-Maoist and the TTP were the only active groups in their conflict while SL commanded 93% of the troops on the battlefield.

Page 13: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

13

exercise self-restraint.

In civil wars where outcomes hinge on the population’s support, groups may see the

ability to terrorize the people who depend on the government for protection as a way to continue

applying political pressure. As groups grow stronger, the threat to impose costs becomes more

credible and certain strategies of terrorism, such as attrition, become more viable. In attrition,

groups use terrorism to signal the ability and resolve to continue imposing costs in the future

(Kydd and Walter 2006, 59-60). Indeed, evidence suggests that rebels use it to “disrupt and

discredit the processes of government” (Crenshaw 1981, 387) or to ratchet up pressure on the

government to make compromises (Thomas 2014), often in tandem with guerrilla warfare.

As one of the notorious perpetrators of terrorism in civil war, Peru’s SL illustrates the

wealth effects associated with market dominance. SL, along with other left-wing groups,

emerged from the Peruvian Communist Party (PCP) in the 1960s. The group split from the PCP

in 1964 because of a disagreement over tactics; while most of the PCP’s members preferred

peaceful means of change, SL sought a violent people’s war to bring about revolution in Peru

(McCormick 1990). When SL began to use terrorism during the 1980 Peruvian elections, it was

already a comparatively strong insurgent group; in 1987, they were estimated to have between

4000 to 5000 dedicated, full-time cadres operating in the country (Central Intelligence Agency

1987). By 1990, SL had administrative control over an estimated 25 to 40% of the country and

had developed a high degree of military and organizational professionalism (U.S. Congressional

Hearing 1992, 13, 17). Rather than eschew terrorism as their strength grew and the conflict

morphed into a long-running guerrilla insurgency, SL embraced it. Between 1982 and 1999, SL

conducted an average of 121 attacks per year; in 1984 alone, they carried out 260 attacks.

SL’s profligate and effective use of terrorism owed not only to its military strength per se

Page 14: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

14

but also to its command of a large market share in the absence of strong competitors. Even after

the emergence of the Tupac Amaru Revolutionary Movement (MRTA) in the mid-1980s as a

rival group, SL’s dominant market position as well as its notoriety for widespread terror

persisted. A comparison between the two rebel groups helps illustrate this: “…overall, the

MRTA lacked the lavish finances of Sendero, and paled in most other respects as well. It had

less than 10 percent of Sendero Luminoso’s membership base, and was responsible for less than

10 percent of the number of violent deaths attributed to Sendero Luminoso” (Rochlin 2003, 73).

Our data also support this assessment. During its peak years (1989-1991), MRTA possessed only

6.5% of all rebel troops in the conflict and carried out around 100 attacks in total, while the SL

controlled the rest of the market and executed more than 600 attacks in the same period.

Taken together, monopoly effects predict that market-dominant groups will use terrorist

tactics to consolidate control and prevent emergence of competition. Wealth effects predict that

as rebel group capabilities increase, so does their ability to conduct terrorism and the willingness

to employ potentially risky strategies. Together, these insights yield our main hypothesis that

groups with more rebel market share will use more terrorism.

4 Research Design and Data

To test our hypothesis empirically, we construct a new dataset of rebel group terrorism in

civil wars, matching rebel groups in the NSA dataset with terrorist organizations identified in the

GTD data. The NSA’s inclusion criterion, based on a 25-battle-death threshold of armed conflict,

allows us to create a more comprehensive sample of rebel actors than similar studies that only

include groups involved in conflicts reaching 1000 battle deaths in a given year (e.g., Stanton

2013; Fortna 2015). We expand the NSA spells into panel data for a rebel group-year unit of

analysis, which captures periods of overlapping rebel group activity as well as changes in rebel

Page 15: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

15

strength over time. We include in the dataset all non-state actors for the years 1970–2011, with

the exception of those involved in a coup d’état, which usually have a short lifespan of a few

days, and groups identified as “irregulars,” which we exclude due to their lack of an

organizational structure. Our final dataset comprises 342 rebel groups and 2,032 rebel group-year

observations.

4.1 Dependent Variable

We use two primary measures of the dependent variable, terror attacks and terror

fatalities.10 The first measure is the annual number of terrorist incidents perpetrated by each rebel

group; the second measure is the number of fatalities from those incidents. We rely on the GTD

coding of terrorist events, selecting only those incidents that match all three of the GTD

terrorism criteria.11 We then filter the GTD dataset to include the following attack types: armed

assault, bombing, hijacking, facility/infrastructure attacks, and hostage taking (barricade and

kidnapping incidents). We exclude incidents that are not clearly perpetrated against a civilian

target (i.e., military, police, and government targets). After applying these filters, we match the

number of incidents and fatalities per year to the groups in our data.12 In most cases, we rely on

exact name matches and only record incidents whose perpetrator in the GTD exactly matches the

rebel group name in NSA.13 Our variables only capture terrorist attacks associated with specific

10 Young (2016, 3) surveys 21 journals to show that the two most common operationalizations of terrorism as a dependent variable are event counts and number of fatalities. He also shows that results are sensitive to alternative specifications, such as number of attacks versus fatalities, which is why we operationalize the dependent variable in two different ways. 11 For inclusion in GTD, incidents must meet at least two of three criteria, namely that attacks have a broader (political, social, etc.) goal, target a larger audience than the immediate victims, and fall “outside the context of legitimate warfare activities” (LaFree and Dugan 2007, 188). 12 Our primary measure follows Fortna’s (2017) coding of terrorist activity by rebel groups. Further details about our coding procedures are in the Appendix. To ensure that our results do not depend on a specific definition of terrorism, we use alternative measures in robustness tests that rely on a more restrictive set of attack and target types. Results are reported in the Appendix. 13 See the appendix for details on name-matching between the datasets in ambiguous cases.

Page 16: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

16

rebel organizations; we do not attribute unclaimed attacks to any rebel group in the target

country in order to avoid overestimating the amount of terrorism perpetrated by civil war actors.

Nearly 42% of the rebel groups (143 out of 342) in our dataset use terrorism at some point in

their lifetime, while around 62% of the civil wars in our sample have at least one rebel group

using terrorism.

4.2 Independent Variable

Our main independent variable—rebel troop share—estimates the distribution of

capabilities among rebel groups concurrently fighting the government. We create this variable

using the NSA data on the size of rebel armed forces. For each state-rebel group dyad year, we

first identify all other groups that were party to the same conflict in the given year. Next, using

the NSA estimates of rebel forces, we calculate the proportion of troops that each rebel group

controls relative to all the rebel troops active in the conflict year. For example, in 1990,

Revolutionary Armed Forces of Colombia (FARC) fought against the Colombian government

alongside two other groups, National Liberation Army (ELN) and Popular Liberation Army

(EPL). Each of these is represented as a separate rebel group-year in the dataset. Using NSA data

on troop number estimates, we estimate the FARC’s troop share at around 74% for 1990, while

the other two groups have 19% and 7% troop shares respectively.

Table 1 presents the ten rebel groups from our dataset that conducted the highest number

of terrorist attacks between 1970 and 2011. During their peak years of terrorist activity, almost

all of these groups commanded more than 75% of all rebel troops available in the battlefield;

most enjoyed complete monopoly over the conflict landscape.14 The observation that the most

14 The Taliban troop share is not calculated due to the presence of at least one other rebel group in the conflict for whom troop data are not available. Nonetheless, the Taliban is well known as a military powerful rebel actor.

Page 17: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

17

notorious perpetrators of civil wartime terrorism are market-dominant and monopolistic rebel

groups provides preliminary support for our theory. One exception to the observed trend of large

amounts of terrorism being perpetrated by groups with high market status is the ELN in

Colombia, who used terrorist attacks extensively even though its market share was significantly

constrained with FARC’s rapid growth in the early 1980s.

Table 1. Rebel Groups with Highest Levels of Terrorist Activity and Their Market Share During Peak Years

This empirical pattern holds when we examine the distribution of the dependent

variables—terrorist attacks and fatalities—across market shares. Figure 1 below displays the

distribution of the dependent variables by market share, where groups with troop share below 0.5

Page 18: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

18

are coded as “low share”; groups with troop share 0.5 or greater, but less than 1, are coded as

market “dominators”; and groups with a troop share of 1 are coded as “monopolists.”

Figure 1. Distribution of Dependent Variables by Market Share

4.3 Control Variables

We include a number of control variables that potentially correlate with both terrorism

and rebel capabilities. We control for the regime type of the government party to the civil war—

democracy—using the Polity2 score (Marshall, Jaggers, and Gurr 2013). The literature suggests

that democracies make attractive targets for terrorists due to their citizenry’s low threshold for

civilian casualties (Hultman 2012; Kydd and Walter 2006; Pape 2003) as well as certain regime

attributes, such as respect for civil liberties, which facilitate the planning and execution of

terrorist operations (Eyerman 1998; Li 2005). Regime type might also be related to rebel

capabilities; in democratic countries where counterinsurgency measures are subject to normative

and institutional constraints, rebels can maneuver more easily to recruit and mobilize supporters.

Page 19: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

19

At the same time, democracies may also offer alternative channels of political expression to the

aggrieved population, thereby hurting rebels’ ability to recruit sympathizers to their rank

(Eyerman 1998; Ross 1993; Schmid 1992).

Another control variable is log-transformed GDP per capita, which captures the level of

economic development in the target country.15 Given that material resource flows are vital for

rebels’ organizational survival, fighting a high-income target country may increase the perceived

effectiveness of particular types of terrorism, such as hostage-taking and kidnapping for ransom,

which provide the rebels a low-cost source of war funding. Operating as a proxy for state

capacity as well as economic well-being, the GDP variable also is plausibly related to rebel

strength. Higher income levels may hurt the rebel ranks; as Fearon and Laitin (2003, 80) argue,

“recruiting young men into the life of a guerilla is easier when the economic alternatives are

worse.”

We also control for factors operating at the conflict and group levels. At the conflict

level, we use a dummy variable for conflict intensity to differentiate between minor conflict

years (25 – 99 battle deaths) and major conflict years (more than 1000 battle deaths). Terrorism

may increase as violence increases across the board. Moreover, conflict intensity and rebel group

strength are likely correlated. Stronger groups are able to mount larger challenges to the

government and may have more ambitious aims, leading to conflicts being fought more intensely

(Buhaug 2006).

At the group level, we control for rebel strength relative to government. We use the

rebel-to-government troop ratio variable, which we create using the NSA troop estimates and

military personnel data from the Correlates of War Project’s National Material Capabilities

15 GDP per capita data are from Gleditsch (2002) and the Penn World Tables (Feenstra, Inklaar, and Timmer 2015).

Page 20: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

20

(NMC) dataset (Singer et al. 1972). We also control for foreign support and group ideology.

Rebel groups who receive financial and military assistance from foreign states are not only better

equipped to execute terrorist operations, but also might undertake “drastic, highly visible

campaigns” to signal to their sponsors their commitment to the rebellion (Hovil and Werker

2005, 7). Data for the foreign support variable come from the NSA dataset. For group ideology,

we include a dummy variable, nationalist-separatist, coded by Polo and Gleditsch (2016)

indicating whether the rebel group pursues a nationalist-separatist agenda. Polo and Gleditsch

argue that ethno-nationalist ideals yield a strong sense of othering within the ethnic community

that rebels claim to represent, thereby increasing the legitimacy of terrorist attacks targeting the

out-group(s) associated with the government. Other studies make the opposite prediction,

suggesting that separatists are less likely to use terrorism as they hope to gain legitimacy and

support for statehood in the international arena (Fazal 2017). Ethno-nationalist ideology may

also influence capabilities, particularly in terms of rebel recruitment. Weinstein (2005), for

instance, argues that ethnic linkages and regional identities are social endowments that help rebel

leaders make credible commitments to deliver the future benefits of joining the rebellion.

Finally, we control for the Cold War time period. Kalyvas and Balcells (2010) argue that

superpower support for both proxy state and rebel actors affected their military capacity, leading

to different warfighting strategies, or “technologies of rebellion.” Different modes of warfighting

might also have influenced rebel decisions over tactics. For example, evidence suggests that civil

war terrorism has increased since the end of the Cold War (Enders and Sandler 1999).

4.4 Model Specification

Given that our dependent variable, the number of rebel group terrorist attacks, is an event

count with overdispersion (i.e., observed variance exceeding the mean) and excessive zeros, we

Page 21: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

21

use zero-inflated negative binomial (ZINB) models to test our hypotheses. Similar work suggests

ZINB models as the appropriate method for dealing with the abundance of zeros in the data,

which Drakos and Gofas (2006a, 74) call “an intrinsic statistical property of terrorist counts.”16

ZINB models combine a negative binomial count model with a logit model that predicts

excess zeros, assuming that there are two separate data-generating processes for the zero

outcomes (Greene 1994). Essentially, this approach distinguishes between “at-risk” groups who

have some probability of using terrorism, and “certain-zero” groups who lack the incentives or

means for terrorism due to structural factors and thus never employ the tactic.17 Of course, it is

debatable whether any groups have a truly zero probability of using terrorism since they could

theoretically adopt the tactic at any time. The highly skewed distribution of our attack count

variable as well as the results from Vuong tests (Vuong 1989) suggest that the ZINB model is

appropriate, however.

We use separate sets of covariates in the inflation and count equations, because there are

theoretical reasons to believe that different variables affect the likelihood of any terrorism

occurring versus the amount of terrorism that occurs. Our approach is informed by Li’s (2005,

293) objection to employing the same variables in both parts, which “implies that the data

generation process is one and the same, apparently inconsistent with the rationale for using the

zero-inflated estimator.” In fact, our data show significant overlap between authoritarian settings

and certain-zero outcomes: 82% of the rebel groups in our dataset who have never used terrorism

operate in a non-democratic country.18 Our data also suggest that rebel strength relative to the

16 Similar studies using this method include Piazza (2011); Findley and Young (2012b); and Santifort-Jordan and Sandler (2014). 17 The intuition behind this distinction is that the set of covariates generating a zero count for at-risk groups is different from those that govern the zero-always process (Drakos and Gofas 2006a). 18 Drakos and Gofas (2006b) attribute this overlap to two potential factors: (i) non-democratic regimes’ tendency to underreport terrorist activities and (ii) structural factors that make authoritarian regimes less

Page 22: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

22

government is a potential predictor of excessive zeros. For the handful of observations where

rebel forces match or outnumber government forces, terrorist activities is near zero for all groups

except for Sierra Leone’s Revolutionary United Front (RUF) and Somalia’s Al-Shabaab.

Accordingly, we include a measure of democracy and a measure of rebel strength relative to the

government in the inflated portion of the models.

5 Results

Table 2 presents the results. In all of our models, robust standard errors are clustered by

rebel group to allow for intragroup correlation of the error terms. Model 1 tests the “market

share” hypothesis using a count of terrorist attacks by rebel group. The coefficient for our main

explanatory variable, rebel troop share, is positive and statistically significant at the 0.001 level,

providing strong support for our hypothesis. The finding demonstrates that rebel groups who

possess a larger share of all troops available in the conflict execute a higher number of terrorist

attacks. The 1.604 coefficient on troop share corresponds to an incidence rate ratio of 4.97; this

means that the incidence rate of terrorist attacks for a monopolistic group (i.e., troop share=1) is

around 5 times that of a group with near-zero troop share. A group that possesses 20% market

share has an incidence rate of 1.38, whereas a market-dominant group with 80% troop share has

a rate of 3.61.

susceptible to terrorism. They also underline that, while the ZINB model addresses the issue of excessive zeros in a “methodologically progressive” way, it cannot disentangle between these two zero-generating factors.

Page 23: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

23

Table 2. Zero-Inflated Negative Binomial Models of the Effect of Rebel Market Share on

Terrorism

In Model 2, we use the terror fatalities count as the dependent variable. Similar to Model

1, the coefficient for rebel troop share is significant at the 0.001 level. Our analysis of incidence

rate ratios shows that the rate of fatalities from terrorist attacks perpetrated by a monopolistic

group is 2.68 times that associated with a militarily powerless group with near zero market share.

Page 24: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

24

In line with our theory, both findings from Models 1 and 2 suggest increasingly more extensive

use of terrorism by rebel groups who establish military dominance in the conflict landscape.

Figure 2 displays the predicted number of attacks and fatalities at different values of rebel

market share.19 For rebels who are substantially weaker than their competitors and command

only a minuscule share of troops in the conflict market, the predicted amount of terrorist activity

is around 2-3 attacks and 7-8 fatalities. In Figure 2a, as we move along the x-axis towards the

50% threshold (the range above which hosts market dominators and monopolists), the predicted

number of terrorist attacks by rebel group more than doubles. Similarly, as displayed in Figure

2b, the predicted number of fatalities for rebel groups with 50% market share is significantly

larger than the predictions associated with groups that have a low market share. Within the %90-

100 range for rebel troop share which corresponds to groups with the highest market power, the

predicted attack count more than quadruples (>10 attacks) and the fatalities count more than

doubles (>20 fatalities) the amount of terrorist activity associated with relatively weaker rebel

groups with troop shares in the %0-10 range.

Figure 2: Predicted Number of Attacks and Fatalities by Rebel Market Share

19 Predicted attacks and fatalities are calculated with all controls held at their median values.

Page 25: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

25

The models also illustrate that that the negative coefficient for rebel-to-government troop

ratio fails to achieve statistical significance (p-values: 0.07 in Model 1; 0.22 in Model 2). This

finding suggests that the “weapon of the weak” thesis holds limited explanatory power when

applied to terrorism in the context of civil war. In line with our expectations, rebel strength

relative to potential competitors appears to be a much stronger predictor of civil wartime

terrorism than rebel strength relative to government. The other control variables behave largely

as predicted. Rebel groups involved in conflicts with more intensive fighting, as well as those

operating in democratic countries, engage in higher levels of terrorist activity, although the

democracy variable is statistically significant only in Model 1. On the other hand, rebel groups

with nationalist-separatist ideology carry out significantly fewer attacks. The coefficients for

GDP per capita, foreign support and post-Cold War variables fail to achieve statistical

significance in either model.

In the inflated models, the polity measure has a negative and statistically significant

coefficient, which suggests that rebel groups operating in countries with higher levels of

democracy are much less likely to be in the certain-zero category. This is in line with our

observation that regime type predicts excessive zeros, with the majority of non-terrorist rebels

operating under authoritarian regimes. While the positive coefficient of the rebel strength

variable is consistent with the literature’s prediction that exceptionally powerful rebels are more

likely to be in the certain-zero group, the variable does not achieve statistical significance. These

two findings confirm our intuition that, while several factors influence the amount of terrorist

activity carried out by rebel groups, rebels’ strategic decision to employ or eschew the tactic

altogether is largely determined by structural factors such as regime type.

Our findings about the relationship between rebel market share and terrorism are robust

Page 26: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

26

to a variety of alternative measurement and model specifications, including more restrictive

measures of terror attacks and fatalities and estimations with fixed-effects negative binomial

regression. Due to space constraints, the results are presented in the Appendix.

5.1 Effects of Market Share in Multi-Party Conflicts

Roughly two-thirds of the observations in our data set represent monopolized conflict

markets, where the rebel group is fighting the government in the absence of any competition.

While we expect monopolistic groups to produce high amounts of terrorism in civil wars, our

theory offers similar predictions about stronger rebel groups operating in multi-actor conflicts,

who possess the capabilities for proliferate terrorist activity while also having incentives to use

terrorism as a means of safeguarding their market-dominant position vis-à-vis competitors. In

order to confirm that our findings also hold for conflicts that resemble competitive markets, we

examine the variation in terrorist activity across rebel groups involved in multi-party civil wars.

Table 3 presents models estimated using a subsample from our dataset that only includes

observations where a rebel group has at least one competitor in the given year. Model 3 looks at

the effect of rebel market share on terror attacks; Model 4 re-estimates the same model using

terror fatalities as the dependent variable. Similar to our earlier estimations with the full sample,

the troop share variable has a statistically significant and positive effect on the number of

terrorist attacks by rebel group. The incidence rate of attacks for market-dominant rebels with

80% troop share is 1.86 times that of groups commanding 20% market share; the ratio from the

same comparison is 2.03 when we look at the incidence rates of terror fatalities. These findings

confirm our expectation: in competitive conflict markets, the rebel organizations that are stronger

than their potential rivals tend to produce larger amounts of terrorism.

Page 27: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

27

Table 3. ZINB Models of Rebel Terrorist Activity in Competitive Conflict Markets

Our closer examination of multi-party civil wars provides the additional insight that,

when studying terrorist activity at the group level, we need to consider how competitive

dynamics may play out differently for rebel organizations depending on their power position

relative to potential competitors. As such, explanatory variables that only focus on the number of

Page 28: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

28

rebel groups in the conflict, commonly employed in empirical analyses of terrorist outbidding,

fail to capture the difference between the strong and weak actors in terms of their military

capabilities and perception of threats posed by rival groups, thereby making only loose proxies

for inter-group competition.

6 Conclusion

In this article, we advance a market theory of civil wartime terrorism that highlights the

relationship between rebels’ military strength relative to other armed groups and their use of

terrorist tactics. Our main proposition is that rebel groups who control a larger portion of

resources in the conflict landscape will employ terrorist attacks more extensively, with increased

market share driving terrorism through a mix of capabilities and strategic incentives. As rebel

groups secure a larger market share in the conflict, they become more capable of carrying out

widespread terrorist attacks and more acceptant of risks associated with the tactic, such as

legitimacy costs. In addition, groups who obtain dominant market shares become incentivized to

use terrorism to maintain their position vis-à-vis potential competitors. We test our argument

using a new measure of rebel troop share on all rebel groups in civil wars between 1970 and

2011. Our hypothesis finds strong support—groups possessing larger shares of military

capabilities perpetrate significantly more terrorism, both in terms of attacks and fatalities.

Our findings diverge from much of the current literature on the determinants of civil war

terrorism. First, we add nuance to the widespread belief that multi-party conflicts are inherently

more likely to produce terrorism. What matters is the distribution of power within the rebel

conflict system, not the number of actors per se. A multi-party conflict where one rebel group

controls a dominant share of the fighting capabilities is more likely to experience terrorism than

a conflict with the same number of groups but a more equal distribution of power.

Page 29: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

29

This insight has important implications for the literature that models conflict as an

economic process. Most of this scholarship assumes that conflict inverts traditional economic

logic, with competition, not monopoly, leading to worse outcomes for civilians (e.g., Metelits

2010; Wood and Kathman 2015). We find instead that conflict markets operate much like

economic ones insomuch as the presence of dominant actors drives inefficiencies (in our

interpretation, violence against civilians in the form of terrorism). This suggests the need for a

more nuanced understanding of the effects of competition in the context of civil war.

We also contribute to the literature by proposing a new measure of rebel market share,

which we operationalize by measuring each group’s troop number as a proportion of the total

rebel fighting force within the same conflict. The market share variable provides important

information about a rebel group’s military standing in a conflict, which cannot be captured in

standard comparisons of strength between rebel groups and the target government. Our approach,

somewhat paradoxically, also makes rebel-to-government measures more informative. For

example, imagine two rebel groups, A and B, with similar troop numbers in two separate

conflicts, each coded as “weaker” than the government. Group A is the smallest of several rebel

actors involved in the conflict and, according to our market theoretical logic, will commit

comparatively little terrorism. Group B, on the other hand, is the sole non-state actor and thus

commands all of the available resources in the conflict market. We demonstrate empirically that

Group B poses a greater threat for extensive and lethal use of terrorism, controlling for other

factors. This insight also has implications for counter-terrorism policies. In particular, it suggests

that relative strength between groups is a better indicator for assessing the threat of terrorism

than relative strength vis-à-vis the government.

Page 30: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

30

Finally, future research should focus on developing even more accurate measures of rebel

group competition, taking into account not only military capabilities, but also other aspects of

intergroup interaction such as geographical overlap and clashes over territory. An important area

for inquiry is how shifts in capabilities drive changes over time. Evidence from the civilian

targeting literature suggests that declines in relative capabilities, induced by new market entrants,

may lead to increased brutality (Wood and Kathman 2015); whether this logic holds in the

context of terrorism is an important empirical question to investigate.

Page 31: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

31

References

Abrahms, Max. 2012. “The Political Effectiveness of Terrorism Revisited.” Comparative Political Studies 45(3): 366–93.

Asal, Victor, and R. Karl Rethemeyer. 2008a. “The Nature of the Beast: Organizational Structures and the Lethality of Terrorist Attacks.” The Journal of Politics 70(2): 437–49.

------. 2008b. “Dilettantes, Ideologues, and the Weak: Terrorists Who Don’t Kill.” Conflict Management and Peace Science 25(3): 244–63.

Blattman, Christopher, and Edward Miguel. 2010. "Civil War." Journal of Economic Literature, 48(1): 3-57.

Bloom, Mia M. 2005. Dying to Kill: The Allure of Suicide Terror. New York: Columbia University Press.

Boyns, David, and James David Ballard. 2004. “Developing a Sociological Theory for the Empirical Understanding of Terrorism.” The American Sociologist 35(2): 5–25.

Buhaug, Halvard. 2006. “Relative Capability and Rebel Objective in Civil War.” Journal of Peace Research 43(6): 691–708.

Central Intelligence Agency. Directorate of Intelligence. 1986. Sri Lanka: The Growing Insurgency. NESA 86-10036. Accessible online at: https://www.cia.gov/library/readingroom/docs/CIA-RDP88T00096R000300340001-2.pdf.

------. 1987. Insurgency and Counterinsurgency in Peru, Columbia, and Ecuador. NI IIM 87-

10005. Accessible online at: https://www.cia.gov/library/readingroom/document/cia-rdp91t00498r000200160001-3

Clauset, Aaron, and Kristian Skrede Gleditsch. 2012. “The Developmental Dynamics of

Terrorist Organizations.” PLoS One 7(11): 1–11.

Crenshaw, Martha. 1981. “The Causes of Terrorism.” Comparative Politics 13(4): 379–99.

Crozier, Brian. 1960. The Rebels: A Study of Post-War Insurrections. Boston: Beacon Press.

Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. 2009. “It Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Journal of Conflict Resolution 53(4): 570–97.

------. 2013. “Non-State Actors in Civil Wars: A New Dataset.” Accessible online at: http://privatewww.essex.ac.uk/~ksg/eacd.html.

Page 32: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

32

Downes, Alexander B. 2008. Targeting Civilians in War. Ithaca, NY: Cornell University Press.

Drakos, Konstantinos and Andreas Gofas. 2006a. “In Search of the Average Transnational Terrorist Attack Venue.” Defence and Peace Economics 17(2): 73-93.

------. 2006b. “The Devil You Know but Are Afraid to Face: Underreporting Bias and its Distorting Effects on the Study of Terrorism.” Journal of Conflict Resolution 50(5): 714-735.

Eck, Kristine, and Lisa Hultman. 2007. “One-Sided Violence against Civilians in War.” Journal of Peace Research 44(2): 233–46.

Enders, Walter, and Todd Sandler. 1999. “Transnational Terrorism in the Post-Cold War Era.” International Studies Quarterly 43(1): 145–67.

Eyerman, Joe. 1998. “Terrorism and Democratic States: Soft Targets or Accessible Systems.” International Interactions 24(2): 151-70.

Fazal, Tanisha M. 2017. “Rebellion, War Aims & the Laws of War.” Daedalus 146(1): 71–82.

Fearon, James D. and David D. Laitin. 2003. “Ethnicity, Insurgency and Civil War.” American Political Science Review 97(1): 75-90.

Feenstra, Robert C., Robert Inklaar, and Marcel P. Timmer. 2015. “The Next Generation of the Penn World Table.” American Economic Review 105(10): 3150–3182.

Findley, Michael G., and Joseph K. Young. 2012a. “Terrorism and Civil War: A Spatial and Temporal Approach to a Conceptual Problem.” Perspectives on Politics 10(2): 285–305.

------. 2012b. “More Combatant Groups, More Terror?: Empirical Tests of an Outbidding Logic.” Terrorism and Political Violence 24(5): 706–21.

Fortna, Virginia Page. 2015. “Do Terrorists Win? Rebels’ Use of Terrorism and Civil War Outcomes.” International Organization 69(3): 519–56.

Gleditsch, Kristian S. 2002. “Expanded Trade and GDP Data.” Journal of Conflict Resolution 46: 712–724.

Greene, William H. 1994. “Accounting for Excess Zeros and Sample Selection in Poisson and Negative Binomial Regression Models.” Working Paper No.EC-94-10, Department of Economics, New York University. Accessible online at: https://ssrn.com/abstract=1293115.

Hirshleifer, Jack. 2001. The Dark Side of the Force: Economic Foundations of Conflict Theory. New York: Cambridge University Press.

Page 33: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

33

Hovil, Lucy and Eric Werker. 2005. “Portrait of a Failed Rebellion: An Account of Rational, Sub-Optimal Violence in Western Uganda.” Rationality and Society 17(1): 5-34.

Hultman, Lisa. 2012. “Attacks on Civilians in Civil War.” International Interactions 38(2): 164–81.

------. 2007. “Battle Losses and Rebel Violence: Raising the Costs for Fighting.” Terrorism and Political Violence 19(2): 205–22.

Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. New York: Cambridge University Press.

Kalyvas, Stathis N., and Laia Balcells. 2010. “International System and Technologies of Rebellion: How the End of the Cold War Shaped Internal Conflict.” American Political Science Review 104(3): 415–29.

Kydd, Andrew H., and Barbara F Walter. 2006. “The Strategies of Terrorism.” International Security 31(1): 49–80.

Lafree, Gary, and Laura Dugan. 2007. “Introducing the Global Terrorism Database.” Terrorism and Political Violence 19(2): 181–204.

Li, Quan. 2005. “Does Democracy Promote or Reduce Transnational Terrorist Incidents?” Journal of Conflict Resolution 49(2): 278–97.

Mampilly, Zachariah Cherian. 2011. Rebel Rulers: Insurgent Governance and Civilian Life During War. Ithaca, NY: Cornell University Press.

Marshall, Monty G., Keith Jaggers, and Ted Robert Gurr. 2013. “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2011.” http://www.systemicpeace.org/polityproject.html.

McCormick, Gordon. 1990. The Shining Path and the Future of Peru. Santa Monica, CA: RAND Corporation.

Metelits, Claire. 2010. Inside Insurgency: Violence, Civilians, and Revolutionary Group Behavior. New York: New York University Press.

Nemeth, Stephen. 2014. “The Effect of Competition on Terrorist Group Operations.” Journal of Conflict Resolution 58(2): 336–62.

Oots, Kent Layne. 1989. “Organizational Perspectives on the Formation and Disintegration of Terrorist Groups.” Terrorism 12(3): 139–52.

Piazza, James A. 2011. “Poverty, minority economic discrimination, and domestic terrorism.” Journal of Peace Research 48(3): 339–53.

Page 34: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

34

Polo, Sara MT, and Kristian Skrede Gleditsch. 2016. “Twisting Arms and Sending Messages.” Journal of Peace Research 53(6): 815–29.

Rochlin, James F. 2003. Vanguard Revolutionaries in Latin America: Peru, Colombia, Mexico. Boulder, CO: Lynne Rienner Publishers.

Ross, Jeffrey I. 1993. “Structural Causes of Oppositional Political Terrorism: Towards a Causal Model.” Journal of Peace Research 30(3): 317-329.

Santifort-Jordan Charlinda and Todd Sandler. 2014. “An Empirical Study of Suicide Terrorism: A Global Analysis.” Southern Economic Journal 80(4): 981–1001.

Schmid, Alex P. 1992. “Terrorism and Democracy.” Terrorism and Political Violence 4(4): 14-

25.

Singer, David J., Stuart Bremer, and John Stuckey. 1972. “Capability Distribution, Uncertainty, and Major Power War, 1820-1965.” In Peace, War and Numbers, edited by Bruce Russett, 19-48. Beverly Hills, CA: Sage.

Stanton, Jessica A. 2013. “Terrorism in the Context of Civil War.” Journal of Politics 75(4): 1009–22.

Thomas, Jakana. 2014. “Rewarding Bad Behavior: How Governments Respond to Terrorism in Civil War.” American Journal of Political Science 58(4): 804–18.

Thornton, Thomas Perry. 1964. “Terror as a Weapon of Political Agitation.” In Internal War: Problems and Approaches, ed. Harry Eckstein. New York: Free Press of Glencoe.

United States. Congress. House. Committee on Foreign Affairs. Subcommittee on Western Hemisphere Affairs. The Threat of the Shining Path to Democracy in Peru: Hearings before the Subcommittee on Western Hemisphere Affairs of the Committee on Foreign Affairs. Washington, DC: GPO, 1992.

Vickers, John. 1996. “Market Power and Inefficiency: A Contracts Perspective.” Oxford Review of Economic Policy 12(4): 11–26.

Vuong, Quang H. 1989. “Likelihood Ratio Tests for Model Selection and Non-nested Hypotheses.” Econometrica 57(2): 307-333.

Weinstein, Jeremy M. 2005. “Resources and the Information Problem in Rebel Recruitment.” Journal of Conflict Resolution 49(4): 598-624.

Wintrobe, Ronald. 2006. “Extremism, Suicide Terror, and Authoritarianism.” Public Choice 128(1/2): 169–95.

Page 35: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

35

Wood, Reed M. 2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace Research 47(5): 601–14.

------. 2014. “Opportunities to Kill or Incentives for Restraint? Rebel Capabilities, the Origins of Support, and Civilian Victimization in Civil War.” Conflict Management and Peace Science 31(5): 461–80.

Wood, Reed M., and Jacob D. Kathman. 2015. “Competing for the Crown: Inter-Rebel Competition and Civilian Targeting in Civil War.” Political Research Quarterly 68(1): 167–79.

Young, Joseph K. 2016. “Measuring Terrorism.” Terrorism and Political Violence. Online first at: https://doi.org/10.1080/09546553.2016.1228630

Page 36: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

36

Appendix for “Weapon of the Market-Dominant: A Market Theory of

Terrorism in Civil War”

I Constructing the Dataset For this project, we constructed a dataset that matches rebel groups in the Non-State

Actor (NSA) dataset (Cunningham, Gleditsch, and Salehyan 2009; 2013) with terrorist

organizations identified in the Global Terrorism Database (GTD) (LaFree and Dugan 2007).

Using the NSA dataset, which expands on the Uppsala-PRIO Armed Conflict Dataset (ACD)

(Melander, Pettersson, and Themner, 2016), allowed us to gather more information on rebel

group characteristics such as military capabilities and organizational structure.20

Matching NSA to GTD

To build our dataset, we expanded NSA’s state-rebel group dyad spells into panel data

rather than using the ACD state-rebel group dyad years. We did so to avoid losing data on rebel

group behavior for years during which a group was active but missing from the dataset due to its

battle death numbers falling under the civil war threshold. The ACD excludes any year in which

the level of conflict falls below 25 battle deaths for each state-rebel group dyad, regardless of

whether the rebel organization involved in that conflict remains operational in the given year,

whereas the NSA dyad spells end only “if there is a period exceeding two calendar years in

which the level of conflict falls below 25 battle deaths in a year” (Cunningham, Gleditsch and

Saleyhan 2013, 519).

The ACD’s restrictive inclusion criterion may introduce selection bias when modeling

the relationship between rebel capabilities and use of terrorism, particularly if intermittent drops

in level of violence are correlated with both variables. For ongoing armed conflicts, excluding

years with low-level or no violence from analyses could mean losing information on how rebel

tactics respond to exogenous shifts in conflict intensity (e.g., the government announces a

ceasefire) or to sudden changes in the distribution of military capabilities among conflict parties.

20 Our approach is similar to Polo and Gleditsch’s (2016) ACD2GTD dataset, which matches GTD perpetrators and incidents with rebel groups in the ACD.

Page 37: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

37

We attempted to overcome these problems by matching the GTD data with NSA dyad spells,

which cover intermittent periods of low-level violence as well as high-intensity conflict years.

Matching Group Names

Within the GTD dataset, many attacks are either unattributed to a specific group—i.e.,

the group name variable (gname) is “unknown” —or attributed to a group identified only by a

vague descriptor (e.g., “Kachin insurgents”). In a limited number of situations, where a vague

group name clearly matched only one group in the conflict, we added those attacks to the group’s

total. In other situations, the GTD group name was not a perfect match, but clearly referred to a

specific group. These situations include cases where one group has two different naming

conventions in the literature or cases where one data source used the group’s full name while the

other used the acronym. The following chart shows which vague identifiers or slightly

mismatched names in GTD (gname) were matched with groups in NSA (side b).

Country GTD gname NSA side b

Myanmar Kachin Insurgents Kachin Independence Army

KIO

Myanmar Shan State Army-North SSPP

Nicaragua Contras FDN

FDN/Contras

Niger Air Azawak Liberation Front FLAA

Rhodesia ZANU ZAPU Zimbabwe Patriotic Front

ZANU-PF (only matched for the period after 1976)

Russia Chechen Rebels Republic of Chechnya

Rwanda Hutus Opposition alliance

Somalia Islamic Courts Union (ICU) ARS/UIC

Soviet Union Armenian Extremists Armenian Guerrillas Armenian Nationalist Armenian Militants

Government of Armenia and ANM

Page 38: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

38

South Sudan South Sudan Liberation Army (SSLA)

SSLM/A

Sudan Sudan People’s Liberation Army (SPLA)

SPLM

Thailand Muslim Separatists Islamic Extremists

Patani Insurgents

Uganda NRM UPM/NRA

Yugoslavia Kosovo Liberation Army (KLA)

UCK

A number of extremely vague descriptors were unable to be matched back to specific

groups in NSA. We delete the following gnames from both the restricted and expanded versions:

● Other

● Unaffiliated Individual(s)

● Miscreants

● Narco-Terrorists

● Youths

● Villagers

● Unemployed Persons

● Terrorists

● Taxi Drivers

● Students

● Strikers

● Strike Enforcers

● Squatters

● Rioters

● Protesters

● Political Group

● Political Activists

● Opposition Group

Page 39: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

39

● Mob

● Individual

● Gunmen

● Armed People

● Anti-Government Group

● Anti-Government Demonstrators

GTD and the Missing 1993 Data

As reported in Lafree and Dugan (2007, 186), GTD is missing data from the year 1993.

During the process of physically transferring PGIS—the data source off which GTD was built—

to the University of Maryland, the authors discovered that the data for 1993 had been lost in an

earlier move by the PGIS team. GTD2 attempts to re-collect this missing data, but the collection

methods vary from the rest of the original data set. Because there is no way to rectify this (and

because we can reasonably treat the data as missing at random), we follow the standard

procedure in the terrorism literature of simply using the GTD data as-is.

Handling the Weakest Groups

Some rebel groups in states experiencing a civil war may be too weak to appear in our

dataset, given the NSA battle death inclusion threshold. In this case, the group would have been

eliminated from our dataset, even if they were recorded in GTD as having committed terrorist

attacks which met our inclusion criteria. This exclusion of the smallest and weakest groups is

standard in the literature on terrorism within the context of civil war (Fortna 2015, 531). We do

not believe that excluding these groups would significantly change our results.

II Data and Coding Procedures Independent Variable: Rebel Troop Share

Data for our rebel troop share variable come from the NSA data, which provides

estimates of rebel organizations’ armed forces. For each state-rebel group dyad year, we first

identify all other rebel groups that were party to the same conflict in a given year, using the ACD

conflict identifier. We then use the NSA estimates of rebel forces to calculate the proportion of

troops that each rebel group controls relative to all the rebel troops active in the conflict year.

Page 40: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

40

Specifically, we use the NSA rebestimate variable, which provides “the best estimate of the size

of rebel armed forces” (Cunningham, Gleditsch, and Salehyan 2012, 4). Because we expand the

NSA data from state-rebel dyad spells into panel data, we impute the troop information, which is

provided at the spell level, to the corresponding panel years.

Dependent Variable: Terror Attacks and Fatalities

The data for our dependent variables—Terror Attacks and Terror Fatalities—are from

the GTD, which contains global information on domestic and transnational terrorist events from

1970 to present. In order for the GTD to include an incident, it must be: (1) intentional, (2)

violent, or have the threat of violence, and (3) perpetrated by a non-state actor. Additionally, an

incident must satisfy two of three criteria (LaFree and Dugan 2007, 188):

(1) The act must be aimed at attaining a political, economic, religious, or social goal. In terms of economic goals, the exclusive pursuit of profit does not satisfy this criterion. (2) There must be evidence of an intention to coerce, intimidate, or convey some other message to a larger audience (or audiences) than the immediate victims. And (3) the action must be outside the context of legitimate warfare activities; that is, the act must be outside the parameters permitted by international humanitarian law (particularly the admonition against deliberately targeting civilians or noncombatants.

Our sample only includes incidents that meet all three criteria in an effort to exclude as many

other forms of violence as possible.

The GTD allows researchers to apply additional filters, including attack and target. For

our primary specifications of the dependent variable, we filter the dataset to include only the

following attack types: hijacking, hostage taking (kidnapping and barricade incidents), armed

assault, bombing/explosion, and facility/infrastructure attacks. This excludes three attack types:

assassination, unarmed assault, and unknown, following Fortna (2017, 12). In particular,

assassinations do not meet our criteria for deliberately indiscriminate attacks, given that they are

the targeted killing of a single politically-relevant individual.

We likewise filter targets to include only the following target types: airports and aircraft,

business, educational institution, food or water supply, private citizens and property, religious

figures/institutions, telecommunication, tourists, transportation, utilities, and unknown. We

exclude target types that are unlikely to be truly indiscriminate, such as journalists and NGOs.

Although these groups are often reprehensibly targeted for violence in conflict zones, we argue

Page 41: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

41

that these attacks tend to be selective and thus not terrorism as defined here. We also exclude

governmental and military targets. Rebel groups almost always are weaker than the government,

which means that by default they use asymmetric means. To mitigate this potential bias, which

would over-represent the use of terrorism, we only count attacks if they are directed against

civilian targets.

III Descriptive Statistics Table 4. Summary Statistics

Page 42: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

42

Figure 3. Distribution of Single and Multi-Party Conflict Years

Figure 4. Distribution of Rebel Group Troop Share

Page 43: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

43

Figure 5. Distribution of Rebel Group Market Share

Table 5. Control Variables

Variable Operationalization Sources Rebel-to-government Troop Ratio

Ratio of rebel to government troops

NSA dataset and the Correlates of War National Material Capabilities (NMC) dataset (Singer et al. 1972)

Democracy

Polity2 variable, which ranges from -10 to 10

Polity IV dataset (Marshall, Jaggers, and Gurr 2013)

Logged GDP per capita

Logarithmized Gross Domestic Product per capita in constant 2005 US$

Gleditsch (2002) and the Penn World Tables (Feenstra, Inklaar, and Timmer, 2015)

Conflict Intensity

Binary indicator that takes a value of 2 in major conflict years (> battle deaths) and 1 in minor conflict years (25-99 battle deaths)

Uppsala-PRIO Armed Conflict Dataset (ACD) (Melander, Pettersson, and Themner, 2016)

Foreign Support

Dummy variable that takes a value of 1 if a group receives NSA dataset

Page 44: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

44

financial or military assistance from foreign states

Nationalist-Separatist

Dummy variable that takes a value of 1 if a rebel group has a nationalist-separatist agenda

Polo and Gleditsch (2016)

Post Cold War

Dummy variable that takes a value of 1 for years > 1989

Number of Competitors

Count of the number of other rebel groups active within the same conflict-year (used in multi-party models)

IV Robustness Checks

We run additional statistical analyses to demonstrate that our findings are robust to

different operationalizations of the dependent variable and alternative model specifications.

Models 5 through 15 present the results from the robustness checks. First, to ensure that our

results do not rely on the particular coding of terrorism, we re-estimate the models employing

different operationalizations of terrorist activity. The first one is an alternative count variable for

terror attacks, which is coded using a more restrictive definition of terrorism. Our restricted

count limits attack types to the two most frequent modes of attack: armed assault and

bombing/explosion. We keep the target types, but exclude any unknown targets from this coding.

The other variable, employed in Model 6, is the restrictive attack count from Polo and Gleditsch

(2016).21 The results, presented in Table 6 and displayed in Figure 6, are largely consistent with

21 Polo and Gleditsch (2016, 821): “Since definitions of terrorism are disputed, we use two operationalizations with different event inclusion criteria. The most inclusive is the GTD definition covering all events that satisfy at least two of the criteria in the GTD codebook…. A more restrictive version requires that events satisfy all three criteria, and excludes all attacks against military targets, even if classified by the GTD as falling outside guerrilla warfare.”

Page 45: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

45

the original findings and support our rebel market share hypothesis.

Next, we examine whether our findings hold with alternative model specifications. First,

we estimate negative binomial regression models, using both the extensive and restrictive

measures of terror attacks and fatalities. Next, we specify fixed-effect negative binomial (FENB)

models, which account for the time-invariant, unobserved group-specific effects. We estimate

these using rebel group dummies (i.e., unconditional FENB) instead of the conditional FENB

method proposed by Hausman et al. (1984), which has been shown to fall short of removing

individual fixed effects in count panel data (see Allison and Waterman 2002, Greene 2005;

Guimarães 2008). The results from negative binomial (Models 7-10) and unconditional FENB

models (Models 11-14) mirror our original findings; in all models, the coefficients for rebel

troop share are statistically significant. One notable difference from original findings is that, in

some of the models where the DV is terror attacks, the rebel-to-government troop ratio variable

has a negative and statistically significant effect on rebel group terrorism. This suggests some

evidence in favor of the traditional weapon of the weak hypothesis, although the finding is not

too robust and the market share variable still appears as the stronger predictor of terrorist attacks.

Finally, in Model 15, we include a quadratic term in the ZINB model of terror attacks to

check whether is a non-linear, U-shaped relationship between rebel market share and terrorism.

The results, displayed in Figure 7, do not indicate the presence of a curvilinear effect.

Page 46: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

46

Table 6. Re-estimating ZINB Models with Alternative Measures of Rebel Group Terrorism

Page 47: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

47

Figure 6: Predicted Number of Attacks and Fatalities by Rebel Market Share (Using Alternative Measures of Terrorist Activity by Rebel Group)

Page 48: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

48

Table 7. Negative Binomial Regressions of the Effect of Rebel Market Share on Terrorist Activity

Page 49: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

49

Table 8. Unconditional Fixed-Effects Negative Binomial Regression of the Effect of Rebel Market Share on Terrorist Activity

Page 50: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

50

Table 9. ZINB Model Examining the Presence of a Non-monotonic Relationship Between Rebel Market Share and Terrorism

Page 51: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

51

Figure 7: Predicted Number of Attacks by Rebel Market Share (ZINB Model with Quadratic Term)

Appendix References Allison, Paul D and Richard P. Waterman. 2002. “Fixed-Effects Negative Binomial Regression Models.”

Sociological Methodology 32: 247-265. Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. 2009. “It Takes Two: A Dyadic

Analysis of Civil War Duration and Outcome.” Journal of Conflict Resolution 53(4): 570–97. ———. 2012. “Codebook for the Non-State Actor Data.” Accessible online at:

http://privatewww.essex.ac.uk/~ksg/eacd.html. ———. 2013. “Non-State Actors in Civil Wars: A New Dataset.” Accessible online at:

http://privatewww.essex.ac.uk/~ksg/eacd.html. Feenstra, Robert C., Robert Inklaar, and Marcel P. Timmer. 2015. “The Next Generation of the Penn

World Table.” American Economic Review 105 (10): 3150–3182. Fortna, Page. 2017. “Is Terrorism Really a Weapon of the Weak? Testing the Conventional Wisdom”

Columbia University. Paper Presented at ISA 2015. Accessible online at:https://www.princeton.edu/politics/about/file-repository/public/Is-T-a-Weapon-of-the-Weak-2016-4.pdf

Page 52: Weapon of the Market-Dominant: A Market Theory of ...aslihansaygili.com/files/Weapon-of-Market-Dominant_Sept2018.pdfNon-State Actor (NSA) data (Cunningham, Gleditsch, and Salehyan

52

Gleditsch, Kristian S. 2002. “Expanded Trade and GDP Data.” Journal of Conflict Resolution 46: 712–

724. Greene, William. 2005. “Functional form and heterogeneity in models for count data.” Foundations and

Trends in Econometrics 1(2): 113–218. Guimarães, Paulo. 2008. “The fixed effects negative binomial model revisited.” Economics Letters, 99:

63-66. Hausman, Jerry, Bronwyn H. Hall, and Zvi Griliches. 1984. "Econometric Models for Count Data with an

Application to the Patents-R&D Relationship." Econometrica 52:909-38. Lafree, Gary, and Laura Dugan. 2007. “Introducing the Global Terrorism Database.” Terrorism and

Political Violence 19(2): 181–204. Marshall, Monty G., Keith Jaggers, and Ted Robert Gurr. 2013. “Polity IV Project: Political Regime

Characteristics and Transitions, 1800-2011.” Accessible online at: http://www.systemicpeace.org/polityproject.html.

Melander, Erik, Therese Pettersson, and Lotta Themner. 2016. “Organized Violence, 1989 to 2015.”

Journal of Peace Research 53(5): 727–742. Polo, Sara MT, and Kristian Skrede Gleditsch. 2016. “Twisting Arms and Sending Messages.” Journal of

Peace Research 53(6): 815–29. Singer, David J., Stuart Bremer, and John Stuckey. 1972. “Capability Distribution, Uncertainty, and

Major Power War, 1820-1965.” In Peace, War and Numbers, edited by Bruce Russett, 19-48. Beverly Hills, CA: Sage.