Allying to Kill

Examples of Terrorist Alliances

We begin our analysis of how alliances influence lethality by presenting four illustrative examples of terrorist group cooperation and how the right alliances increased group lethality. To ensure that groups from one particular region or using one particular tactic are not overly biasing our illustrations, we draw these examples from several different regions and time periods. Each indicates how alliances can enhance group capacity and lead to more dangerous behavior. These short cases also show, anecdotally, some of these processes by which alliances between groups actually form and influence overall group capabilities. The statistical analysis that follows will identify whether what we find in the case studies is merely an aberration or generalizable to a larger phenomenon.

Hezbollah and Hamas/Palestinian Islamic Jihad

Among the best documented alliances between terrorist organizations is the relationship between Hezbollah and Hamas. In 1992, after a period of turmoil, Israel expelled 415 members of Hamas and Palestinian Islam Jihad (PIJ) to Lebanon, where they came into contact with leaders of Hezbollah, specifically Imad Mughniyeh (Ricolfi 2005, 91–92; Westcott 2008). Despite their theological differences, Hamas and PIJ forged cooperative ties with Hezbollah because they saw the opportunity to learn from one of the most successful (in terms of operational effectiveness) terrorist groups in the region. In our view, this is a primary example of the preferential attachment in alliance formation that we described in the theory section of this article—this relationship was formed because of existing capability rather that convenience or ideology.

When most of the expelled members of Hamas and PIJ returned to Israel the next year, the casualties per attack for these groups increased from about two per attack to around ten. This rate was approximately the same as Hezbollah’s, demonstrating the convergence in capability resulting directly from the alliance that our theory predicts. This increased lethality stemmed largely from the increase in the number of suicide missions. Scholars have documented the extent to which Hezbollah taught Hamas and PIJ about the operational nuances of suicide terrorism during this period (Pedahzur 2005; Ricolfi 2005, 91–92). Additionally, Hezbollah has used these ties to spread deadly weapons in the Gaza Strip, using men such as Adnan al-Ghoul as hubs for the distribution of weapons and knowledge (Rees 2001). For example, Hamas apparently used connections and funding from Hezbollah to buy precursor technologies for crude mortars and Qassam rockets (Middle East Intelligence Bulletin 2004, 8–9). ¹⁰

Figure 1 below suggests the impact of collaboration on Hamas’ effectiveness. Note the striking increase in fatalities per attack that occurred after members of Hamas return from their initial collaboration with members of Hezbollah in 1993. After that point, fatalities wax and wane with the severity of Hamas’ overall operations against Israel, but the spikes during periods of higher activity demonstrate what Hamas gained from the collaboration.

OPEN IN VIEWER

Figure 1.

Hamas/Hezbollah collaboration and Hamas lethality

Palestinian and Leftist Organizations

In most cases, the known alliances between terrorist groups have paid tangible dividends to both sides. To take one example, the Japanese Red Army (JRA), a relatively small Maoist organization (perhaps forty-five strong at its apex) with stated goals of overthrowing the Japanese Government and monarchy and fomenting world revolution, trained in communist-friendly countries in the Middle East to develop the skills required to launch attacks within Japan (Kotani 2003).

According to a 1974 Central Intelligence Agency (CIA) report, JRA’s leader, Shigenobu Fusaki, made contact with the Popular Front for the Liberation of Palestine (PFLP) in 1971. Subsequently, he helped produce a film called “The PFLP and the Red Army Declare World War” and published a book entitled The Arab Guerillas and the World Red Army (Central Intelligence Agency 1974). Soon after, the JRA sent members to Lebanon, where they received assistance, shelter, and training from the PFLP (Kotani 2003). After a year of training and cooperation, on May 30, 1972, three members of the JRA launched an attack on the baggage reclaim hall at Lod Airport in Israel. Using guns and grenades in their checked baggage, the terrorists fired indiscriminately, killing twenty-six and wounding seventy-two. The PFLP contributed weapons, safe houses, and other aid to help the Japanese Red Army conduct the assault (Bar-Maoz 1991, 8–9; Clutterbuck 1990, 167–72).

At his trial in Israel in 1972, Kozo Okamoto, the only JRA attacker to survive the Lod airport attack, gave a revealing account of the international network involved in the operation. After some weeks of general training in Beirut, the three terrorists were sheltered, briefed on the specifics of the specific operation, and trained with the type of weapons they would use (Clutterbuck 1990, 167–72). Okamoto said that the JRA cell had allied with the PFLP because “the Arab world lacks spiritual fervor so we felt that through this attempt we could stir up the Arab world” (Time Magazine 1972).

The quid pro quo in the alliance between the two organizations is relatively clear—the PFLP received dedicated manpower while the JRA received safe haven, funding, training, and visibility that allowed them to better pursue their own cause. Moreover, cooperation increased the militancy of the rank and file in both organizations. This again underlines the extent to which these relationships can be highly instrumental in their design and pay tangible dividends.

The Provisional Irish Republican Army (PIRA) and the Revolutionary Armed Forces of Colombia

Another example of the deadly effect of terrorist alliances comes from links between the PIRA and the Revolutionary Armed Forces of Colombia (FARC). The relationship officially came to light when James Monaghan, Martin McCauley, and Niall Connolly, three experienced PIRA operatives, were captured in Colombia while training and working with the FARC. Subsequent congressional hearings in the United States revealed that over a dozen PIRA members had traveled back and forth to Colombia in the preceding years. Indeed, PIRA operatives had been living in Cuba, sheltered by the Castro regime, highlighting the breadth of the network (Majority Staff of the House International Relations Committee 2002).

As our theory predicts, the PIRA–FARC relationship translated into increased effectiveness for both organizations. During one raid, the Colombian army found FARC-operated mortars with similar design principles to those used by the PIRA in Northern Ireland (Murphy 2004, 81). The committee report from the U.S. House of Representatives noted in 2002 that, after interacting with the PIRA, the FARC began to employ new techniques in urban terrorism. Specifically, the FARC adopted the PIRA tactic of “car bombings which target police explosives teams and other first-responders whose job is to dismantle or neutralize these deadly devices” (Majority Staff of the House International Relations Committee 2002). The tactic was startlingly effective—the Columbian police lost more than 10 percent of their bomb technicians in the year after it was implemented (Majority Staff of the House International Relations Committee 2002).

Beyond the attacks on first responders, PIRA explosives training techniques resulted in more effective attacks against the Colombian urban infrastructure and expanded targets to include bridges, power lines, and reservoirs. Moreover, PIRA training contributed to mounting police and military casualties. As a result of their increased capability, the FARC was able to regain the initiative and put the Colombian government on the defensive.

For the PIRA, the relationship yielded primarily financial benefits. Coca money from the FARC helped fund PIRA weapons purchases and pay for personnel (Oppenheimer 2008). Among other things, payments from the FARC enabled the “shipment of 20 highly efficient Russian AN-94 assault rifles that Russian intelligence reported going to the IRA at about the same time that the three IRA experts were in Colombia” (Curtis and Karacan 2002).

This example nicely illustrates extreme geographical, cultural, and ideological divides that terrorist organizations are willing to bridge in their quest to increase their capabilities. It also underlines that the rewards that each side reaps may be very different, but may still, in the end, increase their capability to kill.

al-Qaeda and Jemaah Islamiyah

The origin of the relationship between JI and al-Qaeda goes back three decades to the Soviet war in Afghanistan. One of the key early JI leaders, the late Abdullah Sungkar, spent time in al-Qaeda training camps in Afghanistan during the 1980s. The Southeast Asian militant Islamic groups who trained and fought with al-Qaeda eventually formed a loose alliance including JI, the Islamic Liberation Front of Indonesia, and the Moro Islamic Liberation Front of the Philippines (Ministry of Home Affairs 2003). Al-Qaeda’s motivation for maintaining and expanding its contacts into Southeast Asia was, and still is, the globalization of its cause and the opportunity for training grounds outside the Middle East for Islamic fighters. The Southeast Asian groups sought funding and tactical knowledge.

Events over the last decade demonstrate what JI received from al-Qaeda. As described in the introduction to this article, Jemaah Islamiyah planned attacks on Singaporean targets including embassies, bases, businesses, and American naval personnel that would have been well beyond their capability and aspirations without al-Qaeda support. The attack plans highlight the degree to which the alliance between al-Qaeda and JI effectively joined al-Qaeda’s advanced direction and planning skills with JI’s manpower and militancy. This bolstered JI’s capacity to conduct deadly attacks. It is also worth noting that, with the alliance, JI increased its emphasis on promoting al-Qaeda’s broader jihad against the United States, which contrasts with JI’s traditionally regional goals.

One final but important aspect of the relationship is the extensive training that JI operatives have historically received in al-Qaeda camps. Eight of the JI members arrested in January 2002 had been to Afghanistan and received training on the usage of AK47’s, mortars, and more advanced military tactics including ambush/assassination, sniper missions, and bomb construction (Ministry of Home Affairs 2002, 2003, 8–9). This training and direction from al-Qaeda was designed to harden JI militants, particularly younger members without wartime experience from the Soviet occupation in Afghanistan. Similar to the relationship between Hamas and Hezbollah, the clear beginning of cooperation and sustained activity over time allow us to look at the impact of cooperation from a time-series perspective. Figure 2 demonstrates that the alliance with al-Qaeda in the form of training and operational collaboration was immediately followed by a high fatality rate, prior to its suppression by the government.

OPEN IN VIEWER

Figure 2.

JI/al-Qaeda collaboration and JI lethality

Research Design

The preceding illustrations of the deadly consequences of several alliances represent only a very small sample of the universe of all such relationships. Indeed, in the course of our research, we have identified over 300 such relationships. ¹¹ What is the systematic relationship between this surprisingly vast network of terrorist alliances and lethality? ¹² To answer this question, we turn to a statistical examination of the association between alliance and lethality.

We limit the statistical analysis to relationships among currently active organizations, which avoids conflating ongoing relationships with those that died out in the period of analysis. We note from the outset that the lack of time-series data on alliance formation and dissolution places limitations on the causal claims we can draw from the statistical analysis. However, when assessed in conjunction with the qualitative research and figures presented above, statistical analysis can help us determine whether the broad patterns of association we predict actually exist, and their potential importance.

Operationalizing Lethality

Our dependent variable in the analyses that follow is the number of deaths a terrorist group has generated from its attacks. While fatalities are of prime concern from a counterterrorism perspective, we argue that they are also indicative of underlying capacity. Terrorist groups generally seek to demonstrate their capabilities to the governments they oppose as well as the broader population and the ability to kill is the primary way to do so. In addition, terrorist groups exist as a result of their willingness to use violence to achieve their ends (Pape 2005; Pedahzur 2005; Atran 2004). Finally, lethality influences group status and prestige. While there are various ways that potential members, other terrorists, and governments might assess the “success” of a terrorist operation, the most reliable indicator is death and destruction. ¹³

We test our argument that alliances contribute to lethality using two different formulations of the fatalities measure. The first is simply a count of the number of fatalities an organization inflicts over its lifetime. Since the total number of casualties might be skewed by a single attack that inflicts an especially large number of casualties, we also test the relationship between alliances and the efficiency of attacks—that is, the number of fatalities inflicted per attack. This latter measure can be thought of as a measure of terrorist group “efficiency.”

Operationalizing the Number and Depth of Terrorist Alliances

We assembled data on all terrorist organizational alliances from the Terrorism Knowledge Base (TKB), a collaborative effort between the Memorial Institute for the Prevention of Terrorism (MIPT) and the RAND Corporation. Buried within the TKB is a description of the linkages among organizations that existed between 1968 and 2008, when MIPT ceased adding information. ¹⁴ We limit our analysis below to groups active between 1998 and 2008, when MIPT was gathering data on both international and domestic terrorism. The MIPT describes several types of relationships between terrorist organizations. Prior research has focused on these relationships as positive linkages (founding group, ally, armed wing, political wing, shared members, supported cause, and umbrella organization) and negative linkages (competing faction, enemy, rival, and splinter group). ¹⁵ We argue that only differentiating between positive and negative alliances potentially obscures the genuine interorganizational alliances that are the theoretical focus of this article.

This is because the formulation of positive and negative alliances conflates strategic alliances with ties between groups that stem from a common organizational origin such as the Irish Republican Army (IRA) and the Real Irish Republican Army (RIRA), which the TOPs data identify as a “Founding Group”/“Splinter Group” relationship. We want a more conservative measure in order to increase our confidence that the alliances captured in our data are examples of voluntary cooperation rather than legacy. More significantly, the data generating processes that lead to discretionary alliances between groups are very different from those that produce the ties between the political and military wing of a group or groups that share members. We therefore focus exclusively on the ties described as “alliances” by the MIPT project. ¹⁶

We operationalize the magnitude of each organization’s alliance portfolio in two ways. To start, we simply count the number of alliances each organization has with other organizations. ¹⁷ This is a simple measure of the ties each group has to others in the terrorist universe. However, our theoretical discussion suggests that it matters whether an ally is isolated or well connected. The implication of our argument is that a terrorist group with deep ties to the core of the network (those who maintain dense linkages themselves) is more likely to reap rewards in terms of knowledge and capability than is one that is not—even that group has a high number of first-degree contacts.

We capture the degree to which groups have such consequential alliances with a measure of eigenvector centrality. According to Borgatti (2005), in a diffusion process “an actor who is high on eigenvector centrality is connected to many actors who are themselves connected to many actors,” thus multiplying their exposure. Essentially, the impact of network linkages between terrorist groups may operate in a deeper and more important way than previous research has recognized. Mathematically, eigenvector centrality is proportional to the sum of the scores of all nodes.

x i = 1 λ ∑ j ∈ M ( i ) x j = 1 λ ∑ j = 1 N A i , j x j ,

Where:

x_i is the eigenvector score of the ith node,

This results in a system of equations from which we can derive a measure of eigenvector centrality for each terrorist organization. We anticipate that a terrorist group with alliances to core groups (those who maintain deep linkages themselves) is more likely to reap rewards in terms of capability than one that is not—even if it has some individual alliances. We argue that a primary manifestation of the increase in capability that accompanies increasing alliance depth, measured in terms of eigenvector centrality, is the ability to conduct attacks that kill more people and therefore garner more attention.

The resulting independent variable, Alliance Depth, assigns relative values to all groups in the system based on the principle that connections to well-connected organizations (i.e., those at the core of the alliance network) make a larger contribution to capability than do alliances with less-connected organizations (i.e., those at the periphery of the alliance network; see Borgatti 2005). This is because information, tactical innovations, and resources diffuse more readily in the connected core than the relatively disjointed periphery (and even less so among organizations with no alliances at all). A group with a smaller number of connections that are in turn closely linked to the core will have greater alliance depth than a group with more links that are isolated to the periphery. If our argument is correct, those groups with deeper alliances (i.e., those groups with more links within the core of the overall network, of terrorist relationships) should be able to more effectively kill civilians in their attacks. In contrast, we anticipate that a more modest effect for the simple count of alliances (Alliance Count) because it conflates more and less consequential relationships. In sum, we anticipate that a terrorist group with deep ties to the core of the network (those who maintain dense linkages themselves) is more likely to reap rewards in terms of knowledge and capability than is one that is not—even if the second group has a high number of first-degree contacts.

Other Independent Variables

Many factors may influence the lethality of terrorist groups, only one of which is its alliances with other groups. We want to account especially for variables that might influence both the propensity of groups to ally and their propensity to launch attacks with high numbers of fatalities due to the fact that they will introduce bias into our model specifications. Including these variables does not change our core results, but it does increase our confidence in the validity of our findings.

For example, larger groups might be more prominent, making them more likely to become alliance partners, but their size may also independently correlate with lethality. We therefore incorporate data from Asal and Rethemeyer on organizational size into our models with an organizational size (Size) variable that runs from 0 to 3. Groups with 100 members or fewer are a 0, those with membership between 100 and 1,000 are a 1, those with membership between 1,000 and 10,000 are a 2, and those with membership above 10,000 are coded a 3.

Younger terrorist groups, because they are less established in their organizational routines, might be both more likely to seek alliances partners and more receptive to new and deadly tactics such as suicide bombing. Horowitz (2010) suggests an inverse relationship exists between organizational age and innovative capacity for terrorist groups. In contrast, Miller (2008) argues that for regular operations, organizational age should be positively related to group effectiveness. We therefore control for the age of terrorist organizations (age) with a count variable that measures the number of years that have transpired since the founding of a group, according to MIPT. ¹⁸

Similarly, the ideological background of a terrorist group might influence both its potential alliance partners and desire to forge alliances in general, particularly if it has access to other organizations of similar ideology. At the same time, ideology could influence the lethality of a group by raising or lowering inhibitions against indiscriminate killing, introducing correlation with both the independent and the dependent variables in our models. For example, Juergensmeyer (2000) and Hoffman (1998), among others, argue that the universalist nature of religious motivations makes groups more likely to discount the negative consequences of mass casualties and instead focus on generating the biggest possible symbolic acts of violence. The “cosmic wars” often perceived by religiously motivated terrorist groups make them especially prone to deadly violence. ¹⁹ Even Pape (2005), who believes occupation, rather than religion, leads most immediately to suicide bombing (a particularly deadly form of terrorism) argues that religious differences between groups are an important predictor of violence. We therefore control for organizations centered on religiously based ideologies.

We also identify organizations with ethnonationalist motivations in response to the prominence of groups such as the PIRA in the universe of terrorist organizations. ²⁰ Ethnonationalist groups tend to congregate around territorial and secessionist struggles and therefore may be more likely to find themselves in close proximity to potential allies engaged in the same fight. At the same time, these types of conflict tend to engender strong commitments from members and may be associated with lethality through that mechanism. To operationalize these aspects of group ideology, we gathered data from MIPT on the ideology of terrorist organizations and include controls for religious (Religion) and ethnonationalist (Nationalist) organizations in our models.

Suicide bombing has been linked to a substantial upsurge in the number killed per attack. It may also have an independent relationship with alliance formation aside from its relationship with lethality because it is a visible and prestigious tactical innovation (Horowitz 2010). To ensure that our results are not biased, we control for groups that have adopted suicide bombing as tactic (Suicide). ²¹ Along similar lines, we identify organizations that control territory (Territory). Such organizations have resources at their disposal that may independently contribute to their capabilities and their attractiveness as alliance partners. ²²

In addition to the attributes of the organizations themselves, we must consider the environment in which terrorist groups operate. In order to adjust for changes in the political circumstances in which terrorist organizations must operate, we control for the regime type of the home country of the terrorist group using the POLITY2 data (Sarkees 2000). Our regime type variable ranges from −10 (autocracy) to 10 (democracy) (Polity). Krueger (2007) and Krueger and Maleckova (2003) have demonstrated that an important determinate of terrorist group behavior, including their overtures abroad, derive from the degree of political repressiveness in the county in which they originate. Repressiveness could also put a damper on alliance formation by forcing organizations further underground and putting a premium on the security that would accompany isolation. To account for this, we include two variables from Freedom House, Repressionand Press Freedom.We also include a measure of GDP per capita from the World Bank in order to address issues of state resources, capability, and poverty. Including these variables makes our work comparable with past research, especially Piazza (2008) and Jackson Wade and Reiter (2007). Finally, in response to the concern that terrorist groups ally and kill around particularly salient conflicts, we control for organizations oriented toward the conflicts in Lebanon, Palestine/Israel, and Iraq. ²³ We present descriptive statistics in the Online Appendix.

Findings

Table 1 presents a series of statistical models aimed at uncovering the relationship between terrorist group alliances and the lethality of attacks by these organizations. Because both dependent variables are count variables, we rely on negative binomial models. ²⁴

OPEN IN VIEWER

Table 1.

Alliances and Lethality (Active Groups Only)

		Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
		β/(SE)	β/(SE)	β/(SE)	β/(SE)	β/(SE)	β/(SE)
		Fatalities	Fatalities	Fatalities	Fat./Attack	Fat./Attack	Fat./Attack
Alliance	Alliance Depth	0.347***	0.044**	0.030*	0.084***	0.049***	0.044***
		(0.102)	(0.021)	(0.017)	(0.023)	(0.014)	(0.015)
	Alliance Count		–0.091	–0.015		–0.250***	–0.153*
			(0.091)	(0.085)		(0.075)	(0.081)
Organizational Covariates	Suicide		2.117***	1.257***		0.998***	0.893***
			(0.305)	(0.315)		(0.321)	(0.293)
	Org. Age		0.032***	0.009		–0.003	–0.023***
			(0.010)	(0.009)		(0.008)	(0.009)
	Size		1.411***	0.767***		0.798***	0.456***
			(0.232)	(0.198)		(0.183)	(0.166)
	Territory		1.001**	1.042**		0.118	–0.496*
			(0.419)	(0.409)		(0.314)	(0.281)
	Religious		1.389***	0.714**		2.007***	1.414***
			(0.311)	(0.345)		(0.301)	(0.281)
	Nationalist		0.597**	0.131		0.224	–0.060
			(0.276)	(0.340)		(0.268)	(0.258)
National Covariates	Polity			0.019			–0.037
				(0.045)			(0.030)
	GDP/Per Cap.			0.000			0.000
				(0.000)			(0.000)
	Repression			0.015			–0.107
				(0.269)			(0.200)
	Press Freedom			0.012			–0.002
				(0.017)			(0.012)
	Israel			0.058			–0.697
				(0.433)			(0.534)
	Iraq			0.530			–0.896***
				(0.440)			(0.337)
	Lebanon			–0.095			–0.178
				(0.582)			(0.610)
	N	431	274	122	431	274	122

Note. Models present maximum likelihood coefficients obtained from negative binomial models (one-tailed tests). The terrorist organization is the unit of analysis. Robust standard errors are in parentheses. Alpha values are positive and significant in all models.

*p < .10. **p < .05. ***p < .01.

Models 1 through 3 in Table 1 assess the relationship between terrorist group alliances and total fatalities. Models 4 through 6 focus on how terrorist group alliances influence the number of fatalities per attack. In line with Achen’s (2005) prescriptions, we begin with simple models that exclude the control variables (models 1 and 4). This allows us to assess all active organizations for which we have fatalities data (N = 431). ²⁵ In models 2 and 5, we add the control variables. In models 3 and 6, we limit the analysis to those groups that have inflicted at least one fatality. This accounts for the possibility that there are important unmodeled differences between groups that kill and those that do not. Limiting the analysis to active organizations partially addresses the concerns raised by the aforementioned absence of time-series data because it enables us to avoid conflating active groups with those that died out over the period of analysis. This, combined with the control for the age of each organization, ensures the closest thing to an “apples-to-apples” comparison that the data will allow. Moreover, if alliances do not appear linked to the fatalities a terrorist group can generate in this statistical arrangement, they are very unlikely to matter with more complete data. Thus, answering the question using our setup is a critical first step toward better understanding the role of terrorist group alliances. ²⁶

As we predict, there is a strong and positive relationship between our measure of alliance depth and fatalities. The measure of alliance depth is consistently positive and significant in our models without control variables, when we add in control variables, and when we restrict to only groups that have inflicted at least one casualty. In addition, alliance depth is significant when we switch from the dependent variable being total casualties inflicted to average fatalities per attack. Thus, while our alliance depth measure is certainly not the only factor that drives group lethality, the statistical evidence demonstrates that it plays a prominent role explaining much of terrorist group lethality.

In contrast, the simple count of alliances is insignificant in the fatalities models and negative and significant in the fatalities per attack models, suggesting that many inconsequential alliances can actually detract from the capability to kill. ²⁷ Notably, if we remove our alliance depth measure, the count of alliances becomes significant again in the overall fatalities models (1–3), consistent with Asal and Rethemeyer’s findings. These results demonstrate that only focusing on the number of alliances for a group, while an important first step, may have obscured the nature of the relationship with lethality. While relevant alliances increase group lethality, some alliances do not matter at all—or even detract from the efficiency and deadliness of attacks.

Among the control variables, the suicide bombing, size, and religious motivations have consistently positive and significant impacts across all models. These results are consistent with prior research. Organizational age, control of territory, and ethnonationalist motivations all have positive relationships with fatalities, but not fatalities per attack. It is intuitive that larger organizations kill more people, since the size of membership is perhaps the most basic indicator of strength. Similarly, organizational age has a positive relationship with casualties, likely because these organizations have had more time in which to inflict damage and because stronger organizations are more likely to survive for substantial periods of time. Those that control territory are more capable a priori, and have more opportunity to kill. Ethnonationalist organizations are likely to be more lethal due to the increased commitment that such groups are often able to engender in their memberships. Interestingly, in the context of this model of alliance relationships, none of the country-specific covariates appears to have a consistent relationship with the number of fatalities that terrorist organizations inflict.

Figure 3 demonstrates the substantive effect of terrorist group alliance depth on the casualties inflicted by those groups. We characterize the values of the alliance depth variable as “minimum” (0), “mean” (0.6), “high” (4.7), and “very high” (8.7) because the numerical values for the eigenvector measure are not intuitive. The “high” value is one standard deviation above the mean, and the “very high” measure is two standard deviations above the mean. For example, Aum Shinri Kyo (the Japanese organization that carried out the Tokyo subway gas attacks) has no known alliances and therefore receives the minimum score. The Tamil Tigers, in contrast, have several alliances and ended up with approximately the mean value for alliance depth. The Kurdistan Workers Party (PKK) in Turkey has a “high” alliance depth score, while Hamas has a “very high” alliance depth ranking. It is also worth noting that even the “very high” values reported in Figure 5 are not outliers. The data are right-skewed—twenty-six groups have greater alliance depth scores than Hamas, the highest group depicted in Figure 3. Therefore, these figures significantly understatethe substantive importance of our results.

OPEN IN VIEWER

Figure 3.

Predicted fatalities as alliance depth increases. “Minimum” = 0, “Mean” = 0.6, “High” = 4.7, and “Very High” = 8.7. The “High” value is one standard deviation above the mean. The “Very High” measure is two standard deviations above the mean. The dotted lines are 90 percent confidence intervals.

As alliance depth increases from its minimum (0) to a very high level (8.5), predicted fatalities increase from 12 to 72 and fatalities per attack increase from 1.25 to 8.25. These increases demonstrate the large substantive effect that alliances have on overall lethality and efficiency. Some readers may be concerned that outlier groups such as al-Qaeda, which are disproportionately lethal, may overly influence these results in general. To address this, we replicate the findings presented here (and obtain similar findings) for models with logged dependent variables as well as with models that exclude al-Qaeda and its direct allies. ²⁸

To further address the concern that these findings are not just the result of our model’s ability to predict the behavior of low-level terrorist groups, Figure 4 shows the relative accuracy of our predicted casualty levels compared to real-world casualty levels. Figure 4 specifically depicts the relationship between the log of actual fatalities inflicted by terrorist groups and the predicted log of fatalities.

OPEN IN VIEWER

Figure 4.

Comparing predicted and actual fatalities³⁰

The dispersion pattern indicates that the model fits the data well, especially as the number of fatalities increases. In fact, the model does an especially good job predicting the fatalities of some of the deadliest terrorist groups of the last twenty-five years, which is atypical of most work in this area. Our model captures the relative deadliness of al-Qaeda, Hamas, Hezbollah, and the LTTE. That being said, there are groups it does not accurately model. For example, the model predicts that Islami Chhatra Shibir would be much more deadly than they actually are. Determining why our model did not capture some of these groups is a promising avenue for future research. More generally, Figure 4 shows that incorporating terrorist alliances into models of terrorist group behavior, while limited in the ways described above, does significantly improve our knowledge of terrorist group behavior, and contribute to our understanding of alliances in general.