Abstract
Criminal and terrorist organizations often depend on repeat offenders to maintain the group’s longevity, especially after repeated law enforcement interventions. Yet, little is known about the offenders who perpetrate multiple incidents on behalf of a group. Relying on data for 118 terrorist offenders involved across eight attacks from 2000 to 2005, this study examines the correlates of repeat offending within a terrorist organization. Our main predictor, criminal social capital, is measured by the number and structure of co-offending ties. Poisson regression results demonstrate that offenders with a higher number of connections are more likely to be involved in multiple attacks; while offenders positioned as brokers—bridging otherwise unconnected others—are less likely to reoffend. In addition, being a leader and graduate education was associated with repeat offending. These findings suggest that selection is based on more than an offender’s skill set but also on their embeddedness within the group.
An offender’s illicit network has been identified as an important factor for understanding offending patterns across the criminal career. Offenders with higher numbers of co-offenders tend to reoffend more and persist for longer in their individual offending career, a finding that has been demonstrated across crime types (Piquero, Farrington, & Blumstein, 2007) and among older age groups (Lantz & Hutchison, 2015). However, studies have yet to examine how co-offending ties impact offenders’ continued involvement with a criminal organization. Social ties are not distributed equally across group members. Some offenders are more embedded within an organization, having a greater number of ties to other group members, while others are located on the periphery, having few connections to the overall group. Previous research has suggested that individuals who occupy more peripheral positions in illicit groups may be at a greater likelihood of desistance (e.g., McGloin, 2005). While individuals embedded in the group have been suggested to be more likely to remain, having greater exposure to criminal norms and access to illicit opportunities embedded in social ties (McCarthy & Hagan, 1995). In this sense, the group context matters, because offenders are dependent on not simply their willingness to be involved in future offenses but also on the opportunity to be recruited for a future offense. Although research has advanced our understanding of the structure of criminal network on offending pathways, the extent to which a person’s structural position influences their likelihood of committing multiple offenses on behalf of the group is unknown. An examination of the factors that allow members to be selected to commit offenses over time can provide additional insight into group processes, while also informing the potential for strategic interventions of the most prolific offenders.
To examine repeat offending in a group context, we turn to a terrorist organization—Jemaah Islamiyah (JI)—who conducted eight consecutive attacks over a 6-year period. Our focus on JI extends from the extensive network data available on the organization, but also the unique opportunity it provides to study reoffending across multiple attacks. As an enduring organization, JI was formally established in 1993 in Southeast Asia and has survived for over two decades. The group’s primary ambition has been to create an Islamic State within Indonesia, eventually encompassing all of Southeast Asia. Seizing an opening, after the fall of the Suharto regime, JI increased their presence in Indonesia, perpetrating a series of attacks in the early to mid-2000s. Continuity in members across eight attacks from 2000 to 2005 form the cornerstone of our analysis, allowing us to capture repeat involvement across stages in the group’s evolution and under intense government interdictions.
Building on previous co-offending research, we focus on variation in offenders’ co-offending ties and their structural position within the organization as predictors of repeat offending. The question at the heart of the study is whether individuals who come into their first attack already more socially embedded into the organization end up being the repeat offenders that help sustain future attacks. This approach is consistent with earlier studies that have demonstrated highly connected terrorist offenders are better able to access resources (Pedahzur & Perliger, 2006) and are more likely to remain committed to collective objectives (Stevenson & Crossley, 2014) both factors in a group’s sustained operations.
Factors Associated With Participation in Terrorist Attacks
What could be the factors involved in the selection of specific individuals to carry out an attack? As is the case with the successful coordination of many high stakes crimes (Bouchard & Nguyen, 2011; Cornish & Clarke, 2002; Lacoste & Tremblay, 2003; Malm & Bichler, 2011; McCuish, Bouchard, & Corrado, 2015; Morselli, 2005, 2009; Steffensmeier & Ulmer, 2005), we would look for a combination of trust factors and specific types of skills needed to carry out the attack. While previous research has primarily examined the selection of suitable co-offenders for a criminal event; for the purposes of this study, an important distinction needs to be made between an individual’s selection into their first participation in a terrorist attack and the possibility of being selected again for one or more attacks. The factors that are associated with first participation may differ from those associated with the second attack, after the individual had a chance to prove herself or himself. In other words, one’s performance during the first attack would be a key predictor of being provided with additional opportunities within a terrorist organization. Prior to the first attack, however, we posit that opportunities and resources provided through an offender’s social ties to an organization—who you know—and the perception of competence to carry out one’s role in the attack are some of the building blocks predicting participation.
Perception of competence has been shown to be an important factor in criminology to characterize offenders’ self-efficacy (e.g., Brezina & Topalli, 2012), but not necessarily in the context of recruitment of co-offenders, the way we intend to use it here. Perception of competence is important in a terrorism context because organizing attacks and maintaining operations over time require a range of skills and competencies for all the steps required in the crime commission process (Newman & Clarke, 2006; Shapiro, 2013). Terrorist attacks require skilled offenders not only at the time of the attack but also a range of actors beyond the incident, from those involved pre-event (e.g., bomb makers) to those that are required postevent (e.g., providers of safe houses; Koschade, 2006). Perception of competence may be appraised within informal arenas that provide an opportunity to demonstrate or acquire these skills, such as training camps. The typically longer preparation prior to terrorist attacks of the type conducted by organizations like JI may provide an opportunity to project one’s abilities to perform their roles.
However, performance during the actual offense provides for more accurate assessments of these factors. From this perspective, an offender’s first attack provides a legitimate forum to display their skill level and ability to contribute to an organization’s aims. An offender’s ability to perform under high-risk, high-stress contexts may weigh heavily in whether they are recruited for future attacks. Performing a task successfully highlights competency and trust, both highly valued in illicit contexts (Tremblay, 1993). Further, being selected multiple times also reflects an offender’s ability to avoid being detected and detained by law enforcement. Demonstration of these skills allows groups to organize themselves more effectively, recruiting or excluding members based on demonstrated experience or lack thereof. These crime-specific skills are generally transmitted through social ties, as formal mechanisms to improve criminal skills are not available (McCarthy & Hagan, 2001; Steffensmeier & Ulmer, 2005; Tremblay, 1993; Tremblay & Morselli, 2000). Thus, we would expect repeat offenders to be strategically positioned in their network to acquire this knowledge through fellow offenders. Connections may also serve a dual purpose, providing access to additional opportunities and connecting offenders to the individuals responsible for organizing future attacks. In a terrorism context, trusted ties provide both security and resilience, allowing groups to minimize the number of interactions between members while still maintaining cohesiveness (Krebs, 2002; Roberts & Everton, 2011; Sageman, 2004; Shapiro, 2013). Criminal social capital—the ability to use one’s social network for criminal outcomes—is expected to be a key predictor of selection in terrorist attacks. Below, we further develop on criminal social capital as a predictor of participation in multiple attacks.
Repeat Offenders and Criminal Social Capital
The repeat terrorist offenders considered in this study do not have perfect comparison points in more traditional criminological research. Yet, research on career criminals and the factors associated with persistence provide a useful guide in approaching this type of inquiry. Individuals who have been recruited to a terrorist organization, become a member, and participate in multiple attacks across many years as adults qualify under the terrorist career umbrella (e.g., Amirault & Bouchard, 2015). Previous research on repeat offenders and career criminals has identified co-offending ties as important factors in explaining offending patterns. Piquero, Farrington, and Blumstein (2007) identified a high number of criminal contacts as being associated with increases in both the frequency and duration of offending. Criminal contacts serve as a proxy measure of an offender’s criminal social capital. The number of available co-offenders may increase the number of illicit opportunities, providing a larger pool of accomplices to select or be selected from. Further, these same contacts may increase proficiency at crime, serving as social resources to transmit and receive criminal knowledge (Bouchard & Nguyen, 2011; Lantz & Ruback, 2015; McCarthy & Hagan, 1995; Morselli, Tremblay, & McCarthy, 2006).
Previous co-offending research has found that it is rare for offenders to reuse criminal contacts for future offenses. In the few instances where offenders do develop stable co-offending patterns (i.e., reuse the same contacts), these patterns have been argued to be associated with an offender’s skill set, including access to resources or opportunities that make them a valued asset (McGloin et al., 2008; Morselli, 2001). Lantz and Ruback (2015) recently demonstrated this tendency for co-offending contacts to share opportunities, finding that repeated targeting of the same burglary location, when not involving the same offender, often involved their connected co-offenders. Moreover, research has also demonstrated that access to more nonredundant criminal contacts (individuals who are not all directly connected) also increases one’s crime repertoire (McGloin & Piquero, 2010). The finding that individuals tend to be more diversified in their crime types when they co-offend has also been observed across aggregate data (Andresen & Felson, 2012a, 2012b).
The idea that criminal social capital facilitates offending by providing access to criminal opportunities is rooted in social learning theories of crime, such as Sutherland’s (1947) differential association theory. These theories align with Coleman’s (1988) more general argument that an individual’s structural position (i.e., how they are connected to others within their network) provides them with differential access to resources and opportunities embedded in social ties. While Coleman (1988) was referring to conventional social capital—licit opportunities embedded in social ties—criminal social capital refers to its illicit counterpart—illicit opportunities accessed through criminal contacts. The difference being conventional social capital inhibits offending (Sampson & Laub, 1993), and criminal social capital facilitates it (McCarthy & Hagan, 1995). Criminologists who tested the hypothesis that criminal social capital engenders beneficial criminal outcomes have focused either on access to lucrative crime opportunities (Descormiers, Bouchard, & Corrado, 2011; Morselli & Tremblay, 2004) or on detection avoidance (Bouchard & Nguyen, 2010). For instance, in market crimes, Morselli (2001) found that serving as a broker, bridging disconnected players along the illicit trafficking chain, provided access to more profitable transactions and hands-off roles that protected individuals from detection by law enforcement.
While brokerage can shape a criminal career, by providing access to lucrative opportunities, in high-risk violent crimes the sum of social ties may play an important factor in continuity. Making the decision to adopt violent measures to promote a political cause has been consistently linked to an individual’s social ties (Della Porta, 1988; Sageman, 2004). Social ties not only provide access to additional opportunities but have also been suggested to reinforce radical views, diffuse accountability for violence, and increase the costs of not engaging in violent behavior through social exclusion (McCauley & Segal, 1987). Hence, ties to other offenders may play a salient role in offending, cementing radical beliefs, while providing additional opportunities for action. However, despite serving a number of practical purposes for terrorist organizations, high connectivity may also create risks for offenders. Previous studies have found that high connectivity within illicit networks increases an individual’s exposure and risk of detection (Baker & Faulkner, 1993; Morselli, 2010). Individuals with a high number of connections to a group may have greater commitment and opportunities that could facilitate continued involvement; however, these same connections come at a cost, potentially increasing the risk of detection.
Current Study
To examine the role of criminal social capital in explaining variation across individual’s selection for future attacks, this study analyzes 118 offenders across eight attacks perpetrated by JI in Indonesia. The series of attacks provide us with a unique opportunity to study patterns in selecting terrorist co-offenders. A majority of the JI members were only selected once, but some were involved in as many as six or seven attacks. The question at the heart of this study is whether there are clear differences between the single attack and the repeat terrorist offenders. We posit that criminal social capital—here measured by an offender’s co-offending ties—is likely to be a predictor of participation in multiple attacks. The availability of network data allows us to systematically analyze the structural position of each terrorist offender in the overall network. The hypothesis that an offender’s network size facilitates selection and willingness to be involved in multiple attacks can be tested. The study also considers competence in the form of human capital—highest level of education—along with criminal capital—occupying a leadership position and experience as a militant—as potential confounders in predicting repeat terrorist offenders.
Data and Method
Terrorist Networks and Attributes
To measure the impact of criminal social capital on repeat offenders, data on 118 terrorists involved in at least one of the eight attacks were collected from open sources, including the John Jay and ARTIS Transnational Terrorism (JJATT) database (Atran et al., 2008). JJATT is a public online database that provides network and attribute data for offenders involved in over 20 al Qaeda–affiliated terrorist attacks. 1 For the current study, data on five of the eight attacks perpetrated by JI in Indonesia were available on the JJATT database, while the remaining three attacks were collected from open sources. 2 Data collected from JJATT relied not only on court transcripts, which has been cited as one of the most reliable methods of acquiring terrorist-related data (Freilich, Chermak, Belli, Gruenewald, & Parkin, 2014; Sageman, 2004) and national news reports, but was also supplemented by primary sources such as photos, letters, and interviews (Magouirk, Atran, & Sageman, 2008, p. 4) and has been used across numerous studies (e.g., Magouirk & Atran, 2008; Magouirk et al., 2008; Helfstein & Wright, 2011a, 2011b). Through these sources, JJATT mapped the network of every offender identified as being involved in the attack, with ties coded as present if offenders had been in contact, through phone calls, letters, or in-person meetings. Hence, the current definition of co-offending ties diverges from previous research, which defines a tie between two offenders according to their participation in the same crime event (e.g., Andresen & Felson, 2012a, 2012b; Lantz & Hutchison, 2015; McGloin & Piquero, 2009, 2010). Rather, the current study operationalizes a co-offending tie as a recorded interaction between two offenders who also participated in the same event (i.e., attack). Given the scale and scope of terrorist attacks, requiring a range of offenders directly involved in executing and attack, and others indirectly in its coordination, means that many of the individuals participating in the same event are not directly connected or even aware of other coparticipants. Thus, this definition allows us to capture variation in the structural position of individuals—and thus their criminal embeddedness—across participation in the eight attacks.
For the three additional attacks perpetrated by JI during this same period (2000–2005) in Indonesia, data were derived from open sources using two main strategies. First, reports familiar to the authors on JI, including two International Cisis Group reports, were consulted (International Crisis Group, 2002, 2006). Second, a systematic search of the literature using open sources was conducted. This search relied on the web-based search engines Google and Google Scholar using logical combinations of different key words such as JI, terrorism, attack, bombing, and key words related to each attack (e.g., Bali; Australian Embassy) to collect a wide range of sources, including books, journal articles, and media sources. To maintain consistency, coding procedures followed those described for the JJATT database. Only core members and those offenders that directly contributed to the attack network were included in the data. Lacking access to primary sources precluded us from directly modeling the data collection procedures used by JJATT. However, when actors in the additional three attacks had also been involved in the five attacks listed by JJATT, we relied on this latter source to code ties across the three additional attacks. For instance, if actors were connected in previous attacks (according to the JJATT database), we also coded them as being present in the three subsequent attacks. 2
Across the attacks, 118 offenders were involved in at least one of the eight incidents. In order to examine participation in multiple attacks, 23 individuals were excluded from the sample either because their first attack did not occur until the eighth attack (n = 17) or were involved in a suicide mission (n = 6). This created a final sample of 95 offenders who had an opportunity to participate in a subsequent attack.
The eight attacks in the sample were perpetrated from 2000 to 2005 (Table 1). The first three attacks occurred a few years after the fall of the Suharto regime, during the time JI was increasing their presence in Indonesia. These attacks were perpetrated in the early 2000s and focused on domestic targets, to put pressure on the Indonesian government (e.g., a 2000 Christmas Eve bombing involved 39 coordinated explosions at different religious institutions, a 2000 bombing of a Philippine Ambassador’s residence in Indonesia, and two bombings in 2003 and 2004 at Christian places of worship). However, post-9/11, the group’s focus turned to Western targets, which included the targeting of three tourist facilities. This includes the 2002 Bali bombing of a nightclub, a 2003 bombing of the Marriott Hotel in Jakarta, a bombing of the Australian Embassy in 2004, and a second bombing of the Bali tourist district in 2005. Although domestic pressure in the early 2000s initially hindered the government from implementing repressive countermeasures, the high fatalities caused by the 2002 Bali bombings created a backlash among the public, with authorities creating an antiterrorism task force that resulted in substantial arrests of suspected members and one of the core leaders (National Counterterrorism Center, 2013). These eight attacks form the cornerstone of our analysis, allowing us to capture repeat involvement across stages in the group’s evolution and under intense government interdictions.
Eight Attacks Perpetrated by Jemaah Islamiyah–Affiliated Offenders in Indonesia.a
Note. JJATT = John Jay and ARTIS Transnational Terrorism.
aAttack-specific information was obtained from the National Consortium for the Study of Terrorism and Responses to Terrorism’s online Global Terrorism Database available at http://www.start.umd.edu/gtd/.
We use the co-offending ties across offenders involved in the eight attacks to construct a single network of the organization (Figure 1). Repeat offenders serve to bridge the eight attacks, with experienced offenders creating continuity in co-offending ties over time. These repeat offenders are represented by black nodes and onetime offenders by white nodes. 3 The links that connect each offender represent the presence of a co-offending tie. Offenders’ placement within the network is done using the multidimensional scaling feature in the ORA (Version number 3.0.9.9.36) software suite (Carley et al., 2013), which positions actors according to similar patterns of relations. That is, offenders who share the same set of co-offenders are placed in proximity to one another, while offenders who are different in terms of their set of relations are positioned at greater distances from one another. This network visualization shows that repeat offenders appear to be more central to the network, not only having a higher number of co-offending ties to other offenders, but specifically to repeat offenders. However, this does not appear to be consistent across all repeat offenders, with a few black nodes positioned on the periphery of the network reflecting few co-offending ties. In addition, the network visualization captures offenders’ aggregate set of co-offending ties across attacks; hence, it should be emphasized that repeat offenders are likely to have acquired additional co-offending ties across subsequent attacks. Recognizing this, our analysis only uses measures of network features at time of first attack for all 95 offenders included in the sample. In this way, the potential bias arising from offenders building their network of contacts across multiple attacks is removed, creating a criminal social capital baseline that is comparable across offenders.
Co-offending ties across attacks (2000–2005).
Measures
To assess factors associated with repeat offenders, a number of variables were constructed across the 95 offenders. A full list of these variables and their descriptive statistics are provided in Table 2.
Characteristics of Terrorist Offenders (Imputed Data).
Note. N = 95.
Number of attacks
The number of attacks an individual participated in was used as the outcome variable in the analysis. While JJATT only supplied network information on five of the attacks included in this analysis, the dataset did provide information on whether individual offenders involved in one of these five attacks also participated in additional incidents beyond these attacks (e.g., attacks outside of Indonesia). Thus, if an offender within one of these five attacks also participated in another attack not included in the listed attacks it was included as one of their total attacks. This same procedure was used to collect data for offenders involved in the three additional attacks derived from open source searches. That is, the selected attacks were only the starting point for coding whether offenders were involved in multiple offenses. While all repeat offenders in the sample participated in at least two of the eight attacks, some had participated in an additional attack outside of the incident sample.
Independent Variables
Criminal social capital
To examine the effect of criminal social capital on repeat offending, two network measures reflecting this concept were examined: degree and betweenness centrality. The first, degree centrality, is a simple measure of the sum of direct connections a member has with other members within the network (Wasserman & Faust, 1994). This measure has been suggested to capture an individual’s importance and influence in a network, reflecting their degree of activity and connectivity within the organization (Morselli, 2009; Sparrow, 1991), and more specifically, a greater ability to efficiently disseminate information across members of terrorist groups (Koschade, 2006; Pedahzur & Perliger, 2006). Thus, individuals with high degree centrality may not only have greater access to opportunities through their contacts but may also have a vested interest in maintaining the continuity of the organization. All measures of degree centrality were calculated at the time of first attack. For instance, using the network data on offender’s co-offending ties, if an offender only knew 3 of the 15 other offenders in the first attack, they would have a degree centrality of 20%. Taking an offender’s first offense allowed for a baseline measure across offenders, so as not to discriminate between measures of criminal social capital for those who only take part in one attack and those with criminal social capital acquired across multiple attacks. High degree centrality may increase an offender’s likelihood of being selected for a second event, with connectivity suggesting not only embeddedness within a group right from the start but also greater access to opportunities.
Recognizing that mixed evidence has been found regarding the influence of the number of connections on an individual’s frequency of offenses, with higher ties potentially leading to an increased likelihood of detection (Baker & Faulkner, 1993; Morselli, 2010), a second measure of criminal social capital, betweenness centrality, was also included. Providing a more refined measure of centrality, betweenness centrality, accounts for the fact that not all connections produce the same benefits—tapping into the quality of contacts rather than mere quantity of contacts. Providing a proxy for brokerage, betweenness centrality, measures the degree to which actors serve as bridges between unconnected others (Wasserman & Faust, 1994). Betweenness centrality is measured by calculating the degree to which an actor connects other actors through the shortest path (geodesic). Actors located along many geodesics in the network have higher betweenness centrality scores (Freeman, 1977). Individuals high in brokerage are important in that they connect otherwise unconnected members, controlling the flow of connectivity and information in the group, thus increasing their value and potential selection for a second offense. Consistent with degree centrality, all measures were taken at the time of first offense.
Criminal capital
Sharing a complementary relationship with criminal social capital, criminal capital refers to the crime-specific skills and expertise an individual has acquired (Bouchard & Nguyen, 2010; Loughran, Nguyen, Piquero, & Fagan, 2013; Lussier, Bouchard, & Beauregard, 2011; McCarthy & Hagan, 2001). This measure accounts for the fact that offenders who have gained experience in committing attacks may be more proficient. Two variables reflecting this expertise were used: belonging to the group’s central staff and previous militant experience. Importantly, none of these measures require that offenders show their skills and competence in the context of one of the eight JI attacks studied here. These measures capture the perception of competence we were looking for as a predictor of selection for a first attack.
A leadership position within a terrorist network may reflect a certain degree of competency that allowed them to attain this position. Leaders of terrorist organizations have been cited to be highly skilled actors, requiring a degree of knowledge to coordinate and maintain the illicit group (Hoffman, 2004; Stern, 2003). Thus, these members may not only have acquired the necessary specialized knowledge or experience to occupy this position but may also be among those who are most involved in coordinating attacks. The restricted nature of this position was highlighted within JI, with only a fraction of all offenders holding a position as central staff (12%; n = 11). The central staff position was static, with members assessed as holding this status for the entire duration of their involvement in the eight attacks. While few offenders held central staff positions, many of the offenders selected for one or more of the eight attacks had previous experience in participating in militant operations, which may have impacted their recruitment. Given the political environment in Indonesia, specifically, the degree of conflict during the 1990s and into the 2000s, it is not surprising that the majority of offenders had previous militant experience (76%).
Human capital
Human capital is criminal capital’s conventional counterpart and refers to an individual’s personal attributes derived from experience and training that contributes to career achievement (Becker, 1962). Despite the lack of support for human capital in profit-driven crimes (McCarthy & Hagan, 2001), this form of conventional capital found support in Lussier et al.’s (2011) examination of sexual offending. It may also be pertinent for terrorist offenses. For example, education has been positively associated with terrorist success, with Benmelech and Berrebi’s (2007) finding that bombers with higher education had more success in committing suicide attacks. While it is evident that these actors are precluded from selection into future attacks, these findings suggest that education provides a specialized skill set to conduct successful operations, making these members more valuable to the overall group. This is consistent with studies that have found that suicide bombers are typically recruited as “cannon fodder,” consisting of less educated members (e.g., Ganor, 2000; Pedahzur, Perliger, & Weinberg, 2003; Weinberg, Pedahzur, & Canetti, 2003). Lacking education, these individuals may be considered less valuable to the group and thus more disposable.
Offenders’ highest level of education acquired, a proxy measure of human capital (e.g., Lussier et al., 2011), was included in the analysis. Due to low frequencies for three categories, “certificate” (n = 1), “some graduate” (n = 3), and “doctorate” (n = 2), these categories were merged with others. Certificate was merged with “some college” and doctorate with the “master’s degree” and some graduate categories. This created five categories, with the majority of offenders having little education beyond high school, while a select few had attained graduate-level studies.
Control variables
Two control variables were included in the analysis: participation in first attack and participation following an intervention. Participation in first attack was included in the analysis to account for some individuals who were simply not on the radar until much later on in the time period. Given that attacks are being analyzed for a specific time frame, individuals who participated earlier on have a higher chance of participating in multiple attacks. An intervention variable controlled for the fact that the organization operated with relatively limited law enforcement interference from 2000 to 2002. During this period, the Muslim majority of the country were resistant and skeptical of government antiterrorism efforts. However, the 2002 Bali bombings served as a turning point for support of antiterrorism policies. In October 2002, JI suicide bombers detonated explosives in the densely populated tourist district of Bali causing 202 fatalities and approximately 300 injuries. The high number of casualties created a shift in antiterrorism interventions and in July 2003 detachment 88—an antiterrorism task force—was created with support from the United States and Australia. This Indonesian Police Unit conducted large-scale arrests and killings of terrorist offenders in the months and years that followed (Everton & Cunningham, 2015). To capture the effect of this unit, an intervention variable was created indicating offenders whose first attack occurred either directly before the implementation of this agency (capturing the high number of individuals who were arrested after the 2002 Bali bombings) or any time after its implementation (2002 and onward).
Analytic Strategy
Multiple imputation methods were used to account for missing values in five of the independent variables: education (21%; n = 20), militant experience (19%; n = 18), degree centrality (3%; n = 3), betweenness centrality (3%; n = 3), and central staff (1%; n = 1). Bivariate analyses demonstrated that missing values were distributed across cases, with most cases only having missing values for one variable. One exception was education and militant experience, with multiple cases having missing values for both variables. This may have implications for the imputation, potentially introducing systematic regularities in the imputation of these variables. However, given the random distribution of missing values across the full set of variables, imputing data allowed us to conserve all individuals and variables in the analysis, providing a more comprehensive assessment of repeat offenders and their predictors, as well as reducing the loss of power that occurs from excluding variables. Multiple imputation aims to restore error variance by reflecting the variability that would be found in the original data, minimizing bias in the estimation of parameters (Allison, 2000). The variables used to conduct the multiple imputation included all independent variables in the analysis and the outcome variable. A total of 10 imputations were conducted, and the pooled results are reported.
To examine the impact of criminal social capital on repeat offending, we ran two Poisson regression models. Poisson regression was used, given that the outcome variable (number of attacks) comprised counts of a rare event. As a baseline, the first model examined the control variables (participation in first attack and the impact of the intervention), human capital (educational achievement), and measures of criminal capital (central staff and militant experience). Model 2 added in the measures of criminal social capital (measures of degree and betweenness centrality) to examine its impact on repeat offenders. All analyses were run in SPSS version 22, and robust estimators were used to account for violations of underdispersion and independence of observations. First, the outcome variable suffered from underdispersion as indicated by a deviance/degrees of freedom ratio that deviated significantly under one. Second, our measures of criminal social capital—an offender’s centrality within the network—violated the assumption of independence of observations. The robust estimator provided a means to account for covariance properties of the errors and observations, providing more consistent estimates of the standard errors when dependence is present.
Results
The repeat offenders in our sample counted for nearly half of all offenders involved across the eight attacks (44%; n = 42). These repeat offenders are not homogeneous, with some limiting their participation to two attacks and others involved in up to seven attacks. On average, the repeat offenders were involved in 2.9 attacks (SD = 1.3), with most involved in only two attacks (54%; n = 23). Fewer repeat offenders are involved in three attacks (21%; n = 9) and even fewer in four attacks (13%; n = 6). Lastly, only four offenders are involved in at least five attacks, with two offenders in six attacks, and a final offender in seven attacks.
In addition, repeat offenders are distributed according to when they first participated with the organization. Table 3 distinguishes between repeat offenders who were involved from the very first attack and those who didn’t enter until later on in the organization’s trajectory. Almost all offenders who were involved in three or more attacks were involved from the very beginning (68%). These offenders serve to link the eight attacks, with experienced offenders creating continuity in membership across attacks. In contrast, offenders who didn’t enter until later on were typically only involved in two attacks (91%). These findings may suggest two scenarios. One, the possibility of a cohort effect, with the formation of a cohort of offenders who met at the beginning and follow each other across attacks. In contrast, the two-time offenders appear to have been brought in as needed—representing a pool of potential affiliates who were not consistently selected for attacks. This is supported by the fact that half of the offenders who were only involved in two attacks were brought in at different time points. Alternatively, offenders who are only involved in two attacks may reflect a lack of opportunity, having entered near the end of the group’s trajectory, when the group was more heavily targeted by government antiterrorism forces.
Distribution of Repeat Offenders Across Attacks.
Note. N = 42.
To examine whether there were any differences between the repeat and onetime offenders in our sample, we first run a series of bivariate tests. Table 4 demonstrates the results of the bivariate analysis, finding that repeat offenders were significantly more likely to be involved in the first attack perpetrated by the organization (p < .001) and were less likely to consist of offenders whose first attack happened after the government increased their targeting of the group. In terms of education, repeat offenders were significantly more likely to have a graduate-level studies education as compared to onetime offenders (p < .01). In addition, offenders involved in multiple attacks primarily consisted of central staff members (p < .001). In contrast, individuals with militant experience were no more, no less likely to be involved in multiple attacks. Repeat offenders were also more likely to have a high number of connections to other offenders within the attack network (p < .001). Thus, repeat offenders were more likely, on average, to join the organization at the time of first attack, have a higher number of contacts within the attack network, be part of the group’s central staff, and/or possess a higher level of education.
Bivariate Analysis of Onetime and Repeat Offenders.
aAll bivariate analyses were run using Mann–Whitney U tests or Spearman’s ρ. bThe values represent the median degree and betweenness score.
A series of Poisson regression models were conducted to examine the impact of these predictors on the number of attacks in which offenders were involved (Table 5). As a baseline, the first model examined the effect of all independent variables with the exception of criminal social capital. Consistent with the bivariate analysis participating in the first attack increased an offender’s chance of being involved in subsequent attacks (b = .44; SE = .15; p < .01). In addition, the implementation of the antiterrorism task force—intervention—was negatively associated with the number of attacks (b = −.29; SE = .14; p < .05). Educational achievement also influenced the likelihood of being a repeat offender. Using no high school as the baseline category, results demonstrated that a graduate-level education (b = .54; SE = .23; p < .05), a bachelor’s degree (b = .34; SE = .16; p < .05), and a high school diploma (b = .29; SE = .13; p < .05) were positively associated with the number of attacks in which offenders were involved. However, possessing a college education or a certificate failed to attain statistical significance relative to the reference category. In terms of criminal capital, being a central staff member (b = .48; SE = .12; p < .001) was positively associated with the number of attacks in which offenders were involved, in contrast to militant experience which was not a significant factor.
Poisson Regression of Number of Attacks.
Note. N = 95.
a“No high school” represents the reference category for education level.
†p < .10. *p < .05. **p < .01. ***p < .001.
Model 2 added in measures of criminal social capital to test whether the number or structure of co-offending ties were correlates of repeat offending. Results show that having a higher number of co-offending ties was associated with multiple attacks (b = 1.20; SE = .12; p < .01). In contrast, offenders who occupied brokerage positions—bridging unconnected offenders—were less likely to be involved in multiple attacks (b = −3.09; SE = 1.5; p < .05). Thus, controlling for other factors, offenders that had more connections to the network at the time of first participation were more likely to participate in subsequent attacks, while offenders who were brokers were less likely to be involved in future attacks. In addition, the variables from Model 1 maintained their significance, with intervention (b = −.38; SE = .13; p < .01), graduate-level studies (b = .44; SE = .21; p < .05), high school diploma (b = .31; SE = .14; p < .05), and central staff (b = .44; SE = .12; p < .001) all being associated with repeat offending. One exception was participated in first attack, which did not remain significant when measures of criminal social capital were included. In sum, Model 2 demonstrates that repeat offenders were more likely to have a higher number of connections to other members within the attack network, less likely to be brokers, more likely to occupy a central staff position, and to be better educated.
Ties That Repeat
The above analysis suggested that a high number of contacts at first participation was associated with repeat offending later on. But is it simply about how many people you know? Given the variety of offenders involved in any given attack, it may be that certain types of ties may facilitate involvement in multiple attacks.
Taking an offender’s first attack, we classify the connections they made based on three possible types: (1) connections to experienced or “repeat” offenders at time of the attack, (2) connections to first-time offenders like themselves but who would remain onetime offenders, and (3) connections to first-time offenders who would eventually be recruited in subsequent attacks and become repeat offenders.
Making a distinction between ties to repeat offenders who have previously been involved in attacks and repeat offenders who will be involved in future attacks allows for the potential identification of a “cohort effect”—that is, whether these connections “followed” repeat offenders and were maintained in subsequent attacks. These additional analyses, presented in Table 6, examined the degree to which both onetime and repeat offenders were connected at the time of their first attacks.
Types of Co-offending Ties Across Onetime and Repeat Offenders.a
Note. aAll measures of co-offending ties were calculated at the time of first attack. Analyses were run in UCINET to account for the dependency between actors.
The results suggest that participation in multiple attacks may be influenced by more than the sum of an offender’s ties. Connections to other first-time offenders who would become repeat offenders was positively associated with involvement in multiple attacks. Table 6 demonstrates that repeat offenders were more likely to be connected to offenders who would be involved in subsequent attacks than to onetime offenders (p < .001). However, being connected to experienced repeat offenders—who were previously involved in attacks—did not distinguish between onetime and repeat offenders. In addition, onetime offenders were more likely to be connected to other onetime offenders (53%), while repeat offenders were less likely (24%; p < .001). These findings are consistent with the distribution of repeat offenders in our sample, which suggests that offenders involved at the beginning followed each other across subsequent attacks. The data do not allow us to discern whether our repeat offenders simply knew more of these future repeat offenders even prior to participation in their first attack. Yet, the findings do suggest the presence of clusters of offenders who participated across attacks as a cohort.
Discussion
This study examined the role of criminal social capital in explaining selection for future terrorist attacks. Results demonstrated that criminal social capital, measured by the number of terrorist members you know, predicted participation in multiple attacks. However, criminal social capital was not the only driver of repeat offenders, with central staff members and offenders with graduate-level education also more likely to participate in multiple attacks. This suggests that the selection is based on more than an offender’s skill set but also their embeddedness within the group.
The results regarding criminal social capital and repeat offending in a terrorism context are interesting in part because the size of the network was measured at time of an offender’s first attack, not based on a cumulative advantage that would emerge from repeat participation. Offenders appeared to come in with an advantage or to have been able to develop this criminal social capital leverage in the process of planning and undertaking the attack. Although applied to a new context, these results are consistent with extensive research that underscores the central role of peers in influencing offending patterns more generally (Akers, Krohn, Lonza-Kaduce, & Radosevich, 1979; Matsueda & Anderson, 1998; Piquero et al., 2007; McGloin & Piquero, 2009), but also within terrorist networks, where high connectivity has been shown to reflect high degrees of activity (Sangal, Martin, & Carley, 2012) and commitment (Stevenson & Crossley, 2014). This is reinforced by the finding that clusters of offenders participated with one another across attacks, suggesting that repeat involvement with the same co-offenders may have cemented relationships, and strengthened available opportunities for continued participation.
Yet, terrorist networks do not simply provide just another context. Diverging from profit-oriented crimes (see also Lussier et al., 2011), these offenders first and foremost share violent political ideologies, which provides an even more important role for the number of co-offending ties one has within the organization. A high number of ties to an organization may not only provide more opportunities but also assist in solidifying radical beliefs. Being connected to like-minded extremists may reinforce current views and maintain these radical ideologies over longer periods. In addition, these ties may amplify the costs of leaving, with highly connected desistors potentially facing higher social costs.
This may assist in explaining why offenders who were positioned as brokers within the network were less likely to continue participating in attacks. That particular finding contrasts with previous research on brokers. Using individuals high in betweenness centrality as a proxy for “instigators” of criminal events, Lantz and Hutchison (2015) demonstrated that offenders positioned as brokers start offending at an earlier age and conduct more offenses. The main argument being that brokers (or recruiters) are generally high-rate offenders who offend with low-rate offenders. In contrast to their sample of burglars, individuals who occupied these positions within the terrorist network were found to be more low-rate offenders. This may be related to the unique context of terrorist offending. While brokers are valuable in profit-motivated offenses, bridging supply and demand chains or access to illicit opportunities, in a terrorist context embeddedness in a network may be more important to maintaining ideological orientations necessary to persist. This is consistent with major theories of terrorism that suggest transitions into terrorist groups require small densely, connected groups of offenders (e.g., Sageman, 2004; Nash & Bouchard, 2015), and empirical research that has found offenders who continue with a group over time are more likely to be central, and become more central over time (Stevenson & Crossley, 2014). Lastly, the current study’s finding of a cohort effect supports the importance of embeddedness, showing stable sets of repeat offenders followed each other across attacks.
The results also showed that repeat offenders were more likely to occupy a position as a central staff member, one of our measures of criminal capital. Individuals who have acquired a leadership role not only have the decisional power to get involved in subsequent attacks but have clear incentives to do so to further advance their own cause. Leadership roles have been stated to reflect a commitment to the group (Crenshaw, 1981) and acquisition of a certain degree of experience and knowledge (Carley, Lee, & Krackhardt, 2002). Further, it is worth noting that central staff members involved in multiple attacks also had a higher number of co-offending ties (degree centrality: .50) relative to repeat offenders that did not hold this position (degree centrality: .41), providing the necessary social resources to initiate and organize attacks. 4 This finding also indirectly reveals that central staff members were available for participation in multiple attacks, which suggests that they are also less likely to be involved in roles that physically expose them during the attack, such as bombers or foot soldiers, limiting their risks of detection.
It is worth noting that only one measure of criminal capital, occupying a role as a central staff member, emerged as significant while the second measure, militant experience, was not found to be associated with repeat offenders. While it is possible that experience plays no role in repeat terrorist offending, we feel that the value of experience is dependent on context. In the case of JI, most offenders had acquired some form of militant experience (76%) and thus, given its ubiquity, may not have been as important a criterion when selecting future recruits. In contrast, few offenders occupied a position as central staff (12%). 5 The fact that most offenders had experience does in fact tell us that this appears to very important for selection purposes, so much so that it does not emerge as a key characteristic to predict inclusion in multiple attacks.
Finally, we found that human capital may play a key role within JI, as individuals with graduate-level studies were more likely to be involved in multiple attacks. This extension of human capital into terrorist activities may be attributed to the degree of organization required to conduct elaborate attacks or gain access to legal venues. This is consistent with research on 148 Palestinian suicide bombers by Benmelech and Berrebi (2007) that found university educated bombers were better able to evade detection prior to detonation and target more prized locations. While suicide bombers are typically not repeat offenders (by virtue of their task), this does suggest that they possess a valued skill set and is reinforced by the fact that suicide bombers are typically disposable members that lack education (Ganor, 2000; Pedahzur et al., 2003; Weinberg et al., 2003).
There are a few limitations that should be taken into account when assessing these findings. The analysis of repeat offenders only examined a case study of members of JI, which has been noted to have a distinct network structure that more closely resembles a hierarchy relative to other Islamic groups (Sageman, 2004). With many senior members having received military training in Afghanistan in the early 1980s and late 1990s, the organization has been stated to take on a more formal command structure. In addition, members have been recruited through Islamic school settings (International Crisis Group, 2003). This may not only impact findings on measures of network centrality measures but means findings should also be interpreted within the unique context of JI, which operated in a distinct environment where recruits were often embedded in school environments and connected to others within a hierarchical organization. Thus, further analysis should be conducted across different organizations to examine generalizability. Related to this, this study only examined eight attacks for a specific time period from 2000 to 2005 and thus lacks information on terrorists that participated outside of these eight attacks. While the analysis attempted to be as comprehensive as possible, given the covert nature of attacks, members and some attacks themselves, may have been missed. Thus, individuals who were coded as being involved in the first attack may have been involved in earlier attacks. In addition, members that were coded as being involved in a single attack may have participated in further attacks at a later time period. The data precluded verification of these possibilities.
Further, in terms of analysis, our models only allowed us to capture how an individual’s social embeddedness within an organization at the time of onset influenced the number of attacks an individual participated in but not their longevity with the group. Lacking data on an individual’s end date with an organization, we were unable to properly model the duration for which members stayed. It is possible that individuals remained with the organization but did not involve themselves in future attacks. Future studies may look at members’ continuity with a group and how variation in their structural position over time (e.g., become more or less embedded) influences their survival with the organization. Lastly, in terms of data, the sample size is worth noting. On one hand, it is large—there are few occasions to study the networks of terrorists across eight attacks, with LaFree, Dugan, and Miller (2015) finding that nearly half of all terrorist organizations perpetrate a single attack. On the other, it is quite small, with only 95 offenders representing a smaller sample than ideal for multivariate analyses.
Despite these limitations, the study is among the first to establish a potential set of predictors to assess multiple offenses among terrorists. The findings also have value when thinking about the targeting of terrorist groups and the characteristics of individual members. Distinguishing between repeat offenders and their onetime attack counterparts permits effective allocation of limited resources, while allowing for the creation of tailored strategies to dismantle these groups. This study suggests strategies that focus on offenders who may be the most embedded and thus committed to continuing and maintaining the group through multiple attacks may be the most effective. Focusing resources on individuals with high connectivity serves two purposes: It assists in dismantling the network by taking away those with the most social ties and may serve to reduce the potential lethality of future attacks by removing offenders that are more likely to have acquired experience and perpetrate future attacks. Adopting an approach that examines networks to identify the most connected aligns with research that demonstrates network approaches play an important role in desistance strategies for terrorists (Noricks, 2009). For terrorist offenses, mapping networks can assist in the identification of offenders whose removal would most effectively disrupt the organization. However, the most effective disruption strategies aim to take into account both an offender’s network centrality and their value to the network, often measured as their role or resources they bring (Carley et al., 2003; Roberts & Everton, 2011). This idea has been referred to as “network capital” accounting for an offender’s network importance in terms of both the intangible and tangible resources they bring to the network (Schwartz & Rouselle, 2009; Westlake, Bouchard, & Frank, 2011). Thus, an offender’s structural position along with their skill set should be accounted for to develop the most effective targeting strategies.
Conclusion
In a terrorism context, ties to other offenders play a salient role in providing opportunities. Covert environments reduce the pool of potential accomplices, placing a high importance on criminal social capital. In addition, the skills offenders’ bring to a group whether as central staff members or having acquired advanced education demonstrate that an individual’s value plays a role in selection to multiple attacks. Thus, not only an offender’s network position but also the context of their connections may assist in explaining terrorists’ trajectories and offending patterns.
While an individual’s degree of connectivity to a group should not be underestimated, repeat offending should not be reduced to the sum of an offender’s social ties. Being a repeat offender reflects two conditions: (1) selection for a future offense and (2) willingness to be involved in a future offense. The finding that connections, central staff and level of education are all drivers of repeat offenders, may satisfy both these conditions. Offenders who have a high number of connections and occupy positions of status within a group may be more willing to continue participating, as they face greater costs should they choose to leave the group: losing both these social ties and the associated benefits of status. Consideration of these findings collectively suggests that selection to a terrorist organization may be a rational process, with individuals and groups weighing the benefits and costs of selection to further attacks. This approach is consistent with recent studies that have emphasized the rational processes of terrorist attacks (e.g., Hsu & Apel, 2015; Perry & Hasisi, 2015). Under this framework, a high number of ties to an attack network may serve to alter the perceptions of costs of behavior. Individuals who would not otherwise engage in a behavior may be more likely to do so in the presence of a high number of co-offenders; diffusing responsibility and increasing the costs of leaving. In addition from a group perspective, recruiting central staff and educated offenders with experience and specialized skills, aims to increase group capacity. This potentially suggests for future studies to look at selection in a terrorist organization as a rational process to understand terrorists’ trajectories and offending patterns. Informing us of the cost and benefit structuring of decision-making in terrorist offenses, this framework could provide crime-specific measures that decrease costs associated with desistance and target those most at risk of repeat offending.
Footnotes
Authors’ Note
The content is solely the responsibility of the authors and does not represent the official views of Public Safety Canada.
Acknowledgments
The authors would like to thank Cameron McIntosh, Karl Klockars, and Brett Kubicek for their helpful comments on an earlier version of this article.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by Public Safety Canada as part of the Kanishka Project.