Master Thesis
Claiming Responsibility for Terrorism: The Islamic State
Willemijn Bertels Student number: 1367315
Master Crisis & Security Management Faculty of Governance and Global Affairs
Institute of Security and Global Affairs Leiden University
Thesis supervisor: Dr. Bart Schuurman Second reader: Dr. Tahir Abbas
June 9, 2019 Total word count: 18.522
Acknowledgements
I would like to thank Dr. Schuurman for his clear and coherent supervision on this thesis, and Dr. Abbas for his keen eye in assessing this document. Moreover, I am very grateful to my loved ones for the continued patience and moral support I received during this research process.
Table of contents List of tables 4 List of figures 5 Abstract 6 1. Introduction 7 1.1 Claiming terrorism 7
1.2 The Islamic State 8
1.3 Societal and academic relevance 10
1.4 Research question and hypotheses 11
1.5 The concept of terrorism 11
1.6 Reading guide 12
2. Theoretical framework 13
2.1 The communicative nature of terrorism 13
2.2 The strategic logic of responsibility claims 13
2.3 Variation in responsibility claiming 15
2.4 The Islamic State’s claiming behaviour 18
2.5 Added value of this research 19
3. Methodology 21 3.1 Research design 21 3.2 Case selection 21 3.3 Operationalization 23 3.3.1 Dependent variable 24 3.3.2 Explanatory variables 24
3.4 Methods of data collection 29
3.5 Methods of data analysis 30
3.5.1 Logistic regression 30
3.5.2 Assessing the model fit 31
3.5.3 Interpretation of the coefficients 32
3.6 Assessment of validity issues 33
4. Results 36
4.1 Description of variables 36
4.2 Logistic regression analysis 38
4.2.1 Democratic constraints 41
4.2.2 Competitive context 41
4.2.3 Signalling strength 41
5. Discussion 44
5.1 Conclusion 44
5.2 Discussion of research findings 45
5.3 Reflection on limitations 46
5.3.1 Use of the GTD 46
5.3.2 Explanatory factors 47
5.3.3 Features of the Islamic State 48
5.4 Suggestions for further research 49
References 51
Appendix A: Syntax commands 55
Part 1: Case selection process 55
Part 2: Recoding of variables 57
Part 3: Regressions 63
List of tables
Table 1. Operationalization scheme 27
Table 2. Responsibility claims by year 36
Table 3. Distribution of data on the categorical fatalities variable 37 Table 4. Logistic regression results predicting responsibility claiming 40
List of figures
Figure 1. Flow scheme of case selection 24
Abstract
Publicity is vital to the success of terrorism. Therefore, claiming responsibility for terrorism seems to convey a strategic logic. However, most terrorist attacks are never claimed. This thesis investigates what factors may explain variation in responsibility claiming by the Islamic State, between June 29, 2014, and December 31, 2017. Using data from the Global Terrorism Database, logistic regressions are performed to test the explanatory value of existing theories. Findings suggest that competition, suicide, up to twenty fatalities, and the number of injuries increase the probability of claiming, while kidnapping and targeting civilians suppress responsibility claims by IS.
1. Introduction 1.1 Claiming terrorism
Terrorism is a highly contested and politicised concept (Schmid, 2013, p.15). Due in part to its varying manifestations, its partial overlap with other forms of political violence and the widely varying contexts in which it takes place (Schmid, 2004, p.413), hundreds of definitions exist and are used within academia and governments (Schmid, 2013, p.15). Although no universally accepted definition of terrorism is available, many scholars agree on its communicative nature as one of its defining elements (Davis, 2012, p.9; Min, 2013, p.4; Schmid, 2004, p.407; Weinberg, et al, 2004, p.781). Terrorism is broadly understood as a public act intended to coerce an audience (Wright, 2011, p.2), or even as a method of violent communication (Min, 2013, p.4). According to Schmid, the direct victims of terrorist violence serve as message generators, that, combined with news value and mass media, serve to reach audiences that identify with either the victims or with the terrorist cause (2012, p.159). Publicity is thus deemed vital to terrorists: without it, their message will not reach the target audience, and their actions will not instill fear (Wright, 2011, p.2).
According to Cordes, et al., claiming responsibility for an act of terrorism is the principal method used by terrorists to influence their target audience (1985, p.15). However, most terrorist attacks remain unclaimed by their perpetrators: a mere 83.4 percent between 1996 and 2017 (calculation derived from GTD (START, 2017a)). Moreover, from the 1970s onwards, the number of unclaimed terrorist attacks has risen(Wright, 2011, p.2).This reality does not seem to coincide with the idea that terrorists seek public attention to spread their message and gain support.
Some attempts have been made to explain for variation in responsibility claiming, but an overarching theory remains absent today (Min, 2013, p.2). Several factors may play a role in the decision to claim - or not claim - an attack. Scholars have systematically looked into types of targets (Rorie, 2008; Abrahms, & Conrad, 2017) and target states’ democratic constraints for retaliation (Wright, 2011; Min, 2013), types of tactic (Hoffman, 2010; Rorie, 2008; Abrahms, & Conrad, 2017), magnitude of attacks (Wright, 2011; Min, 2013; see also Rorie, 2008; Abrahms, & Conrad, 2017) intergroup competition (Hoffman, 2010; Wright, 2011; Min, 2013; Abrahms, & Conrad, 2017), and ideology (Hoffman, 2010; Min, 2013; Abrahms, & Conrad, 2017) to explain the existing variation in responsibility claiming.
For example, Hoffman (2010) argues that terrorist groups use public claims of
responsibility as a method to distinguish themselves from other, competing groups that could otherwise be falsely attributed responsibility for particular acts of terrorism. Wright (2011) poses that terrorists are more likely to claim responsibility for attacks against democracies, as these states are legally constrained, and terrorists, therefore, have less to fear from retaliation efforts.
Not only do these studies generally have low external validity due to a limited geographical scope, but their results are inconclusive and in some cases contradictory. For example, Hoffman (2010) and Wright (2011) reach conflicting results with regard to the influence of intergroup competition. Moreover, the most recent data used in analyses of the relationship between these various factors and claim variation originate from 2011, and thus fail to account for the deadliest terrorist organisation of the past years: the Islamic State.
1.2 The Islamic State
The Islamic State (IS) is known under various names, including Islamic State of Iraq and al-Sham (ISIS), Islamic State of Iraq and the Levant (ISIL) and the Arabic acronym DAESH (Semati, & Szpunar, 2018, p.1). This thesis refers to the name IS, as it is the organisation’s current name. Moreover, it only looks into the ‘core organisation’ of IS, while its
predecessors, affiliates and inspired individuals are excluded from the analysis (Miller, 2016, p.1).
IS’ roots can be traced back to the 1990s and to the ideas of its spiritual leader, Abu Musab al-Zarqawi, to “challenge the Western presence in the Middle East” (Tziarras, 2017, p.97). The development of what came to be IS, however, was made possible by specific circumstances in the beginning of the twenty-first century, most notably the war in Iraq, the Arab revolutions, and the Syrian civil war (Oosterveld, & Bloem, 2017, p.5). Exploiting the turmoil that followed the U.S. invasion of Iraq after the 9/11 attacks, al-Qaeda in Iraq (AQI) was established in 2004 when al-Zarqawi’s jihadist group pledged allegiance to al-Qaeda. This organisation grew into the Islamic State of Iraq (ISI) two years later, until it gradually moved away from its original ties with al-Qaeda (Tziarras, 2017, p.98). During these years, the civil and sectarian conflict in Iraq provided a fertile breeding ground for the organisation. The Arab revolutions and developing Syrian civil war allowed ISI to move into Syria and become ISIS or ISIL in 2013 (Oosterveld, & Bloem, 2017, p.7). On June 29, 2014, its new
leader Abu Bakr al-Baghdadi declared a Caliphate in the territory it had gained in Iraq and Syria and renamed the organisation to Islamic State (Tziarras, 2017, p.97-98).
Since its official birth in 2014, IS gained momentum and has perpetrated and inspired thousands of attacks worldwide. Between 2014 and 2017, 3654 attacks in 24 countries were attributed to IS, resulting in over 24,000 deaths and more than 20,000 injuries (calculations derived from GTD (START, 2017a)). When attacks by its predecessors AQI, ISI, and ISIL are included, the numbers are significantly higher. Moreover, IS has managed to create a network of affiliates, inspiring other terrorist groups and individuals across the globe to pledge allegiance and carry out attacks in its name (Miller, 2016, p.7). In addition, large swaths of territory in Syria and Iraq were taken by IS, up to 90,800 square kilometer in 2015 (Oosterveld, & Bloem, 2017, p.19), while tens of thousands of foreigners travelled to Syria and Iraq to fight for the self-proclaimed caliphate (Van der Heide, Winter, & Maher, 2018, p.7).
An international anti-IS coalition that started in August 2014 conducted over 8,000 airstrikes over the next year, while regional allies gradually recaptured key IS strongholds. By December 2017, IS had lost 95 percent of its territory, including two of its most important cities in Iraq and Syria (Glenn, et al., 2019). It would take until March 2019 to recapture all territory of the self-proclaimed caliphate. However, IS’ territorial defeat does not imply eradication of its ideology, and “the terrorist threat it presents is real and enduring” (Van der Heide, Winter, & Maher, 2018, p.12).Despite having suffered severe territorial and financial losses, IS maintained its status as the most active and most deadly terrorist group worldwide in 2017 (Institute for Economics and Peace, 2018). Underground survival of the organisation after the ‘fall of the caliphate’, as well as threats posed by returning foreign fighters, are of major concern to policymakers (National Coordinator for Security and Counterterrorism, 2018, p.62).
Media have given much attention to claims of responsibility issued by IS, portraying the image that it does claim many of the terrorist attacks it perpetrates. In some cases, the terrorist organisation even seems to falsely claim responsibility, for attacks it did not commit (Bearak, 2016). For example, IS’ involvement in the Las Vegas shooting in 2017 is widely disputed (Smith II, 2017), as the organisation failed to provide evidence of any links with the perpetrator (Europol, 2019, p.18). In reality, however, IS only claimed 41.9 percentof attacks attributed to them between its official birth in 2014 and the end of 2017 (calculations derived
from GTD (START, 2017a)). Although this claiming rate is more than twice as high as the average of 84.2 percent over the same period (calculations derived from GTD (START, 2017a)), it still leaves over half of IS claims unaccounted for.
Despite the large amount of media attention IS claims have received, little research has been conducted on IS’ claiming behaviour. One example is a recently published ICCT policy brief, which considers the ways in which IS has claimed attacks in the West between 2016 and 2018. Although the outcomes are valuable for understanding IS’ claiming
behaviour, the study does not look into the factors that may explain the fluctuations in
frequency and claiming method that are found (p.13). In fact, no prior research seems to have systematically looked into the factors that influence IS’ claiming behaviour. This is an odd discrepancy, given the organisation’s impact on worldwide security, and the overwhelming attention it has received from media and policymakers over the past years.
1.3 Societal and academic relevance
Understanding terrorism and its perpetrators are key to countering it. As terrorism is an ever-evolving phenomenon, effective counterterrorism efforts require timely insights (Combs, 2018, p.xix). Although IS has lost its territory, it is not yet defeated (Van der Heijde, Winter, & Maher, 2018, p.11). In fact, its territorial downfall enhances its unpredictability, as there are multiple future pathways for the organisation (National Coordinator for Security and Counterterrorism, 2018, p.56). Therefore, understanding IS strategic decisions over the past years, including those to claim a terrorist attack - or remain silent - is still of importance, as a better understanding of IS may be valuable in countering it in the future. Moreover, providing new insights on a global phenomenon that impacts both public and private security actors and society as a whole, can be a relevant contribution to the governance of security.
Furthermore, at the time of writing, previous research appears not to have looked into the factors that may determine the claiming behaviour of this specific terrorist organisation yet, and no comprehensive theory is available to explain for its variation in responsibility claims. This study is academically relevant, as it will test the explanatory value of existing theories with respect to IS. Moreover, previous literature is partly marked by a limited geographic scope, contradicting results and data up until 2011 - years before IS or even ISIL was officially born, and ISI was only starting to reemerge. After years of IS violence, the theoretical debate on claiming terrorism deserves consideration once again. Despite looking
into a single organisation, the results of this study may inform the larger debate: Perhaps IS, as the most striking terrorist organisation of recent years, could provide new insights into the factors that may explain variation in responsibility claims.
1.4 Research question and hypotheses
This thesis will investigate the question: What factors explain variation in responsibility claims for terrorist attacks by the Islamic State, between June 29, 2014, and December 31, 2017? The research objective here is to test the explanatory value of factors derived from existing theories in relation to the case of IS. Therefore, the following hypotheses are tested: H1: Responsibility claims are more likely when the target state is democratically constrained. H2: Responsibility claims are more likely when attacks take place in a competitive context. H3: Responsibility claims are more likely when attacks signal organisational capability.
H3A: Responsibility claims are more likely when attacks involve suicide tactics. H3B: Responsibility claims are more likely when attacks are more deadly.
H3C: Responsibility claims are more likely when attack methods involve assassination, bombing or hostage-taking.
H4: Responsibility claims are less likely when civilians are targeted.
The literature from which these hypotheses are derived will be explored in detail in the theoretical framework set out in the next chapter. To answer the research question and test the hypotheses, a quantitative analysis will be conducted on data derived from the Global
Terrorism Database (hereafter, GTD). This database provides detailed information on every IS-related attack between 2014 and 2017 and allows for this type of large N-analysis.
1.5 The concept of terrorism
Terrorism remains a highly contested concept (Schmidt, 2012, p.158). As this study looks into responsibility claims for acts of terrorism, the lack of a straightforward, universal definition is problematic. As this issue cannot reasonably be resolved for the purpose of this study, the decision is made to employ the definition that is used in the GTD, which defines the concept of terrorism as follows: “the threatened or actual use of illegal force and violence by a
or intimidation” (START, 2017b, p.10). For measurement purposes, this definition is further operationalized in chapter 3.2.
Admittedly, the use of this definition is crude, as it is broadly defined and does not directly eliminate other forms of political violence. However, it is this definition that underlies the raw data of the GTD, which is the primary data source of this thesis. Therefore, using a different definition would not make logical sense. Moreover, most of the claiming research that forms the theoretical framework of this thesis is based on the GTD. Using the same database and accompanying definition thus likely provides the most accurate test for these theories.
1.6 Reading guide
The remainder of this thesis is organised in four chapters. Chapter 2 provides the theoretical framework for this research, considering the communicative nature of terrorism, the role of responsibility claims, available theories and empirical research on the factors that influence responsibility claims, and the added value of this research. Chapter 3 considers the methods that are used to conduct this research. Chapter 4 discusses the results of the analysis. Chapter 5 involves the conclusion to the research question, as well as a discussion of the limitations and policy implications of this research.
2. Theoretical framework
This chapter provides the theoretical framework on which this study builds. Before turning to the specific theories that will be tested, the communicative nature of terrorism and the
strategic logic of responsibility claims are considered. Then, theories on variation in claiming behaviour are discussed and related to the four hypotheses given in the previous chapter. In addition, consideration is given to the scarce literature on IS’ claiming behaviour. Finally, the added value of this research is discussed.
2.1 The communicative nature of terrorism
Despite the lack of a universal definition of terrorism, its communicative aspects are often highlighted (Davis, 2012, p.9). Martha Crenshaw (1981, p.386) for example, describes
gaining recognition or attention as the most basic reason for terrorism. It has been argued that terrorists use violence to communicate a message to audiences that cannot be reached through mainstream media channels (Rorie, 2008, p.3). While terrorist violence is directed against its primary victims, it is intended to simultaneously influence other audiences, among which adversaries, societies, ‘neutral’ distant publics, potential supporters and sympathisers, rivals, and the media (Schmid, 2004, p.375). The violence, theatricality, suspense, and threat of danger that are implicit to terrorism have the ability to provide the perpetrators with attention and subsequent publicity (Crenshaw, 1981, p.386).
Publicity has proven to be a determinant factor for the success of terrorism, as sufficient attention is a necessity for terrorist organisations to inflict emotional trauma and create a state of terror (Wright, 2011, p.2). Moreover, publicity is an essential element to acquire legitimacy, justification, attention to the cause, support and recruitment (Davis, 2012, p.9). A failure to gain such publicity may lead to the downfall of terrorist organisations (Davis, 2012, p.10).
2.2 The strategic logic of responsibility claims
Given the importance of publicity for the success of terrorism, it is not surprising that terrorist groups throughout history have shown a willingness to publicize their responsibility for acts of terror (Hoffman, 2010, p.616). Moreover, media seem more likely to report on terrorist attacks when the responsible perpetrators are known (Weimann, & Winn, as cited by Rorie, 2008, p.8; and Davis, 2012, p.10). In the past two decades, however, scholars have pointed
out that the assumption of terrorist attacks being claimed is not reflected in the empirical world.
David Rapoport (1997, p.11-12) notes that the focus in terrorism studies has been on claimed terrorist attacks, while the fact that most attacks have remained unclaimed has been largely ignored. This has created the misconception that responsibility for terrorist attacks is usually claimed by the perpetrating organisation afterwards. In reality, anonymous attacks are the norm (Abrahms, & Conrad, 2017, p.279). Moreover, the number of unclaimed attacks has increased in the past half-century (Davis, 2012, p.5). This discrepancy between conventional theory and the empirical world presented terrorism scholars with a puzzling question, as it meant a need for rethinking their understanding of terrorism (Rapoport, 1997, p.12).
To search for explanations for variation in responsibility claiming by terrorist organisations means to consider the nature of terrorism itself (Rapoport, 1997, p.12). Although terrorism remains a contested concept, the dominant scholarly view regards terrorists as rational, strategic actors (Abrahms, & Conrad, 2017, p.279). Consequently, the conviction that terrorism is based on a strategic logic forms the starting point for many theories considering responsibility claims by terrorist organisations.
The strategic logic is easily recognized in Melissa Rorie’s application of a rational choice perspective on responsibility claiming (2008). She argues that terrorists deliberately calculate the costs and benefits before deciding whether to claim responsibility for an attack (Rorie, 2008, p.65). Potential benefits could be that terrorist organisations become feared after a claim of responsibility. The threat they pose earns them publicity, giving them an
opportunity to justify their actions, to gain support and sympathy from an audience, and to gain attention to their causes (Rorie, 2008, p.8-9). Potential costs could be that the publicity after a responsibility claim may backfire: Instead of providing legitimization and attention to the grievances and causes of the terrorist organisation, media may just report on the tactics and aftermath of the attack, fueling moral outrage. The message to the target audience may not come across as intended by the terrorist organisation, possibly resulting in a loss of support. Moreover, policymakers generally focus on punishing the perpetrators, and a responsibility claim inevitably leaves terrorist organisations exposed and vulnerable to counterterrorism measures, having given up their anonymity (Rorie, 2008, p.10-11).
Similarly, Max Abrahms and Justin Conrad (2017, p.281) stress the strategic logic of claiming credit for acts of terrorism. In their view, the leadership of a terrorist organisation
will only claim responsibility if the expected political return is positive. When the expected consequences of a perpetrated attack are negative, the leadership will be reluctant to issue a responsibility claim (Abrahms, & Conrad, 2017, p.281). Other authors take on a strategic perspective to explaining variation in responsibility claiming as well (i.e. Hoffman, 2010; Wright, 2011; Min, 2013). Thus, in theory, responsibility claims are only expected when they make strategic sense.
2.3 Variation in responsibility claiming
As shown above, both the decision to claim responsibility as well as the decision to remain anonymous can be argued for from a strategic perspective. So why are some attacks claimed, and not others? Several attempts have been made to uncover what factors may account for the existing variation in responsibility claiming.
Austin Wright (2011) looks at the political properties of states to explain for variation in responsibility claiming. He poses that attacks targeting democratic states are claimed more often, since those states are constrained by strong legal systems and are thus stable and predictable in terms of counterterrorism efforts. On the other hand, autocratic states can be seen as unpredictable and unstable, as they are unconstrained in their use of retaliatory violence against terrorists (Wright, 2011, p.10). In addition, Eric Min (2013, p.13) poses that democracies constitute an opportunity for terrorists to propagate information and fear, as they usually feature free and open press, that readily reports claims. Both Wright (2011, p. 20) and Min (2013, p. 30) find evidence in their research that democratic constraints incentivize responsibility claiming. Based on this literature, the following hypothesis will be tested: H1: Responsibility claims are more likely when the target state is a democracy.
Aaron Hoffman (2010) focuses on the competitive context of terrorist organisations to explain for variation in responsibility claiming. To be able to generate support, terrorist organisations want their target audiences to know about their identity and activities (Hoffman, 2010, p.616). However, in an environment where multiple terrorist organisations are active, potential supporters may be confused about who is responsible for an attack. Then, publicly claiming responsibility shows to be the easiest and cheapest strategy for terrorist organisations to advertise their accomplishments and distinguish themselves from others, according to Hoffman (2010, p.616-617). In his research, Hoffman finds evidence for the positive effect of
competition on responsibility claiming (2010, p.616). However, Hoffman’s study is limited to the Israeli context, and while Min’s research confirms his theory (2013, p.39), Abrahms and Conrad find competition to be an unreliable predictor for responsibility claiming (2017, p.293), while Wright finds no empirical evidence at all (2011, p.21). Based on Hoffman’s theory, the following hypothesis will be tested:
H2: Responsibility claims are more likely when attacks take place in a competitive context. Wright (2011) additionally looks at the type of attack to explain for variation in responsibility claiming. He states that “credit-taking is primarily a consequence of terrorists’ attempts to signal strength by claiming ownership of particularly dramatic acts of terrorism” (Wright, 2011, p.4). In his analysis, he finds the signalling strength of a particular attack, captured in its magnitude in terms of casualties, and use of dramatic tactics such as suicide terrorism, to be a significant predictor for responsibility claiming (Wright, 2011, p.17). Min explains that suicide attacks signal strength, as the organisation is apparently capable of recruiting people that are willing to die for their cause. Moreover, acts of martyrdom are considered highly honorable in Islamic contexts (Min, 2013, p.10). The suicide thesis is supported by previous findings of Hoffman (2010, p.622), and later by Min (2013, p.39).
On the note of signalling strength, Wright considers the magnitude of the attack, in terms of lethality, to be a factor that contributes to the signalling strength of an attack (2011, p.5). The ability to inflict many casualties or employ particularly gruesome tactics sends a message to governments and other audiences about the power a particular group possesses. The level of damage is then interpreted as a proxy for the group’s resources (Overgaard, as cited by Wright, 2011, p.5). Similar to Wright, Min finds empirical evidence for the number of fatalities to improve the chances of responsibility claiming (2013, p.39). However, Rorie (2008, p.48) and Abrahms and Conrad (2017, p.292) find quite meager support for the effect of casualties.
Abrahms and Conrad follow Wright’s thesis on signalling strength and purport that “certain methods of attack require, and thus signal, more capability than others” (2017, p.289). They specifically mention armed assault, assassination, and hostage-taking in this regard, arguing that these methods are inherently more sophisticated than unarmed assault or other attack methods in terms of planning, training, and logistics (2017, p.289). In their analysis, however, they do not find evidence that assassination or armed assault enhances the
probability of a responsibility claim being issued. Still, some empirical support is found for the positive effect of hostage-taking on claim-making (Abrahms, & Conrad, 2017, p.293). Min (2013, p.29) considers assassination, as well as bombing methods to affect claim-making due to their conspicuous and costly nature. His results point to a negative relationship,
however (Min, 2013, p.27). Based on the idea of signalling strength and its different manifestations, the following hypotheses will be tested:
H3: Responsibility claims are more likely when attacks signal organisational capability. H3A: Responsibility claims are more likely when attacks involve suicide tactics. H3B: Responsibility claims are more likely when attacks are more deadly.
H3C: Responsibility claims are more likely when attack methods involve assassination, bombing or hostage-taking.
Some scholars concentrate on the type of target as a predictor of responsibility
claiming. Rorie (2008, p.52) distinguishes between attacks targeting specific targets and those targeting the general public. She argues that attacks on the general public often involve a higher number of casualties and increased counterterrorism efforts, which raise the costs of claiming responsibility. An alternative explanation would be that terrorists may seek to maximize the element of unpredictability to maximize fear, by choosing random targets and additionally, remain anonymous (Rorie, 2008, p.52). However, the study conducted by Rorie is limited to terrorist attacks in the United States, and no significant results are found with regard to the target type (Rorie, 2008, p.48).
Abrahms and Conrad (2017, p.282) find that when terrorist organisations target
civilians, their leaders are less likely to claim the attack. They explain this by proposing a new theory, borrowing from a principal-agent framework. Instead of assuming terrorist
organisation work as unitary actors, Abrahms and Conrad distinguish between leaders and subordinates, who may not own the same commitment to the official goals of the
organisation, or the knowledge and skills to achieve them. The subordinates may sometimes act in defiance of their leaders in terms of targeting choices. This presents the leaders with an agency problem, as indiscriminate violence against civilians can be counterproductive in terms of strategic utility. Hence, leaders are inclined to remain anonymous when civilians are targeted, as found by Abrahms and Conrad in their study (2017, p.284-285). Based on the latter theory on target type, the following hypothesis will be tested:
H4: Responsibility claims are less likely when civilians are targeted.
Several other factors have been taken into consideration in the handful of studies conducted on responsibility claiming, such as tactical success of the attack (Rorie, 2008), military responses, state sponsorship (Hoffman, 2010), national capability (Wright, 2011), proximity to elections (Min, 2013), state repression, organisation size (Abrahms, & Conrad, 2017, p.292) and ideology (Hoffman, 2010; Min, 2013; Abrahms, & Conrad, 2017).
However, these factors generally did not yield significant results or provide comprehensive explanations.
In an attempt to overcome geographical limitations of previous studies, Min (2013) conducted a large-N quantitative analysis to test existing theories. His study provides
evidence that supports the theories on signalling strength, competitive context and democratic constraints as predictors of responsibility claiming (Min, 2013, p.1). Most notable, however, are Min’s findings that the overall ideology and goals of terrorist organisations account for a ‘baseline of claim-making’: groups with more limited and specific ends tend to claim their attacks more often than groups with broader, revolutionary or jihadi aims (Min, 2013, p.39).
2.4 The Islamic State’s claiming behaviour
Applying Min’s findings on the baseline of claim-making on IS is not straightforward.
Interestingly, IS’ aims are ambiguous and not easily categorized. IS’ ideology is both political and religious, and its goals are specific in certain countries, but broad and unstructured
globally (Snyder, 2016, p.2). Assuming a baseline of claim-making based on IS’ aims would, therefore, be questionable. In fact, at the time of writing, barely any academic research seems to have been conducted on IS’ claiming behaviour.
Just recently, however, a policy brief written by an anonymous Europol operative was published by the International Centre for Counter-Terrorism, which looks into IS
responsibility claims in the West between 2016 and 2018. The report mainly focuses on the relationship between external operations in the West and IS’ main leadership (Europol, 2019, p.1), highlighting internal and external communication channels. In addition, the report points out the global competition with al-Qaeda as an incentive to claim as many attacks as possible and to maintain credibility through a reliable communication system. As IS adopted a strategy to incite attacks by lone individuals in the West in 2016, gaining timely and accurate
information about attacks proved to be problematic, especially as the remaining media
apparatus came under increasing pressure by anti-IS offensives in Iraq and Syria over the next years (Europol, 2019, p.3-15).
Most importantly, the report finds considerable variation over time in the type of attacks, their circumstances, and strategic decisions that resulted in responsibility claims. However, the report does not look into the factors that may explain the temporal fluctuations in frequency and claiming method that are found (Europol, 2019, p.13). Moreover, due to its focus on attacks in the West between 2016 and 2018, the large majority of attacks perpetrated (and claimed) by IS are overlooked. Although the findings of the report are valuable for understanding the dynamics of IS’ communication channels, the factors that may explain variation in responsibility claiming by IS, are yet to be examined.
2.5 Added value of this research
The review of theoretical and empirical research into responsibility claiming above shows that no comprehensive explanation is available for the variation that exists in empirical reality. Some factors that may partially account for responsibility claiming are identified by a handful of studies, but the results are mostly inconclusive. Therefore, the subject remains of scientific interest. Moreover, the most recent data used in previous research originates from 2011, when terrorist organisation IS just started to grow stronger (Tziarras, 2017, p.98). Undoubtedly, a gap in knowledge exists on IS’ claims of responsibility.
This study will add to the literature by examining claims of responsibility by IS, testing the value of existing theories. In particular, theories on democratic constraints, competitive context, signalling strength, and target type are examined, as these theories have gained the most attention in previous research, and are easily applicable to the case of IS. Some of the factors mentioned by previous scholars are not included in this thesis, due to questions of feasibility in terms of operationalization (i.e. state sponsorship or organisation size) or because they are expected to remain constant with regard to IS (i.e. ideology).
Moreover, although numerous other factors - both in general and specific to IS - may play part in the decision to claim an attack or remain anonymous, identifying an exhaustive list of factors lies beyond the scope of this thesis.
After this overview of the communicative nature of terrorism, the strategic logic of claiming responsibility, variation in responsibility claiming, IS claiming behaviour, and the
added value of this research, the theoretical framework for this thesis is set out. In the following chapter, its methodology will be outlined.
3. Methodology
To investigate the factors that explain for variation in responsibility claims by IS, this study takes on a quantitative approach. In this chapter, the research design, case selection,
conceptualization and operationalization of variables, methods of data selection and methods of data analysis are described. To conclude, potential validity issues of this thesis are
discussed.
3.1 Research design
This thesis constitutes a quantitative study, for which an observational, large N-design is used. This type of design is suitable for the following reasons. The goal of this research is
explanatory in nature. It aims to test theories on the relationship between different independent variables, (namely democratic constraints, competitive context, signalling strength, and type of attack) and a dependent outcome; the absence or presence of a
responsibility claim for a terrorist attack. Since conducting a controlled experiment would in this case neither be feasible nor ethically desirable, causality cannot be established. Using a large-N design however, causality can be inferred instead by studying correlations in the empirical world (Toshkov, 2016, p.160-161). This constitutes the main advantage of large-N designs over comparative or single-case studies, as only the former are suited to identify and estimate causal relationships through quantitative, statistical analysis (Toshkov, 2016, p.200).
3.2 Case selection
This thesis examines the absence or presence of responsibility claims by IS after terrorist attacks. The basic unit of analysis for this study is confined to terrorist attacks attributed to IS. Figure 1 provides a schematic overview of the selection process. Further, Appendix A
provides the Syntax commands that were used to select the final sample.
Due to limited time and resources, it would not be feasible to single-handedly collect all the relevant details of the thousands of terrorist attacks perpetrated by IS. Regardless, conducting fieldwork to collect primary data among terrorists could be dangerous (Schuurman, & Eijkman, 2013, p.3). Because of these practical concerns, the target
population of this study is restricted to events that are included in the GTD (START, 2017a). The latest version of this database ranges from 1970 to 2017. For the purpose of this research,
only a subset of the GTD that ranges from 2014 to 2017 is needed. This data file provides a sample of 56.355 cases.
As was explained in chapter 1.5 on the definition of terrorism, in this study, terrorism is defined in accordance with the GTD definition, which is also implicated by the use of this database. For practical use, this definition is operationalized for the GTD as follows: An event is classified as a terrorist attack when it meets three criteria. First, the incident must be
intentional. Second, the incident must entail some level of violence or the threat of violence. Third, there must be sub-national perpetrators, meaning that they are not exercising authority. In addition to these criteria, at least two of the following elements must be present in each case in order to be classified as a terrorist attack: (a) the act is aimed at attaining a political, economic, religious, or social goal, (b) there is evidence of an intention to coerce, intimidate, or convey some other message to a larger audience (or audiences) than the immediate victims, (c) the action takes place outside the context of legitimate warfare activities (LaFree, & Dugan, 2007, p.188).
In some cases, determining if an event constitutes a terrorist attack is problematic, because of a definitional overlap with other forms of crime or political violence, or because sources provide insufficient or conflicting information (START, 2017b). The GTD includes a variable that indicates whether an alternative designation of the event is possible, namely insurgency or guerilla action, a different crime type, intra- or intergroup conflict, a lack of intentionality, or the involvement of a state actor. For the purpose of this research, events that do not exclusively qualify as terrorism, are excluded. This brings the sample size down to 45.560 cases.
As the interest of this thesis lies with the terrorist organisation IS, the time frame of the case selection is adjusted to its lifespan and the available data. Thus, as the name ‘Islamic State’ was called into existence on June 29, 2014, the time frame ranges from this date to December 31, 2017, which is the most recent date for which data from the GTD is available (START, 2017a). This time frame restricts the number of cases to 38.514 cases.
This study focuses on all terrorist attacks that are attributed to IS, irrespective of the presence or absence of a responsibility claim. Attribution of an attack entails that the
perpetrator group has been identified in the GTD (START, 2017b). As mentioned in chapter 1.3 on the Islamic State, this thesis only investigates IS’ core organisation. Resultantly, attacks that are attributed to IS predecessors, affiliate organisations and inspired individuals
are disregarded. Although IS rebranded itself into its current name on June 29, 2014, the GTD continues to employ the name ISIL beyond this date. Because ISIL is technically a
predecessor organisation of IS, the use of this name could lead to confusion. Thus, for the purpose of this thesis, attacks that are attributed to ISIL from June 29, 2014, and beyond are considered to be attacks attributed to IS.By selecting only the cases that are attributed to IS, the sample size is reduced to 3654 cases.
Figure 1. Flow scheme of case selection.
Finally, since the presence or absence of a responsibility claim is the dependent outcome that is of interest here, the case selection is further limited by excluding cases for which data on responsibility claiming is missing. However, none of the remaining cases lack this information. Thus, the final sample size remains large (N=3654).
3.3 Operationalization
This thesis investigates what factors can explain variation in terrorist attack responsibility claims by IS. Based on the claiming literature that was outlined in the previous chapter, the focus of the investigation will be on the following factors: democratic constraints, competitive
context, signalling strength, and type of target. To be able to measure responsibility claims and the different factors that may explain them, these concepts must be translated into observable indicators (Toshkov, 2016, p.83). Below, the main concepts of this research are operationalized into measurable variables. Table 1 provides an overview of the concepts, related variables and their values of measurement.
3.3.1 Dependent variable
Claim of responsibility
Evidently, the claim of responsibility is the primary concept of interest of this thesis, and serves as the dependent variable. A claim of responsibility is its essence an ‘announcement of culpability’ for an attack, issued by someone allegedly representing the perpetrator(s)
(Hoffman, 2010, p.620). In this study, a terrorist attack is understood to be claimed when a person or group has claimed responsibility for the attack, through a letter, call (post- or pre-incident), e-mail, note left at the scene, video, post to website, blog or social media, personal claim, or by other (unknown) means (START, 2017b). The absence or presence of such a claim of responsibility is measured with a binary variable (0/1), retrieved from the GTD. As the original coding of the variable in the GTD is suitable for the purpose of this research, it needs no alterations.
3.3.2 Explanatory variables
Democratic constraints
In their research, both Wright (2011, p.22) and Min (2013, p.29) have found that
responsibility claims are issued more often when the target state of an attack is a democracy. Following these authors’ examples, democratic constraints are measured using the Polity IV score for each country in each year (Center for Systemic Peace, 2018a). The Polity IV Project makes a distinction between 21 regime types, ranging from -10 (hereditary monarchy) to +10 (consolidated democracy) (Center for Systemic Peace, 2018a). Hence, democratic constraints are considered to be indicated by a democratic regime type. A polity score of 6 or higher is considered to signify a democratic regime type. As polity scores are not available in the original GTD dataset, this information is derived from the Polity IV dataset (Center for Systemic Peace, 2018b) and combined with existing GTD data on year and target state, to create a Polity IV variable.
For the purpose of this thesis, a variable indicating the Polity IV score for each country in each year is added to the dataset. A score of 6 or higher serves as an indicator for democratic constraints, while a score of 5 or lower indicates the absence of democratic constraints. Table 1 provides an overview of the meaning of values on the Polity IV variable. To make a clear distinction between the presence and absence of democratic constraints, an additional dummy variable (0/1) based on the abovementioned Polity IV variable is created, indicating whether a country in a given year was identified as a democracy.
Competitive context
Competition between terrorist groups is identified as an incentive for claim-making by Hoffman (2010, p.624), and Min (2013, p.29), while Wright (2011, p.21), and Abrahms and Conrad (2017, p.293) do not find similar results. While previous versions of the GTD included a variable for intergroup competition, the recorded data proved to be inconsistent, and the variable was removed from the dataset since 2012 (Min, 2013, p.21). Some authors have hence resorted to using a dichotomous proxy variable to indicate whether more than one distinct terrorist group is attributed an attack within a country, within each year. However, in the sample used here, measuring competition with such a variable would be meaningless, as competitive context would be a constant, rather than a variable (as calculated from GTD data).
Thus, for the purpose of this research, competitive context is measured by the number of competing terrorist organisations within each country, within each year. Importantly, IS itself, as well as groups that are affiliated to IS, are not considered as competing terrorist organisations. A list of all IS affiliate organisations, derived from the GTD ISIL-auxiliary dataset (Miller, & Kane, 2016), can be found in Appendix B. To be able to exclude IS affiliate organisations from the equation, a dummy variable is added to indicate whether an
organisation that is attributed an attack, is an affiliate to IS. It should be noted here that to create this variable, information is needed about attacks that are not attributed to IS. Thus, in order to obtain this information, the final exclusion criterion of the sample selection (see Table 1) is lifted temporarily, resulting in a subset of 38.514 cases. From this subset, the number of groups that are attributed at least one terrorist attack is calculated for each country and each year, excluding IS and IS affiliates. This data is then used to create a new variable that combines this information, indicating the number of competitors per country and year.
Table 1
Operationalization scheme
Concept Variable Values Value label
Responsibility claim Claimed 0/1 0 = no, 1 = yes Democratic
constraints
Polity IV -10 to 10 -10 to -6 = autocracy -5 to 5 = anocracy 6 to 10 = democracy
Democracy* 0/1 0 = no, 1 = yes
Competitive context Competitors Continuous Number of competitors per country/year
Signalling strength Suicide 0/1 0 = no, 1 = yes Fatalities Continuous Number of fatalities
Fatal.cat.a* 0 to 9 Categorical variable for the number of fatalities
Injuries* Continuous Number of injuries Barricading 0/1 0 = no, 1 = yes
Kidnapping 0/1 0 = no, 1 = yes
Assassination Armed Assault 0/1 0/1 0 = no, 1 = yes 0 = no, 1 = yes
Bombing 0/1 0 = no, 1 = yes
Type of target Civilian 0/1 0 = no, 1 = yes
Note. a Fatal. cat. includes the number of fatalities, divided into 10 categories. Each category contains
at least two observations. *Control variables.
Although competition may exist even when no other distinct terrorist groups are attributed attacks in the GTD in any given country and year, Min (2013, p.21) argues that the very
proposition of claiming responsibility to compete with rivals would assume terrorists to seek public attention for their attacks, and resultantly end up in the GTD.
Signalling strength
As Wright (2011) and Min (2013, p.39) find in their research, terrorists seem inclined to claim responsibility for attacks that have the power to ‘signal their strength’ in terms of
organisational capability. Combining the literature, this signalling strength is operationalized into three separate indicators; the use of suicide tactics, the magnitude of the attack, and the use of sophisticated attack methods.
First, the use of suicide tactics is believed to signal strength, as it involves high costs in terms of human capital, and demonstrates the ability to recruit people willing to die for the terrorists’ cause (Min, 2013, p.10). The GTD contains a binary variable (0/1) that indicates whether there is evidence that the perpetrator of the attack “did not intend to escape from the attack alive” (START, 2017b, p.27). This variable is used here without alterations to measure the presence or absence of a suicide attack.
Second, the magnitude of the attack in terms of lethality is thought to signal an
organisation’s strength. A higher number of casualties is found to predict responsibility claims (Wright, 2011, p.18; Min, 2013, p.24). To measure the magnitude of the attack, a continuous variable derived from the GTD is used to indicate the total number of fatalities. In addition, the total number of non-fatal injuries is used as a control variable, as injuries may serve to indicate the magnitude of the attack as well. Both of these continuous variables can be used without alterations for the purpose of this research.
However, an initial observation of the distribution of fatalities in the data sample shows a large number of outliers with hundreds of casualties. As deletion of these outliers would result in significant data loss, an additional variable is created (fatal. cat.). This control variable presents the number of fatalities, but divided into 17 categories. The value of zero represents zero casualties, whereas the next ten categories encompass the ten consecutive numbers of fatalities. Values 11 through 14 contain the subsequent fifty numbers of fatalities. Value 15 includes casualties from 301 to 400, while the last category includes each case with over 400 deaths. Including this variable allows the analysis of all frequencies of fatalities, without a distortion caused by the high number of outliers.
Third, the method of attack could be of importance. As Abrahms and Conrad (2017, p. 289) pose, some attack methods require more organisational capability and sophistication than others. The GTD distinguishes between nine broad categories of attack types, notably
assassination, armed assault, bombings or explosions, hijacking, hostage-taking through barricading or kidnapping, facility or infrastructure attacks, unarmed assault and unknown methods of attack. Of these categories, assassination, armed assault, and hostage-taking are considered to be more sophisticated than other methods, and thus to indicate a higher degree of signalling strength (Abrahms, & Conrad, 2017, p.289). In similar fashion, Min (2013, p.21) considered the costliness and conspicuous nature of bombings and assassinations.
To measure the value of these ‘sophisticated’ methods, the original GTD variable for attack type is recoded into binary variables (0/1) that indicate whether the attack was an assassination, bombing, hostage-taking or armed assault. Since Abrahms and Conrad only find support for the effect of hostage-taking (2017, p.293), this type of method is taken as an explanatory variable, while armed assault is added only as a control variable. Likewise, Min finds significant (albeit negative) results for bombing and assassination (2013, p.27), which justifies adding these types of attack methods as explanatory variables.
As the GTD distinguishes between two subtypes of hostage-taking (barricade incidents and kidnappings), both subtypes are separately recoded into binary variables to indicate hostage-taking. Cases where the attack method was coded in the GTD as ‘unknown’, are defined as missing for the purpose of this thesis; otherwise, they could lead to an
overrepresentation of cases coded as being not a hostage-taking, assassination, bombing or armed assault.
Type of target
Lastly, the primary target of an attack could play a role in the decision to claim responsibility for an attack. Claims seem to be issued less frequently when a civilian, general public is targeted, as opposed to military, government or other specific targets (Rorie, 2008, p.52; Abrahms, & Conrad, 2017, p.291). For the purpose of this research, the type of target is measured with a dummy variable (0/1), indicating whether the primary target is civilian or not. This dummy variable is created by recoding an existing GTD variable that categorizes 22 general target types. A target coded as civilian includes the categories private citizens,
telecommunications, transportation and utilities as defined in the GTD (START, 2017b, p.32-36). This coding is similar to the coding used by Abrahms and Conrad (2017, p.289), and thus allows the most exact test of their theory in this research. As the original GTD variable on target type contains a category of ‘unknown’ target types, this category is defined as missing. Otherwise, these cases would simply be defined as ‘not civilian’, which could potentially influence the analysis.
3.4 Methods of data collection
The data selected for this research are directly derived from the GTD (START, 2017a). This database collects its data from publicly available, open-source materials, using automated and manual collection strategies. Published media articles are filtered automatically until a subset of articles is left to be reviewed and coded in accordance with a codebook by GTD
researchers (START, 2017c).
The GTD is the most suitable data source for the purpose of this research for a number of reasons. First, it provides information on various attributes of worldwide terrorist attacks. Second, the data it contains is suitable for quantitative research analysis, as is required by the research design of this study. Third, the GTD is more suitable than other terrorism databases because it contains information on claims of responsibility, is transparent in terms of coding and use of sources, and has a broad definitional scope to overcome the lack of a universal definition of terrorism. Moreover, the database is fully accessible online at no charge (Sheehan, 2012, p.29-35) and contains data information up to 2017 (START, 2017a).
The GTD contains most data that is needed to test the hypotheses stated in the first chapter of this thesis. Some variables require recoding to suit them to the purpose of this research (i.e. target type is recoded into a binary variable for civilian/non-civilian), and other concepts of this thesis require variables to be combined to create new variables (i.e. group name is combined with the date of the event to create a variable for IS). Only one concept requires an additional data source; the data on democratic constraints are derived from the Polity IV Project. This database provides information on regime authority that allows for distinction between democracies and non-democracies, and is easily compatible with the GTD data.
3.5 Methods of data analysis
To analyze the data obtained from the GTD and test the hypotheses, statistical analysis will be employed. For the purpose of this thesis, logistic regression analysis is the most suitable analysis method, as it fits the research design for multiple reasons. In this type of analysis, the dependent variable is non-metric and binary, while the dependent variables can be metric or non-metric (Hair, et al., 2014, p.314). This corresponds with the binary coding of the
dependent variable in this study (claim of responsibility), as well as the independent variables, of which some are continuous (i.e. number of fatalities), and others are binary (i.e. suicide). Many tests used in logistic regression are based on large samples, with a minimum sample size of 300 cases. As the sample used in this study consists of 3654 cases, this boundary is well exceeded. Likewise, the minimum recommended sample size of 10 per category of the dependent variable is easily met. Moreover, logistic regression analysis can serve both to predict and explain the outcomes of the dependent variable. As the research objective of this thesis is to explain variation in responsibility claiming, logistic regression analysis serves this purpose (Hair, et al., 2014, p.314).
3.5.1 Logistic regression
Logistic regression is especially designed to predict or explain the probability of the
occurrence of an event (in this case, the issuing of a responsibility claim) (Hair, et al., 2014, p.317). This type of analysis estimates the probability of a certain outcome of the dependent variable, given a set of independent variables. However, the binary nature of the dependent variable (0 or 1) means that assumptions of normality and homoscedasticity do not apply. Moreover, only predicted values (probabilities) that fall between 0 and 1 can be meaningful for interpretation (Sieben, & Linssen, 2009, p.1). In logistic regression, the dependent variable is transformed into a logit model, to ensure that predicted values always fall between 0 and 1.
In this model, the natural logarithm (ln) of the odds
(
1−pp)
of the dichotomous dependent variable is estimated, using an intercept (a) and parameters (b) for each of the independent variables (X) (Sieben, & Linssen, 2009, p.2). The logit model is expressed in a formula as:n
In this study, the specific logit model estimates the probability of an attack being claimed, using the given information on democratic constraints, competitive context, signalling strength of the attack, and type of target. The final model can be expressed as:
n
l
(
p not claimedp claimed)
=P olityIV Democracy Competitors Suicide F atalities b F atal.cat.
a + b1 + b2 + b3 + b4 + b5 + 6
Injuries Assassination Bombing Barricade incident
+ b7 + b8 + b9 + b10 + b Kidnapping 11
Civilian + b Armed assault 12 + b13
One of the biggest advantages of logistic regression is the general lack of required assumptions (Hair, et al., 2014, p.319). In general, scholars do not check these assumptions, partly because they are complicated by the logistical function, and partly because a breach of assumptions automatically translates into a bad model fit. Moreover, some exceptions to the assumptions are possible (Bijleveld, & Commandeur, 2012, p.180-183). Thus, for the purpose of this research, the assumptions are presumed to be met, and further checks are disregarded.
3.5.2 Assessing the model fit
Before interpreting the results of logistic regression, it must be estimated how well the predicted model fits the actual data. The basic measure to estimate the overall model fit is the likelihood ratio test (Hair, et al., 2014, p.323). This is done by taking minus two times the log of the likelihood value (-2LL) of the occurrence of the event (claim). The model fit is perfect when -2LL equals to zero, and thus a value closer to zero corresponds with a better model fit. To detect improvement of the overall fit between models, the significance of the difference between -2LL of the models can be examined using Chi-square testing (Hair, et al., 2014, p.323).
In addition to the overall model fit, it may be useful to assess the amount of variation on the dependent variable that can be explained by the predicted model. Unlike linear regression, logistic regression does not provide an absolute measure for explained variance (Bijleveld, & Commandeur, 2012, p.173). Instead, the model fit can be estimated in two ways; using ‘pseudo’ R2 values, or examining the predictive accuracy of the model (Hair, et
al., 2014, p.323). For the purpose of this thesis, the Nagelkerke R2, will be used to compare fit size between models; this pseudo R2 is easily interpreted because of its 0-1 range.
For examination of the predictive accuracy of the model, two tests are available. The classification matrix measures how well the group membership on the dependent variable is predicted by the model, providing a percentage of correctly classified cases (Hair, et al., 2014, p.324). The Hosmer-Lemeshow 𝝌2-test divides the sample into classes to measure the
predictive accuracy of the model. The model fits well if the difference between the observed and predicted distribution is insignificant (Bijleveld, & Commandeur, 2012, p.178).
Although it is preferable to investigate as many measures of model fit as possible (Hair, et al., 2014, p.325), not every test is equally reliable in each circumstance. The Hosmer and Lemeshow 𝝌2-test, for example, is sensitive to small or very large sample sizes, where it may incorrectly report significant differences (Sieben, & Linssen, 2009, p.7). However, the likelihood ratio test is always correct (Bijleveld, & Commandeur, 2012, p.179). Therefore, this test will be used as a threshold for the estimation of the model fit.
3.5.3 Interpretation of the coefficients
When the model fit is assessed, the coefficients provided by the logit model can be
interpreted. Before evaluating these individual coefficients, their significance can be assessed through the Wald statistic. In this way, each hypothesis can be tested individually. Only when the Wald test shows statistical significance can the coefficient be interpreted in terms of directionality and magnitude (Hair, et al., 2014, p. 325).
The model coefficients (B) indicate the change in the dependent variable for each independent variable. However, as explained, the dependent variable is expressed in logit odds, which are difficult to interpret (Bijleveld, & Commandeur, 2012, p.168). Instead, the exponentiated logit coefficient (exp(B), which equals to the odds ratio) is used for
interpretation. Still, the original coefficient unambiguously indicates the direction of the relationship between the independent and dependent variable: a positive relationship simply translates into a positive coefficient, whilst a negative relationship is expressed by a negative value (Hair, et al., 2014, p.326).
Assessing directionality is less straightforward when the exponentiated coefficient is used. In this case, a value above one indicates a positive relationship, whereas a value between zero and one indicates a negative relationship. As the exponentiated coefficient
actually reflects the change in the odds value of the dependent variable, it is useful for interpretation of the magnitude of the relationship. The exponentiated coefficient is multiplicative, which means that for each one-unit change in the independent variable, the odds value of the dependent variable is multiplied by the exponentiated coefficient. Indeed, when no relationship exists (exp(B) = 1.0), the odds remain equal (multiplication by 1.0 equals to no change) (Hair, et al., 2014, p.327). When multiple variables are used in one model, the adjusted odds ratio presents the odds for each independent variable, relative to the other independent variables in the model (Bijleveld, & Commandeur, 2012, p.171). Moreover, the percentage change in odds can easily be calculated by extracting 1 from the exponentiated coefficient, and multiplying the result with 100 (Hair, et al., 2014, p.327).
3.6 Assessment of validity issues
Some potential validity concerns are worth noting. First of all, using the GTD comes with a possible bias that may threaten the internal validity of this study. The GTD relies solely on open sources, so the attribution of attacks to IS may be biased or false in some cases. This issue is addressed by GTD coders through assessment of the validity of sources (START, 2017b, p.8). Furthermore, using the GTD definition of terrorism comes with its own difficulties, as no universally accepted definition of terrorism exists (Schmidt, 2013, p.15). Although the GTD purposely uses a broad definition to overcome the definition issue, this implies that the events that are included in the database, are not universally regarded as (exclusively) terrorism, as different definitions are still used in practice. As mentioned in chapter 3.2 on case selection, the GTD contains a variable that indicates whether an event qualifies exclusively as terrorism, which can be used as an additional filtering mechanism (START, 2017b, p.11). However, defining terrorism remains contested and the labeling of events as such should always be considered with caution.
Second, when terrorists remain anonymous after an attack, it may never be attributed to the organisation. Moreover, media are more likely to report on attacks for which
responsibility has been established (Weimann, & Winn, as cited by Rorie, 2008, p.8; and Davis, 2012, p.10).Hence, there may be an overrepresentation of claimed attacks in the GTD data. This will need consideration when interpreting the results of this study. Conversely, an under-reporting bias of claims may also exist, because more autocratic regimes may have ill-equipped or state-controlled media that neglect to report on claims (Min, 2013, p.19). Min
counters this argument by posing that terrorists have an interest in making sure they gain public attention, and that media often proactively look for the organisation responsible, after an attack has taken place (2013, p.20).
Third, terrorists frequently lie about their involvement in attacks. According to Kearns, et al., false claims of responsibility for terrorism are relatively common, and it is difficult to ascertain the truthfulness of responsibility claims in general (2014, p.426). While the focus of this thesis is not on whether IS is truthful in its claiming behaviour, the selection of events attributed to IS implies that claims of responsibility are truthful, if the attribution has been accurate. Moreover, the selected sample implicitly excludes attacks that are claimed by, but not attributed to IS, and thus are considered to be falsely claimed by IS. This may be the case when despite an IS claim of responsibility, an attack is attributed to another perpetrator, or when a claim is simply deemed unconvincing due to suspiciousness or lack of confirmation. Thus, the sample may be biased towards ‘true’ claims. However, the truthfulness of
responsibility claims by terrorists should always be treated with a degree of scepticism (Kearns, et al., 2014, p.436).
Fourth, when interpreting the results of this study, the distribution of cases across countries should be taken into consideration. Although the sample contains IS attributed attacks in 24 different countries, two of them dominate the distribution of cases: Iraq takes up a mere 85.0 percent of the data, while Syria accounts for 10.5 percent. Although this may largely influence the results, these countries’ large share of IS attributed attacks reflects the reality of attack distribution on the ground. Iraq and Syria have been the primary operation zone and part of the IS’ ‘caliphate’, and thus provide the best information on IS claiming behaviour. Nonetheless, awareness of this unequal distribution of data is necessary when interpreting the results of this thesis.
Finally, since this study only focuses on IS, the results will not be generalizable to the wider population of terrorist organisations. However, the primary aim of this study is to gain insight into responsibility claiming by IS, and it is therefore not concerned with generalizing the results onto other terrorist organisations. Still, insights derived from this study may serve to inform the larger debate on responsibility claiming for terrorist attacks.
This chapter has set out the methodology of this thesis, including its research design, case selection, operationalization of variables, methods of data collection - and analysis, as
well as an assessment of potential validity issues. The next chapter will describe the results of the conducted research analysis.
4. Results
This chapter displays the results of the conducted analysis. First, the distribution of data across the variables included in the analysis is described. Then, the results of the logistic regression analysis are presented and their meaning in relation to the hypotheses are discussed.
4.1 Description of variables
Of the 3654 attacks attributed to IS from its official birth in 2014 to the end of 2017,
responsibility was claimed in 1530 cases (41,9%). The proportion of claimed attacks was not constant over the years, as is shown in Table 2, which provides an overview of claim rates for each year in the sample. The percentage of claimed attacks fluctuated from 30.1% in 2014 up to 50% in the next year. In 2016, the percentage of claimed attacks dropped to 38.0%, while it rose again to 47.1% in 2017. Remarkably, the amount of claimed and unclaimed attacks in 2015 was equally divided, with exactly 478 observations in each category.
Table 2
Responsibility claims by year
2014 2015 2016 2017
n % n % n % n %
Unclaimed 473 69.9 478 50 705 62.0 468 52.9
Claimed 204 30.1 478 50 432 38.0 416 47.1
Observations 677 100 956 100 1137 100 884 100
Note. Attacks and claim rates in 2014 only include cases from June 29, 2014, and beyond.
The vast majority of attacks were perpetrated in Iraq (85,0%), while 10,5% of attacks took place in Syria. The remaining countries under attack by IS included Australia, Bahrain, Belgium, Egypt, France, Georgia, Germany, Indonesia, Iran, Jordan, Lebanon, Libya, Malaysia, the Philippines, Russia, Saudi Arabia, Somalia, Tunisia, Turkey, the United
Kingdom, the West Bank and Gaza Strip, and Yemen. On average, these countries scored just below the threshold of 6 for democracy on the Polity IV scale (M = 4.18; SD = 4.8). However, as Iraq was qualified as a democracy according to the Polity IV scale, the large number of attacks in Iraq resulted in 86,7% of attacks being coded as perpetrated against a democracy.
