Economic Inequality, Ethnic Group Dynamics, and Domestic Terrorism: The Destructive Effects of Collectively Experienced Grievances

(1)

Department of Peace and Conflict Research Uppsala University, January 2020

Bachelor Thesis – 15 credits

Supervisor: Espen Geelmuyden Rød

Economic Inequality, Ethnic Group Dynamics, and Domestic Terrorism

The Destructive Effects of Collectively Experienced Grievances

Tim Gåsste

(2)

1

Abstract

This paper studies how economic inequality between ethnic groups affects domestic terrorism. Drawing on theories concerning horizontal inequality, ethnic group dynamics, political violence, and domestic terrorism, the following hypothesis is developed: Increased economic inequality between ethnic groups increases domes- tic terrorism. In order to test the hypothesis and answer the research question, an ordinary least squares (OLS) regression analysis is conducted using a country-year dataset containing relevant control variables and a total of 2837 observations. The results yielded evidence that supports the hypothesis and the supposition that the economic inequality between ethnic groups, on average, increases domestic terrorist attacks. This central finding could, due to its novelty, yield significant academic and policy implications. However, as a consequence of the inherent challenges of the complicated object of study and of limitations in the research design, more rigorous investigation is needed in order to compliment, rectify or consolidate the reliability and validity of the findings.

Keywords:

Ethnicity, economic inequality, horizontal inequality, domestic terrorism, collective grievances, ethnic group dynamics, violent group mobilization

(3)

2

List of Tables

Table 1: Descriptive Statistics... 25

Table 2: The Effects of Economic Ethnic Inequality on Domestic Terrorism ... 27

List of Figures

Figure 1. Causal Graph – The relationship between the phenomena of interest ... 15

Figure 2. Histogram Independent Variable ... 26

Figure 3. Histogram Dependent Variable... 26

Figure 4. Scatterplot – Visualizing Model 1 ... 28

Figure 5. Effect plot – Visualizing Model 3 ... 28

Figure 6. Causal Graph - Social Causal Mechanism ... 32

Figure 7. Data diagnostics based on Model 3 ... 43

Abbreviations

ARSA Arakan Rohingya Salvation Army

EPR Ethnic Power Relations

GDP Gross Domestic Product

GTD Global Terrorism Database

GTI Global Terrorism Index

IEP Institute for Economics and Peace

ITARATE International Terrorism: Attributes of Terrorist Events

MAR Minorities at Risk

OLS Ordinary Least Squares

PRIO Peace Research Institute Oslo

PWT Penn World Table

UCDP Uppsala Conflict Data Program

STHDA Statistical tools for high-throughput data analysis

(5)

4

1. Introduction

1.1 Research Relevance and Research Question

During the last decade, scholars have utilized cross-country data to examine the effects of various phenomena – such as minority discrimination, ethnic fractionalization, and interregional inequality – on domestic terrorism. However, to the best of my knowledge, no previous research has studied how economic ethnic inequality, defined as the economic inequality between all relevant ethnic groups within a country, affects domestic terrorism. Since there are various reasons to believe that the two phenomena are related there are also reasons to believe that the omission may be important. For instance, robust findings within the peace and conflict literature have established economic inequality between ethnic groups as a significant predictor and cause of other manifestations of political violence such as civil wars and rebellions (Stewart, 2002, 2010; Cederman et al., 2013, 2015). Moreover, the two phenomena have also shown a tendency to coincide. For example, on the one hand, during the extreme ethnic inequality in South Africa during the 1980s, the country experienced a surge in terrorism that resulted in 725 terrorist attacks in a five-year period between 1984 and 1989, which subsided rapidly after the end of apartheid (MacFarlane, 2003; Cederman et al., 2015). On the other, the same type of estimate in South Africa between 2001 and 2006 yielded 2 terrorist attacks. Additionally, Chechen’s terrorist has been responsible for some of the most devastating terrorist attacks in modern times (CFR, 2010) but have also been one of the most economically disadvantaged and persecuted ethnic groups in Russia. This relative disadvantage has been the primary cause of many of the grievances that has escalated into violent group mobilisation and domestic terrorism (See for example: Derluguian, 2005)

Furthermore, while the perception that inequality generates violence within societies dates back thousands of years, group-based inequalities coinciding with identity cleavages such as ethnicity or religion – also known as horizontal inequalities – has shown to facilitate violent group mobilization and generate political violence to a significantly greater extent (Stewart, 2002; Østby, 2008; Cederman et al., 2013). Thus, the principal purpose and contribution of this thesis is to address this gap by examining the relationship between economic ethnic inequality and domestic terrorism by drawing on reputable theoretical arguments on horizontal inequality and political violence.

Terrorism and economic inequality are two of the most polarizing and destructive phenomena in modern times. During the last decades, terrorism has grown to be the most critical

(6)

5 national security threat for many countries, and in 2014 the number of casualties from terrorism reached an all-time high (Ezcurra and Palacios, 2016, p.60; GTI, 2015, p.2). This surge in terrorist activity has threatened political and economic stability, undermined security and peace- keeping efforts, and been responsible for an enormous number of casualties (Enders and Olson, 2012; GTI, 2015; Abadie, 2006, p.50; Enders et al., 2011, pp.322–323). While transnational terrorism has been the primary object of both scholarly and medial focus, domestic terrorism stands for the majority of all terrorist attacks and is naturally intertwined with the inequalities and grievances between sub-national groups (Abadie, 2006, p.50; Enders et al., 2011, pp.320, 322–323; Ezcurra and Palacios, 2016, p.61). Terrorism is typically considered domestic when the perpetrator, venue and target are all from the same country (Enders et al., 2011, p.321).

Furthermore, economic inequality in the form of wealth and income disparities has increased rapidly since the 1980s which has dampened growth, as well as fuelled economic and political volatility, corruption, and conflict (Dabla-Norris et al., 2015, pp.5–9; Alverado et al., 2018, p.7,12). Thus, a better understanding of the intricacies of both phenomena in general and the relationship between them, in particular, is crucial for politicians, policymakers and the research community to produce effective countermeasures and ameliorate these devastating effects (Ezcurra and Palacios, 2016, p.60). Especially since research has shown that accurately targeted social welfare policies (Burgoon, 2006, p.176; Krieger and Meierrieks, 2010, p.902), as well as ethnic inclusion and accommodation efforts (Gleditsch and Polo, 2016), tend to decrease the risk of domestic terrorism.

Despite its substantial impact on other forms of political violence such as rebellion and conflict (Stewart, 2002, 2010; Østby, 2013; Cederman et al., 2010, 2013, 2015), the effects of economic ethnic-based inequality on domestic terrorism remain largely understudied. Previous empirical research on domestic terrorism has examined the effects of political and economic discrimination of ethnic minorities (Piazza, 2011, 2012; Choi and Piazza, 2016; Ghatak et al., 2019), interregional inequality (Ezcurra and Palacios, 2016), as well as ethnic polarization and fractionalization (Foster et al., 2013; Danzell et al., 2019). However, no empirical research has to the best of my knowledge utilized data on all relevant ethnic groups – including ethnic majority groups – to produce a country-year comparative study that examines how the economic inequality between ethnic groups affects domestic terrorism. Thus, the purpose of this thesis is to address the defined gap by conducting a large-N quantitative study that examines this rela- tionship. This primary aim and purpose yield the following research question: How does eco- nomic inequality between ethnic groups affect domestic terrorism?

(7)

6 The thesis complements the existing literature in three primary ways. First and foremost, by making an empirical contribution by providing evidence that supports a correlation between the phenomena of interest and thereby both acknowledging a gap in the literature and partially filling it. Furthermore, shedding light on the relationship also generates avenues for future research and enables more robust and rigorous investigation that could yield invaluable findings for policymakers and society at large. Second, the thesis makes a minor methodological contribution by aggregating more nuanced data from the Cederman et al. (2015) dataset to better measure horizontal inequality between ethnic groups and study its effect on domestic terrorism on a country-year level of analysis. Third, the thesis also makes a minor theoretical contribution by combining the utility of theories on horizontal inequality, ethnic group dynamics, grievances and political violence, as well as expanding and contextualizing them to produce a plausible causal pathway.

1.2 Argument and Findings

The main theoretical argument, as well as the findings from the regression analysis, will now be summarized. I argue that economic inequality between ethnic groups fuels collectively experienced grievances along ethnic lines, which in turn enables violent group mobilisation among both the privileged and disadvantaged ethnic groups in society. Among the disadvantaged groups as a direct consequence of the relative deprivation and among the privileged groups as a consequence of the fear of losing the relative advantage. The violent group mobilisation, in turn, enables the grievances to be addressed through the most cost-effective violent method available to most ethnic groups; domestic terrorism. These processes are, in turn, accelerated and intensified by the nature of ethnic group dynamics which will be covered more in-depth in the theory section.

The OLS models yielded findings that support the hypothesis and the theoretical argument by establishing a systematic correlation between economic ethnic inequality and domestic terrorism, which implies that economic inequality between ethnic groups on average increase domestic terrorism.

1.3 Disposition

Following the introduction, the rest of the thesis will be structured thusly. First, previous research related to the relationship between economic ethnic inequality and domestic terrorism

(8)

7 will be reviewed, and the theoretical concepts, definitions and theories that together constitute the basis for the theoretical argument will be discussed. Subsequently, the theoretical argument and the hypothesis will be presented. Second, in the research design, the empirical data, cases, and scope conditions are discussed followed by a presentation of the variables used to test the hypothesis. Third, in the results section, the descriptive statistics and regression models are presented, followed by an analysis and substantiation of the findings. Furthermore, an assess- ment of the quality of the research design and how the thesis can provide avenues for future research is conducted. Finally, a summary of the thesis purpose, findings and contributions are made in the conclusion followed by the bibliography and the appendix which includes a diag- nostic plot that further demonstrates the structure and fit of the data.

2. Theory and Previous Research

2.1 Previous Research

Because of the mentioned scholarly, medial and political focus on transnational terrorism, the literature on domestic terrorism is comparatively limited (Berkebile, 2017, p.1,2). However, during the last decade, there has been a sizable increase in interest and scholarly focus on the specific dynamics of domestic terrorism, which has generated more empirical research and a better understanding of the phenomenon.

First and foremost, the literature on the economic and political discrimination of minorities have yielded significant findings. Economic discrimination of minorities has been deemed to increase the likelihood of experiencing domestic terrorist attacks and been established as a stronger and more substantive predictor of domestic terrorism than for instance economic development (Piazza, 2011, 2012). Countries in which ethnic minorities are politically discriminated, marginalized and excluded have also been found to be significantly more likely to experience terrorist attacks (Piazza, 2012; Choi and Piazza, 2016; Ghatak et al., 2019). However, while this sub-section of the literature on domestic terrorism studies the effects of ethnic group marginalization and discrimination, they do not capture the inequality between all relevant ethnic groups and how the inequality affects domestic terrorism. For example, Gurr’s (1993) Mi- norities at Risk (MAR) dataset that the articles are primarily based on have been criticised for only taking the ethnic groups that are disadvantaged, discriminated and marginalized into consideration. These circumstances arguably make the data biased and do not provide a solid basis for nuanced comparisons (Cederman et al., 2015, p.808).

(9)

8 Second, Ezcurra and Palacios (2016) have concluded that high inter-regional inequalities within countries tend to increase the number of domestic terrorist attacks. Their results show that higher levels of inequality within a country typically alter the cost-benefit calculus of the terrorist by increasing their perceived benefits of using violence (Ezcurra and Palacios, 2016, p.71). Moreover, Ezcurra (2019) has found that ethnic segregation has a positive and significant effect on domestic terrorism which means that societies, where ethnic groups are more spatially concentrated, have a higher risk of domestic terrorism. As mentioned in the article by Ezcurra and Palacio (2016) and further emphasised in the article by Ezcurra (2019) the disparities between regions seem to be dependent on the concentration of, and segregation between, ethnic groups, which supports the notion that ethnic inequality is important and could have a considerable effect on domestic terrorism.

Third, Studies have also shown that terrorism is more likely to occur in societies that are ethnically polarized (Krieger and Meierrieks, 2010; Python et al., 2017) and that this effect is strengthened when the economic conditions are worse (Danzell et al., 2019). This does not mean that ethnic fractionalization and the presence of many ethnic groups have been established as a predictor of terrorism, contrary to that notion ethnic fractionalization has shown to be a negative predictor of domestic terrorism in some cases (Choi and Piazza, 2016, p.51).

Fourth, armed conflict acts as the most consistent predictor and driver of domestic terrorism with 95 per cent of all casualties from terrorism occurring within countries engaged in armed conflict (GTI, 2019, p.2). When combined with countries with high levels of repression, political terror – involving for example imprisonment without trial, extra-judicial killings, and torture – and government mass killings which are other potent drivers for domestic terrorism (Avdan and Uzonyi, 2017; Piazza, 2017), that same number is increased to 99 per cent (GTI, 2019, p.2).

Finally, partially attributed to the complexity of the phenomena, the effect of many factors on domestic terrorism is dependent on the country-specific circumstances or the different aspects of the phenomena. For instance, the impact of economic growth, development, and education on domestic terrorism has been determined to vary based on country-specific circumstances and on what aspects of each phenomena that is measured (Gries et al., 2009; Krieger and Meierrieks, 2011; Choi, 2015; Brockhoff et al., 2015). Furthermore, while some other relevant alternative explanations such as population size and regime type will be discussed at length in the research design, there are naturally various phenomena that has been omitted from

(10)

9 the regression analysis because of the multidimensionality of terrorism and the limitations of the thesis.

Conclusively, while the majority of the discussed studies are related to the topic of the thesis, none of them looks specifically at all relevant ethnic groups and how the economic inequality between them affects domestic terrorism, which constitutes a considerable research gap that this thesis aims to partially fill. However, what the majority of these studies have in common is that they base their logic and theoretical arguments on the same theoretical structures and assumptions, namely in the seminal theories on inequality, grievances and violence. Hence, the origin, development and contemporary adaptation of this theoretical framework will now be discussed and applied to examine and explain the relationship between the phenomena of interest.

2.2 Inequality, Violence, and Ethnicity

Grievance theory has primarily been devoted to explain political violence in the form of conflict and rebellion but has proven to have explanatory value when it comes to domestic terrorism as well. Theories on the relationship between inequality, grievances and violence can be dated back many centuries and have been addressed in the works of historical figures such as Plato and Machiavelli (Ezcurra, 2019, p.61). However, the more direct origin stems from different theoretical approaches such as Marxist theory on class struggle and revolution (Marx and Engels, 1887), theory on ethnic conflict and structural inequality (Galtung, 1964; Hechter, 1975; Horowitz, 1985), and theory on relative deprivation (Davies, 1962; Gurr, 1970). While these approaches are diverse, they all share the understanding that violence and conflict is a consequence of grievances among the relatively disadvantaged within society.

Sigmund Freud (1922) expanded on Karl Marx theories on economic exploitation and political violence by putting emphasis on the concepts of frustration and alienation. Based on the work of Freud, frustration and aggression theory was developed which would lay the foun- dation for relative deprivation theory and thereby the modern understanding of grievances and political violence (Østby, 2008, pp.208–209). The core assumption of frustration and aggression theory is that frustration occurs when individuals experience grievances and in turn cause all types of aggression and violence (Dollard et al., 1939, p.1). Ted Gurr’s (1970) seminal book on relative deprivation utilized these theoretical underpinnings and the work on the subject already refined by Davies (1962) to further develop the theory. Gurr defined relative deprivation as the discrepancy between people’s expectations and their actual situation and argued that

(11)

10 when this gap becomes too great that constitutes sufficient frustration and motivation to produce violence and conflict (Gurr, 1970, p.13). This type of inequality that measures the dispar- ities between either individuals or households is commonly referred to as vertical inequality and has been the bulk of all the research studying the connection between inequalities and violence (Østby, 2013, pp.206–207). Gurr (1993; 2000) later expanded his work to focus on the socio-economic and political discrimination of ethnic minorities and found that the ethnically based grievances generated by these factors dramatically increased the risk of collective violence by increasing ethnic mobilization. However, while being one of the most prominent explanations that link inequality with conflict and violence, relative deprivation and grievance theory did not hold up well in the contemporary empirical literature.

The pioneering articles by Fearon and Latin (2003) and Collier and Hoeffler (2004) largely dismissed grievances as a substantive cause of conflict. Collier and Hoeffler (2004, pp.587–588) found little to no support for their proxies for grievances in their seminal article on the causes of civil wars which exposed the inadequate practical utility of relative deprivation, grievance theory and vertical inequality as a predictor for civil wars and political violence.

Conversely, contemporary research on the relationship between inequality and violence has established that the inequalities experienced by societal groups in general and by ethnic groups in particular are considerably more likely to lead to violent group mobilisation and armed conflict (Cederman et al., 2013; Stewart, 2010; Østby, 2013). However, while the relationship between ethnicity, group dynamics, and grievances has been utilized to explain the occurrence and characteristics of armed conflict and rebellion, its effect on terrorism and other manifestations of political violence have not been given the same attention.

During the last decade, the research on ethnic groups and their role in political violence and armed conflict has been given considerably more focus within the field of peace and conflict. Ethnic cleavages have been found to affect the use of one-sided violence against civilians, the likelihood of civil war onset, as well as for instance the intensity and duration of armed conflicts (Fjelde and Hultman, 2014; Cederman et al., 2010; Eck, 2009; Wucherpfennig et al., 2011). Research on horizontal inequality pioneered by Frances Stewart (2002, 2010) that em- phasise the importance of ethnic group dynamics re-established grievance theory as one of the most prominent explanations for the relationship between inequality and armed conflict. The central argument of this new wave of grievance theory is that socio-economic and political horizontal inequality produce collective disadvantages and collective privileges among different culturally defined groups. While the disadvantaged groups generate collective grievances

(12)

11 as a direct consequence of the relative deprivation, the relatively privileged groups generate collective grievances through the fear of losing their collective advantage (Stewart, 2002, 2010;

Østby, 2013, pp.216–217). These grievances, in turn, produce a strong motive for collective action and the shared group identity act as a mobilizing agent and catalyst for addressing the grievances through political violence (Stewart, 2002, 2010; Østby, 2013, pp.215–217;

Cederman et al., 2013, pp.37, 44). The link between horizontal inequalities, collective grievances and political violence has since then been further supported by a rich body of evidence and literature (Regan and Norton, 2005; Østby, 2008, 2013; Cederman et al., 2010, 2013, 2015;

Buhaug et al., 2014; Mitra and Ray, 2014).

Conclusively, while the theoretical framework on horizontal inequalities and political violence once again is considered one of the most prominent theories on the occurrence of armed conflict, the framework has not been contextualized and expanded to explain the effects of economic ethnic inequality on domestic terrorism which will be the core theoretical argument driving this thesis.

2.3 Domestic Terrorism

To discuss domestic terrorism, one must first understand what terrorism is, how it is defined in the literature, and what challenges the complicated phenomena entail. A conceptualization or theoretical definition of terrorism that encompasses all the underlying meanings and attributes inherent to the multidimensional phenomena have neither been produced nor is it likely to be in the future (Silke, 2004, pp.2–3). Silke (2004, pp.2–6) argues that this failure to reach an agreed definition has halted and compromised a lot of the comparative research on the topic, while the tireless search for one entails the seriousness of the issue. That does not mean that the commonly used conceptualizations of terrorism are severely flawed or lack utility. Conversely, different conceptualizations and definitions of terrorism can have great utility if they are ade- quately designed and contextualized for their particular purpose. While solving the philosoph- ical question of what terrorism is seems impossible, being transparent and utilizing the same criteria when conducting research is possible and a necessity to produce comparable findings on terrorism (Berkebile, 2017, p.5).

There exists noticeable patterns in the literature that outlines the phenomena’s core components that are typically included in the most prominent definitions. These core components are the target and target audience of the action, as well as the premeditation, method, goal and

(13)

12 identity of the perpetrators, which together forms the most imperative criteria to separate terrorism from other similar phenomena such as crime, riots or armed conflict (Enders et al., 2011, p.321; Berkebile, 2017, pp.6–11; GTD Codebook, 2019, p.10).

First, the methods used have to be either violence or the threat of violence to be considered terrorism. Second, in order to differentiate terrorism from what is deemed as “legitimate warfare” the target of the attack have to be a non-combatant. Third, the target audience has to be bigger than the immediate victims of the attacks. Otherwise, the attack can be considered private instead of political and is better characterized as, for example, a criminal act. Fourth, terrorist attacks are not spontaneous acts, but rather premeditated with a particular object in mind. Fifth, for something to be considered terrorism, the goals of the perpetrators have to be either socio-economic, political or religious. Sixth, terrorism is perpetrated by sub-national groups or individuals. Acts committed by states and governments are thereby not considered terrorism by this definition (Enders et al., 2011; Berkebile, 2017; GTD Codebook, 2019).

Drawing on these core components and criteria for terrorism developed and utilized by for instance Enders et al. (2011, p.321) and the Global terrorism database (GTD, 2019, pp.10- 11) this thesis will define terrorism as the premeditated use of violence or the threat of violence by sub-national actors that targets non-combatants to influence and intimidate a larger audience in order to obtain a religious, socioeconomic or political objective. Furthermore, terrorist acts where the perpetrators, target and venue all share the same nationality are considered as domestic, while terrorist acts where any of these nationalities do not match are considered transna- tional (Enders et al., 2011, p.321). Hence, building on the prior definition of terrorism domestic terrorism will be defined as the premeditated use of violence or the threat of violence by sub- national actors that targets non-combatants to influence and intimidate a larger audience in order to obtain a religious, socioeconomic or political objective where the nationality of the venue, victims and perpetrators are the same.

2.4 Economic Ethnic Inequality

As previously discussed, the research on horizontal economic inequality has provided the most robust evidence in relation to ethnic dynamics and political violence, and will thereby be the type of inequality studied in this thesis (Cederman et al., 2013; Østby, 2008; Stewart, 2002, 2010). Studies have shown that economic disparities along ethnic lines are deeply intertwined with the political and social dimensions of inequality as well, which makes it a suitable general

(14)

13 indicator and predictor for overall inequality (Stewart, 2010; Cederman et al., 2013). Further- more, because of the lack of large-scale country-year datasets covering the social or political dimensions of horizontal inequality, the quantitative methodology utilized in this thesis would not have been as useful when examining those types of inequalities.

The theoretical literature on the role of ethnicity in the relationship between economic inequalities and political violence has as previously pointed out, experienced an upsurge. The seminal works of Stewart (2002, 2010), Østby (2008, 2013), and Cederman et al. (2010, 2013, 2015) among others has established that the economic inequalities experienced between societal groups in general and ethnic groups in particular are considerably more likely to lead to violent group mobilisation and armed conflict. The economic inequality between ethnic groups has also been shown to negatively correlate with economic development, which entails that countries where economic performance is divided along ethnic lines tend to have a lower level of development (Alesina et al., 2016, p.482). In this study economic ethnic inequality is defined as the relative disparities in assets, wealth and income between ethnic groups in society. Ethnic groups are defined as groups that are characterised by their sense of commonality based on shared culture and ancestry. Markers for the shared culture and ancestry could, for example, be a combination of shared phenotypes, languages or religions (Vogt et al., 2015, p.1329; OECD, 2019; Alesina et al., 2016, p.429).

As previously mentioned, different aspects of ethnic group dynamics, inequalities and cleavages have been found to, for instance, affect the use of one-sided violence against civilians (Fjelde and Hultman, 2014), the likelihood of civil war onset (Cederman et al., 2010;

Wucherpfennig et al., 2011), as well as the intensity (Eck, 2009) and duration of armed conflicts (Wucherpfennig et al., 2012). Furthermore, while societies are organized along different lines – for instance, ethnic, class, religious, ideological, political or geographic – when they engage in civil war it is typically between ethnic groups (Blattman and Miguel, 2010, p.12,16). Denny and Walter (2014, pp.199-200,207) argue that this is because of three particular reasons. First, ethnic lines within societies tend to historically coincide with access to power which increases nepotism, inequality, cleavages and thereby grievances and political violence. Second, ethnic groups are more effective at mobilising support to demand change because they tend to have higher group cohesion, as well as be more tightly clustered geographically. Third, ethnic groups tend to face bargaining challenges to a greater extent since their identity naturally is more identifiable and fixed compared to other groups because their commonality is primarily characterized by shared ancestry and culture (Denny and Walter, 2014, p.200).

(15)

14 Conclusively, while the literature primarily focuses on armed conflict as the outcome and man- ifestation of horizontal inequalities and ethnic group dynamics the same reasons and theoretical structure, arguably apply to domestic terrorism (Piazza, 2011, 2012; Ghatak et al., 2019). This argument and the proposed relationship between the phenomena will be discussed in depth below.

2.5 Theoretical argument

The aim of this thesis is to examine how the economic inequality between ethnic groups affect domestic terrorism. The theoretical argument – as illustrated in figure 1 – builds on and pre- dominantly adheres to the logic and assumptions of the previously discussed literature on horizontal inequalities, grievances and political violence. The central theoretical framework on horizontal inequalities is built on the assumptions that group-based relative disparities generates collectively experienced grievances along group lines which in turn increases violent group mobilisation and thereby also political violence (Stewart, 2002, 2010; Østby, 2008, 2013;

Cederman et al., 2010, 2013, 2015). While following the same logic and similar theoretical structure the causal mechanism presented in this thesis is structured to fit its specific circumstances and purpose of examining the effect of economic inequalities between ethnic groups on domestic terrorism.

Economic inequality between ethnic groups can arguably be considered the archetype of a volatile horizontal inequality that can generate political violence in general and domestic terrorism in particular. The primary driver behind this characteristic is the ethnic group dynamics that has been discussed above, but that will be further illustrated by the following argument.

First, following the logic of the grievance literature, horizontal inequality – in the form of economic ethnic inequality – produce and consolidates relative disadvantages and privileges among ethnic groups within society. Second, these relative disparities between ethnic groups produce grievances among both the disadvantaged and the privileged groups. In the disadvantaged groups as a direct consequence of the relative deprivation, and among the privileged groups as an indirect consequence of the fear of losing their relative position of supremacy (Østby, 2013, pp.216–217). Furthermore, due to the nature of ethnic group dynamics and the historical distribution of power along ethnic lines, these grievances tend to be multiplied (Denny and Walter, 2014, pp.199-200,207) and lead to more violence against civilians (Fjelde and Hultman, 2014). Third, the accelerated increase of collective grievances as a consequence of economic horizontal inequality and ethnic group dynamics acts as a powerful mobilising

(16)

15 agent that increases violent group mobilisation (Stewart, 2002, 2010; Cederman et al., 2010, 2013, 2015; Østby, 2008, 2013, p.2; Denny and Walter, 2014). This step is crucial since the decision to use violent tactics to address the group’s grievances is dependent upon the ability to first mobilize a group capable and willing to use such tactics (Ghatak et al., 2019, p.443).

Figure 1. Causal Graph – The relationship between the phenomena of interest

Furthermore, because ethnic groups tend to be spatially clustered, as well as share culture and heritage, they have a considerably easier time mobilizing support to attempt to force change through violence. Consequently, the risk of violent group mobilisation in the form of terrorist organisations or rebel groups is thereby increased significantly (Denny and Walter, 2014, p.199).

Finally, the risk of mobilised violent ethnic groups to address their grievances by utilis- ing the cost-effective and powerful method of domestic terrorism increases. The theoretical and empirical literature on horizontal inequalities primarily suggests that these circumstances cause rebellion and armed conflict. However, it is more common that ethnic groups use the method of terrorism to address their grievances and pursue their goals through force because it is a significantly more cost-effective and plausible strategy to most groups (Choi and Piazza, 2016, pp.37–38). Domestic terrorism is a powerful tool for weaker actors to engage in asym- metric warfare. It demands less organizational sophistication, military capabilities and re- courses, which makes it ideal for most ethnic groups to address their grievances if there are no non-violent alternatives (Choi, 2015, pp.37–38). An empirical example of this process can arguably be seen in Myanmar, where institutionalized discrimination against the muslin ethnic

(17)

16 minority group Rohingya has generated severe collective grievances that, in turn, has enabled the mobilisation of the Arakan Rohingya Salvation Army (ARSA). ARSA conducted several terrorist attacks, which culminated in a brutal military counter-campaign by the Myanmar military that destroyed nearly 100,000 Rohingya settlements, killed at least 6000 civilians and in- itiated a mass exudes of Rohingya from Myanmar in 2018 (Albert and Maizland, 2019).

Furthermore, because ethnicity is typically a more fixed and identifiable group identity, ethnic groups are more likely to encounter bargaining challenges and are thereby also less likely to address the grievances in a non-violent manner which further increases the risk of violent outcomes (Denny and Walter, 2014, pp.199–200).

Conclusively, I argue that increased economic inequality between ethnic groups fuels collectively experienced grievances along ethnic lines which significantly increases violent group mobilisation among both the disadvantaged and privileged groups within society. The violent group mobilisation then enables the grievances to be addressed through the most cost- effective violent method; domestic terrorism. The theoretical argument yielded the following hypothesis: Increased economic inequality between ethnic groups increases domestic terror- ism.

3. Research Design

3.1 Data, cases, and scope conditions

The principal aim of this study is to examine the effect of economic ethnic inequality on domestic terrorism. As illustrated by the theoretical argument and the hypothesis, the effect of economic ethnic inequality on domestic terrorism is expected to be positive. Meaning that an increase in economic inequality between ethnic groups is expected to be related to an increase in domestic terrorism after the alternative explanations have been controlled for.

The dependent variable is based on a country-year aggregate of domestic terrorism by Piazza (2011) derived from a dataset developed by Enders et al. (2011). Enders et al. (2011) undertook several steps to clean up and refine the comprehensive data from the GTD in order to produce a less biased dataset that holds transnational and domestic terrorism separate. Fur- thermore, the data on the independent variable originates from a dataset developed by Cederman et al. (2015) that triangulates different sources of data on economic ethnic group inequality. To match the unit of analysis of the country-year aggregate of domestic terrorism,

(18)

17 the triangulated group data is aggregated to a country-year unit of analysis. Moreover, this ag- gregation and data transformation was necessary to make a nuanced comparison because of the great lack of data on the topic.

The dataset used in the regression analysis contains country-year data on 107 countries from 1970 to 2006. After the exclusion of incomplete observations, the data yields 2837 complete observations containing data on the dependent variable and all of the independent variables. The same number of observations are used in all models for better comparison. The population and time period are primarily determined by the data-availability. The GTD started gathering data on terrorism in 1970, which marks the start of the used dataset. Enders et al.

(2011) utilized data from 1970 until 2007 when developing their dataset. However, the utilized aggregate ranges between 1970 and 2006, which marks the time period of this study.

The scope of this thesis will now be further discussed. First, as previously pointed out, horizontal inequality is the type of inequality that is primarily studied in this thesis, as opposed to vertical inequality which is held constant with a control variable. Second, while political and social inequality is of great importance, this study focusses solely on the economic dimension of horizontal inequality. Third, while the literature on horizontal inequality encompasses various societal groups based on for example religion, geography or political leanings, this study focusses exclusively on ethnic groups. Finally, this thesis only studies domestic terrorism which excludes all terrorism that is defined as transnational or uncertain by the Enders et al. dataset (2011, p.323).

To test the hypothesis and examine the relationship between the phenomena of interest, a set of linear regressions, also known as OLS regressions will be used. OLS regressions are typically the first type of regression models to consider when studying a continuous dependent variable since they tend to provide great utility. Linear regressions can predict outcomes, as well as determine the existence of a statistical relationship between variables while at the same time controlling for confounding variables that could make the relationship spurious. However, there are some limitations to this type of regression models, particularly when studying a dependent variable such as domestic terrorism. OLS tend to be sensitive to outliers and multicol- linearity, which have to be taken into consideration when assessing the models and the findings.

First, a bivariate regression model will be used that tests the relationship between the independent and the dependent variable. Second, a multivariate regression model containing all the control variables will be used to isolate their effects on the dependent variable. Third, a multivariate regression model will be carried out that includes the independent variable and the

(19)

18 control variables in order to capture the effect of ethnic inequality on domestic terrorism while controlling for the relevant alternative explanations. The statistical models determine if the hypothesis is supported.

3.2 Dependent variable

Domestic terrorism is operationalised as a continuous count variable that reflects the number of domestic terrorist attacks per country-year following the operational inclusion criteria established by the GTD and Enders et al. (2011). As previously discussed, the dependent variable has been derived from a dataset developed by Enders et al. (2011) and aggregated into country- year units by Piazza (Piazza, 2011, p.344) for the period between 1970 and 2006. Enders et al.

(2011, pp.321–322) separated the domestic from the transnational terrorist attacks in the GTD data between 1970 and 2007 by matching the country where the attacks took place with the nationality of the perpetrators and victims or targets. If the nationality of the venue, target and perpetrators matched the terrorist attack was considered domestic (Enders et al., 2011, pp.322–

323). Besides the partitioning which according to the authors themselves is the most complete and comprehensive of its kind, Enders et al. (2011, pp.321–322) also cleaned the data from all noticeably mischaracterized or dubious terrorist attacks, eliminating approximately one-fifth of all GTD incidents between 1970 and 2007. For incidents to be considered terrorist attacks and be included in the GTD¹ in the first place, the following three operational criteria must be present. The perpetrators must be sub-national actors, the incident must be intentional, and the incident must entail violence or the immediate threat of violence. Besides these criteria, at least two of the following criteria must be met in order for the incident to be included in the database:

(GTD Codebook, 2019, pp.10–11).

1. The act must be aimed at attaining a political, economic, religious, or social goal.

2. There must be evidence of an intention to coerce, intimidate, or convey some other message to a larger audience (or audiences) than the immediate victims.

3. The action must be outside the context of legitimate warfare activities.

1 For further information on the definitions, coding and processing of the data, access the GTD 2019 Codebook and other important information on the projects homepage: https://www.start.umd.edu/research-projects/global- terrorism-database-gtd

(20)

19 When dealing with such a complex and multidimensional phenomenon as terrorism, achieving both validity and reliability is challenging, to say the least. While the reliability of the operational definition is questionable, the validity between the conceptual definition and the measurement is arguably solid. Both the conceptual definition and the operationalisation criteria for terrorism originates from the same sources (Enders et al., 2011; GTD Codebook, 2019) and focus on the same core attributes – method, perpetrator, goal, target, target audience and premeditation – which make the measurement accurately reflect all the most central attributes of the concept.

Furthermore, both the conceptual definition and the operational inclusion criteria define the matching of the nationality of the venue, perpetrators and target as the central attribute that separates domestic and transnational terrorism (Enders et al., 2011; GTD Codebook, 2019).

However, the operationalisation has some significant flaws when it comes to reliability. While the GTD uses many criteria to separate terrorism from other phenomena better and to cover all the attributes in the conceptual definition, the number of criteria and the complexity of them produce ambiguity. For example, how does one delineate if a goal is social or a threat should be considered immediate? What is considered a large audience and legitimate warfare? This operational ambiguity increases the risk of the same observation being interpreted differently by different people and thereby lowers the reliability of the operational definition. To illustrate the point, the GTD has been deemed to have some significant measurement biases during cer- tain time periods which is believed to be a consequence of different interpretations, as well as changes in the operational criteria (Enders et al., 2011, p.322).

Conclusively, a concise operational definition interlinked with various criteria that en- compass many of the attributes of the underlying conceptual definition can provide great validity. However, the complexity and scope of the operationalisation can make it ambiguous and thereby less reliable. I argue that with such a complex phenomenon as terrorism, a minimum operational definition would not represent the concept accurately and thereby be of little use.

Furthermore, while the reliability is compromised, I argue that in the case of studying terrorism, this is the lesser evil because operationalisation that does not manage to capture the multidimensionality of the phenomena are inadequate.

While the open-source GTD and Enders et al. (2011) dataset are frequently used by top scholars in the field and constitute the most reliant, accessible and extensive data available on domestic terrorism, the data has some critical flaws and idiosyncrasies that make findings and inferences based on it weaker. This could potentially produce inept policy recommendations

(21)

20

:

and have severe consequences (Enders et al., 2011; Cubukcu and Forst, 2018, p.111). The coding procedures changed in 1998 to a narrower definition of terrorism which means that the structure of the data diverges, however, no documentation show in what way the earlier definition was broader which adds to the confusion (Enders et al., 2011, p.322). Furthermore, Enders et al. (2011) also argue that GTD overcounted events between 1991 and 1997, but undercounted transnational events before 1977, which entail some significant measurement biases. The GTD data is also based primarily on news sources which make the data only as reliable as the original sources even though there exist thorough vetting in the coding process. While Enders et al.

(2011) cleaned up substantial parts of the data it probably remains partially biased and plagued with analogous measurement errors. However, as previously pointed out, it still provides the most extensive and robust country-year data on domestic terrorism currently available.

3.3 Independent variable

Economic ethnic inequality is operationalised as a country-year ratio between the average per capita income of all the ethnic groups in the country and the per capita income of the poorest ethnic group(s). The deviation from the country mean is a positive number of one or higher.

For example, if all groups within a country have the same or similar per capita income the ratio would be 1(1:1). If the poorest group(s) is/are three times poorer than the ethnic group country average the ratio would be 3 (3:1) and if the poorest group(s) is five times poorer than the ethnic group country average the ratio would be 5 (5:1), and so on and so forth.

𝑬𝒕𝒉𝒏𝒊𝒄 𝑮𝒓𝒐𝒖𝒑 𝑰𝒏𝒆𝒒𝒖𝒂𝒍𝒊𝒕𝒚 𝑹𝒂𝒕𝒊𝒐 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 𝑖𝑛𝑐𝑜𝑚𝑒

𝑜𝑓 𝑎𝑙𝑙 𝑒𝑡ℎ𝑛𝑖𝑐 𝑔𝑟𝑜𝑢𝑝𝑠 𝐴𝑣𝑒𝑟𝑎𝑔𝑒 𝑝𝑒𝑟 𝑐𝑎𝑝𝑖𝑡𝑎 𝑖𝑛𝑐𝑜𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑝𝑜𝑜𝑟𝑒𝑠𝑡 𝑒𝑡ℎ𝑛𝑖𝑐 𝑔𝑟𝑜𝑢𝑝(𝑠)

The independent variable – and all the independent control variables – have been lagged for one year (t-1) to mitigate endogeneity problems caused by reversed causality. Furthermore, structural asymmetries such as economic inequalities between ethnic groups tend to persist over time which also helps diminish the risk of reversed causation (Tilly, 2009; Stewart and Langer, 2008). However, this measure and these circumstances are not enough to guarantee a time-order between the independent and dependent variable, which is crucial for making credible arguments based on the results of the regression analysis.

The dataset was developed by Cederman et al. (2015) to address the scarcity of nuanced and quality data on economic horizontal inequalities and their effect on the likelihood of ethnic

(22)

21 groups engaging in conflict. The results from the triangulated data showed clearly that both privileged and disadvantaged ethnic groups have a higher risk of engaging in armed conflict (Cederman et al., 2015, pp.818–819). However, while Cederman et al. (2015) studied the effects on conflict and rebellion on a group level, this thesis, as previously argued, aggregates the data to study the overall effect of economic inequality between ethnic groups in societies on domestic terrorism. This measurement could justly be criticised for studying the difference between the country ethnic group average and the maximum deviation since this could potentially paint an unrepresentative picture where one disadvantaged group could alter the ratio of an otherwise equal society. However, I argue, and the literature suggests that these extreme cases of ethnic group disadvantages are one of the most substantive and important drivers for political violence and domestic terrorism, and thereby are great indicators of domestic terrorism (Piazza, 2012; Choi and Piazza, 2016). Hence, an operationalisation measuring the maximum disparities between the country average and the most disadvantaged ethnic group(s) arguably provides the greatest utility.

Cederman et al. (2015, p.806) combine and weigh data from three primary sources based on their strengths and weaknesses under country-specific circumstances and find that the triangulation appears to provide considerably more robust findings than each isolated source.

The main types of data are survey estimates, geocoded data from the G-econ project, and night- lights emission from satellites. First, survey estimates have been proven to be a great source of nuanced data when it comes to horizontal inequalities and has been used by for example Østby (2008) in her seminal article on the subject. However, data limitations prevent worldwide coverage of survey data and because of that survey estimates are not currently sufficient to make a large-scale quantitative analysis containing over a hundred countries (Cederman et al., 2015, p.807). Second, the G-econ data set gathers geophysical data on economic activity around the world which provides truly encompassing worldwide coverage. However, the quality of the G- econ data in a lot of developing countries is compromised because the official statistics are poor. Furthermore, the disparities between ethnic groups when their settlements overlap are hard to decipher based on the G-econ data alone. Third, to address and compliment the other two data sources, night light emission data is used as a proxy for economic activity. While the nigh-light emissions have considerable weaknesses and act as a rather blunt indicator of economic activity it works best in predicting the economic activity in developing countries which is the primary weakness of the two other sources of data (Cederman et al., 2015, pp.807–809).

(23)

22 The dataset is based on the ethnic group structure from the Ethnic Power Relations (EPR) dataset which identifies and includes all politically relevant ethnic groups from all countries in the world where ethnicity has been politicized, and that has a population over 250,000 (EPR Codebook, 2019, p.1). The dataset considers an ethnic group politically relevant if it is either claimed to be represented by at least one political actor such as an active political organization or if group members are systematically politically excluded or discriminated against (EPR Codebook, 2019, p.3). Hence, the data includes not just the marginalized and excluded ethnic groups like for example the popular Minorities at Risk (MAR) dataset (Gurr et al., 1993), but also includes all other relevant political actors which makes it considerably more nuanced and less biased.

Both the validity and reliability of the operationalisation are arguably high. First, the conceptual definition focus on the economic disparities in form of income and wealth between ethnic groups which overlaps completely with the operational definition that uses the ethnic group inequality ratio, which measures ethnic group disparities in income and wealth. Further- more, the method of triangulation is known to effectively mitigate potential biases and improve measurement validity which strengthens the overall credibility of the data (Höglund and Öberg, 2011, p.191). Second, because of the simplicity of the operationalisation, the same type of measurement of the same data should yield identical results if applied multiple times which entails high reliability.

Conclusively, while the different data sources on economic ethnic inequalities all have significant flaws and would produce questionable results if used separately to conduct a large- scale country-year comparative study, instead a synergetic effect is achieved through the process of triangulation that makes the findings considerably more robust. Furthermore, while the triangulated data-set still has considerable holes and weaknesses where none of the mentioned sources are available, it is the most robust data available due to the complexity of gathering comparative data across time and space on horizontal inequalities.

3.4 Control variables

Due to the complexity of domestic terrorism, there exists a multitude of alternative explanations and relevant control variables. However, some have been found to have a greater impact than others and are thereby more frequently used by prolific scholars exploring the topic. The purpose of controlling for alternative explanations is to isolate the variables of interest as much as possible to prevent confounding variables from making the relationship spurious and the results

(24)

23 false or misleading. Drawing on previous research, the five control variables used in this thesis are the population size, GDP per capita, GINI coefficient, Level of Democracy and Armed conflict. All the control variables have been lagged one time period (t-1) to mitigate any confounding effects on the independent variable as well as to establish time-order in relation to the dependent variable.

First, larger populations have been associated with higher levels of domestic terrorist incidents (Savun and Phillips, 2009) and because of that, the natural log of the population of each country based on data from the Penn World Table (PWT) is added as a control variable (Heston et al., 2012). The primary argument behind this inclusion is that larger populations have been shown to increase the difficulty and cost of providing security which generates higher levels of vulnerability and thereby also a greater risk of terrorism (Eyerman, 1998). Further- more, larger populations naturally tend to increase the potential number of terrorists (Freytag et al., 2011; Kis-Katos et al., 2011), as well as cause more demographic stress and thereby greater recruitment pools for terrorist organizations (Krieger and Meierrieks, 2011). Finally, larger populations often entail greater heterogeneity which could affect both ethnic inequality and domestic terrorism.

Second, while the literature on the effects of poverty, low levels of development, and economic vertical inequality on domestic terrorism is inconclusive (Abadie, 2006), some studies suggest that the effects on domestic terrorism are substantial (Lai, 2007; Bird et al., 2008;

Dërin-Gure, 2009). Hence, existing disparities in development between countries will be controlled for using the natural log of GDP per capita based on data from the Penn World Table (Heston et al., 2012). Economic vertical inequality in the form of income and wealth disparities will be controlled for using the GINI coefficient retrieved from the World Bank (2019).

Third, While the effect of regime type in general and democracy in particular on domestic terrorism also have yielded inconclusive findings, it is one of the central phenomena that seems to have an impact on both ethnic inequality and domestic terrorism in one way or the other (Li, 2005; Choi and Piazza, 2016). Hence, the level of democracy will be added as a control variable using the Revised Combined Polity Score from the Polity IV Project (Marshall et al., 2018, p.17). On the one hand, studies have concluded that a higher level of democracy increases the risk of both domestic and transnational terrorism (Eubank and Weinberg, 2001, 2007). Abadie (2006) argues that individual freedom and political rights, which is heavily associated with the higher levels of democracy allow more terrorist activity and thereby increase the terrorist risk ratings. On the other, studies have shown that democracy decreases

(25)

24 terrorism because it provides different and non-violent channels for conflict resolution (Eyer- man, 1998; Schmid, 1992). Conclusively, while the results remain inconclusive, the level of democracy seem to have a significant impact on domestic terrorism and will thereby be included as a control variable.

Finally, as previously discussed, armed conflict has shown to greatly affect the risk of experiencing domestic terrorism (GTI, 2019, p.2). Civil wars are particularly likely to increase the number of domestic terrorist attacks since terrorism is a cost-effective method to fight and demoralize stronger foes, which makes it an ideal tactic for rebel groups and other violent organizations or ethnic groups to use, particularly in urban areas that are better protected (Merari, 1993). Furthermore, interstate armed conflict typically severely decrease the resources available to address other security concerns, which makes societies engaged in interstate armed conflict vulnerable to terrorist attacks (Lai, 2007). The Global Terrorism Index (GTI) developed by the Institute for Economics and Peace (IEP) has since the projects inception in 2012 deemed armed conflict as the most central driver of terrorism and conflict zones as the most likely places for terrorist attacks to occur which clearly indicates the importance of controlling for the phenomena (GTI, 2019, p.2). As previously discussed, ethnic inequality and armed conflict are also intertwined, which makes the control variable even more important. Armed conflict is measured using a country-year dummy variable based on data from the UCDP/PRIO2 Armed Conflict Dataset (Version 19.1) that is coded as 1 if the government has been engaged in an intrastate or interstate armed conflict and 0 if it has not. To be included in the data, a conflict must have resulted in at least 25 battle-related deaths in one calendar year (Gleditsch et al., 2002; Pettersson et al., 2019).

Conclusively, even after adding these prominent control variables, there is still a considerable risk of omitted variable bias because of the complexity and multidimensional characteristics of domestic terrorism. For instance, ethnic fractionalization, regime durability and geographic area are some other commonly used control variables for measuring effects on domestic terrorism that have not been included in the regression analysis (Piazza, 2011, p.345;

Choi, 2015, pp.45–46; Ezcurra, 2019, p.51)

2 The dataset is produced by The Uppsala Conflict Data Program (UCDP) and the Peace Research Institute Oslo (PRIO) and can be accessed along with other seminal datasets on internal and external armed conflicts at:

https://ucdp.uu.se/downloads/

(26)

25

4. Results and analysis

This section will present and interpret all the results from the linear regression, as well as assess the findings and their implications. First, the dataset that is used for the linear regression will be presented using descriptive statistics of the variables used. Second, the results of the regression models will be presented. Initially, a bivariate model is used to evaluate the relationship between the lagged independent variable and the dependent variable. Following the bivariate model, two multivariate models are presented that includes the lagged control variables. One which looks at the effect of all the control variables on the dependent variable and another which examines the relationship between the primary phenomena of interest while keeping the control variables constant. Third, the results and findings from the linear regressions will be interpreted substantively. Fourth, the causal mechanism presented in the theory section will be scrutinized and assessed based on the findings and what elements that constitute a plausible causal relationship. Finally, the limitations of the research design, as well as avenues for future research, will be discussed.

Table 1: Descriptive Statistics (1970 – 2006)

4.1 Descriptive statistics

To test the hypothesis, an OLS regression is used on the incidence of domestic terrorism utilizing a country-year dataset containing 2837 observations of 107 countries from 1970 to 2006.

To address overlap discrepancies in data availability between variables, all observations with missing data on any of the 7 variables have been removed from the regression to guarantee comparable results and findings.

First, the Independent variable – operationalized as the ethnic group inequality ratio – is based on the relative economic disparities between ethnic groups from the triangulation dataset developed by Cederman et al. (2015). The descriptive statistic indicates that the relatively poorest

(27)

26 ethnic groups measured in the dataset has the same average per capita income as the ethnic group country average when the disparities are the smallest, and is 6 times poorer than the average when they are the largest. Furthermore, the mean of 1,532 indicates that the majority of the observations has a low value within that range – which figure 2 illustrates further.

Second, the dependent variable – operationalized as the country-year count of domestic terrorist attacks – which is based on the dataset developed by Enders et al. (2011) indicates that the maximum number of domestic terrorist attacks per country-year is 524 and that the minimum is 0. The comparatively low mean value indicates that there probably exist high- value outliers among the observations, as well as that there is a trend of lower values and in particular an access number of zero’s – which figure 3, as well as figure 7 (which is the data diagnostics plot located in the appendix) clearly illustrates. This was expected because the dependent variable is an event count variable³ representing a relatively sporadic phenomena.

However, because of the access number of zeros and the dispersed distribution, studying it using ordinary least square estimates can produce biases, be inefficient and inconsistent, which has to be taken into consideration (Ghatak et al., 2019, p.450; Long, 1997).

Figure 2. Histogram Independent Variable Figure 3. Histogram Dependent Variable

3 Count variables are typically individual pieces of count data – which is a type of statistical data where the observations only take non-negative values (e.g. 0, 1, 2, 3, 4). Count outcome variables tend to make the results from OLS regressions biased and unreliable. More information on the topic: https://www.theanalysisfactor.com/regression-models-for-count-data/ (Mulvey and Shaw, 1995)

(28)

27

(1970 – 2006)

4.2 Results

The first model (model 1) presents the results of the bivariate regression between the independ- ent variable (Economic Ethnic Inequality) and the dependent variable (Domestic Terrorism).

The result is statistically significant and indicates a positive relationship between the phenomena. First, the statistical significance is specified by the p-value, which is a measurement of the level of confidence in the systematic correlation between the variables of interest. The p-value shows us the probability that the studied relationship is caused by random chance. Furthermore, the rule of thumb in the social sciences is that all p-values that are smaller than 0.05 are typically considered statistically significant since they roughly represent a 95 per cent confidence level in that the relationship is not caused by chance. In model 1, the p-value is in the p<0.01 bracket

Table 2: The Effects of Ethnic Inequality on Domestic Terrorism

(29)

28 and the confidence level roughly 99 per cent. Second, in line with the theoretical argument, the positive coefficient indicates that countries with higher levels of economic ethnic inequality on average have higher levels of domestic terrorist attacks. Specifically, the coefficient indicates that on average, the number of terrorist attacks increases by 9.386 when the ethnic inequality ratio increases by one unit. This bivariate relationship, the fitted regression line and the disper- sion of observations is clearly visualized in the scatterplot in figure 4. However, since the re-

Figure 4. Scatterplot – Visualizing Model 1 Figure 5. Effect plot – Visualizing Model 3

gression is bivariate, there are no controls in place to isolate the two variables from other confounding variables and alternative explanations that could make the relationship spurious. Con- sequently, two multivariate regressions will be carried out to better understand and isolate the relationship between the two variables of primary concern. In model 2, all the control variables are added without the inclusion of the independent variable to examine their effect on the dependent variable separately. All the control variables had a positive coefficient. However, the statistical significance and the effect size of them varied considerably. The relationship between domestic terrorism and the population size, level of democracy, and armed conflict were all statistically significant with a p-value of p<0.01. The logged GDP per capita had a p-value in