Spatial Distribution of Minority Communities and Terrorism: Domestic Concentration versus Transnational Dispersion

Defence and Peace Economics ISSN: 1024-2694 (Print) 1476-8267 (Online) Journal homepage: http://www.tandfonline.com/loi/gdpe20 Spatial Distribution of Minority Communities and Terrorism: Domestic Concentration versus Transnational Dispersion Bryan J. Arva & James A. Piazza To cite this article: Bryan J. Arva & James A. Piazza (2016) Spatial Distribution of Minority Communities and Terrorism: Domestic Concentration versus Transnational Dispersion, Defence and Peace Economics, 27:1, 1-36, DOI: 10.1080/10242694.2015.1055091 To link to this article: http://dx.doi.org/10.1080/10242694.2015.1055091 Published online: 09 Jul 2015. Submit your article to this journal Article views: 207 View related articles View Crossmark data Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalinformation?journalcode=gdpe20 Download by: [Penn State University] Date: 30 March 2016, At: 12:16

Defence and Peace Economics, 2016 Vol. 27, No. 1, 1 36, http://dx.doi.org/10.1080/10242694.2015.1055091 SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM: DOMESTIC CONCENTRATION VERSUS TRANSNATIONAL DISPERSION BRYAN J. ARVA AND JAMES A. PIAZZA* Department of Political Science, The Pennsylvania State University, University Park, PA, USA (Received 21 July 2014; in final form 9 May 2015) Qualitative studies of terrorist movements frequently highlight the importance of diaspora communities as important factors in producing and sustaining terrorist activity in countries. The underlying theoretical argument is that bifurcation of tight-knit minority communities between countries nurtures separatist or irredentist sentiments among affected community members, thus prompting terrorist activity, while minority community members in other countries might mobilize financial and political resources to support terrorist activity among their compatriots. In this study, we empirically test whether transnational dispersion, versus domestic concentration, of minority communities in countries produces higher incidents of terrorism. Conducting a series of negative binomial estimations on a reshaped database of around 170 countries from 1981 to 2006, derived from the Minorities at Risk database and the Global Terrorism Database, we determine that both transnational dispersion of kin minority communities and domestic concentration of minorities within countries increase terrorism and that transnational dispersion is a particularly robust predictor of terrorist attacks. Keywords: Terrorism; Minority groups; Diasporas; Geographic dispersion; Geographic concentration This study empirically examines a thus far unexamined factor in the production of terrorism 1 : the geographic distribution of ethnic communities within and across countries. It investigates several questions. Are countries containing geographically concentrated minority communities within their borders, whose members suffer from past or present political, economic and cultural discrimination, more likely to experience terrorist attacks? What effect does transnational dispersion of a country s minority groups have on the likelihood that it will experience terrorism? Are countries with transnationally dispersed minority communities where communities within the country have kin in other countries more or less likely to suffer from terrorist attacks at home? These questions motivate the following study. *Corresponding author: Department of Political Science, The Pennsylvania State University, University Park, PA 16802, USA. E-mail: jap45@psu.edu 1 In this study, we adopt the definition of terrorism used by Enders, Sandler, and Gaibulloev (2011), Terrorism is the premeditated use or threat to use violence by individuals or subnational groups against noncombatants in order to obtain a political or social objective through the intimidation of a large audience beyond that of the immediate victims (321). 2015 Taylor & Francis

2 B. J. ARVA AND J. A. PIAZZA Casual observation seems to validate both concentration and transnational dispersion as frequent drivers of terrorism. 2 A long list of countries with geographically concentrated ethnic enclaves have seen major terrorist campaigns emerging from separatist and ethnonationalist terrorist movements since World War II. For example, the liberation of s Basque population, with its distinct language and history of demands for greater cultural autonomy, concentrated in the northeast of the country has been the obsession of the ETA (Basque Homeland and Freedom) terrorist organization. Kurdish enclaves in s southeast and northern Iraq have provided motivation, basing opportunities and various levels of support for terrorist activity by the Kurdish Workers Party (PKK) and Kurdish Democratic Party. The lion s share of contemporary terrorism in Russia is motivated by demands for independence for Chechnya, a Muslim-majority region in southern Russia. Muslim ethnic enclaves in Mindanao in the Philippines and in the far-south Pattani Province of have been the source of Islamist separatist terrorism since the 1970s. In South Asia, the mountainous regions of northeast in the so-called Seven Sister states and the Balochistan province in western have both been the sites of active separatist terrorist movements. has also sustained terrorist activity from the Polisario Front, a radical separatist movement that seeks the independence of Western Sahara from Moroccan rule. At the same time, countries with significant minority communities that have kin in other countries or in regional or global diasporas have experienced notable terrorist campaigns and transnationally dispersed ethnic minority communities often fuel terrorist activity usually by providing political support and financial resources to terrorist groups that are active in their home countries. There are, likewise, many prominent examples of this scenario. The Tamil and Lebanese Shi i diaspora communities in Europe and North and South America have been a significant factor complicating Sri Lankan and Lebanese government efforts to address Liberation Tigers of Tamil Eelam (LTTE) and Hezbollah terrorist activity. Both the LTTE and Hezbollah rely on kin communities outside of their home countries to raise funds and to agitate for political support on their behalf. The Kosovo Liberation Army (KLA) is alleged to have relied upon ethnic Albanians living across the border in Albania and further away in and Italy for monetary support, while the KLA s drive for Kosovar independence was motivated, in part, by cultural identification with neighboring Albania. Finally, terrorism within the Kurdish minority communities in Iran, Iraq, and is motivated both by the concentrations of these communities in the geographic extremes and the political and economic margins of their respective countries as well as by the support given to it by kin communities located across borders. The Kurdish territory in northern Iraq, for example, has provided important basing, training, and revenue-generating space for PKK terrorists within. In this study, we empirically evaluate the proposition that countries characterized by geographically concentrated ethnic enclaves experience more terrorist activity than countries with more internally dispersed and geographically integrated ethnic populations. We also examine whether or not countries with ethnic groups that have kin in other countries experience more terrorism. The study contributes to scholarly understanding about the impact of ethnic group geographic concentration and transnational dispersion on terrorist activity by conducting a systematic, empirical cross-country analysis on a topic that has previously been treated using individual case studies and descriptive analysis. In the next section, we explore the relevant literature linking geographic distribution of aggrieved minority 2 Historically, terrorist attacks by groups motivated by ethnic nationalist/separatist and leftist ideologies account for almost 70% of terrorist incidents worldwide (see Piazza 2009, 63).

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 3 communities to political violence more generally. From the relatively scant literature on the impact of domestic concentration and transnational dispersion on political violence, we discuss the hypotheses we test in the study. The following section explains our empirical tests and presents the results. We conclude with the implications of the findings to scholarship on terrorism and to security policy. COMMUNITY CONCENTRATION, TRANSNATIONAL DISPERSION, AND TERRORISM The impetus for our study rests on empirical findings in a slew of recent studies revealing that the presence of aggrieved minority populations within countries makes them significantly more likely to experience terrorism. Theoretical work by Gurr (1993, 1996, 2000) argues that ethnic minority group experiences of discrimination and alienation from mainstream political, economic, and cultural life fosters strong group identities and gives rise to grievances. In instances where political elites can manipulate those grievances and overcome collective action barriers (DeNardo 1985; Hegre et al. 2001) political violence within countries will occur. Crenshaw (1981) and Ross (1993) use Gurr s essential model to depict ethnic minority group grievance as a crucial contributor to terrorist activity within countries. Minority Grievance and Terrorism Recent empirical work seems to validate this theoretical model. Piazza (2011) finds that countries with minority communities that experience economic discrimination experience, on average, six times the level of domestic terrorist attacks than the typical country, eight times the level of terrorism than countries containing minority groups that experience no present economic discrimination, and ten times the number of attacks than countries lacking any minority communities at all (Piazza 2011, 350). Further research by Piazza (2012), building off of a rich body of qualitative literature on ethnic-based terrorist movements, finds generally that the presence of any aggrieved minority community within a country makes it more likely that that country will experience both domestic and transnational terrorism and that countries featuring higher levels of economic and political discrimination, as opposed to forms of cultural, religious, and linguistic discrimination, are particularly likely to suffer from high terrorism rates. Related studies have produced corresponding findings. These results are consistent with control findings by Lai (2007), Wade and Reiter (2007) and Eubank and Weinberg (1994) who also find empirical evidence that poor treatment of minority groups within countries increases terrorism. Minority Communities, Geography, and Terrorism So, we begin our investigation expecting the presence of minority groups, with grievances, to be a significant predictor of patterns of terrorist attacks in countries. But how might this dynamic be affected by population geography? While there have been many qualitative case studies on the topic, little has been done quantitatively to assess the importance of diaspora and transnationally dispersed ethnic or religious minority communities as important factors in producing and sustaining terrorist activity in countries. We find this to be a curious omission, given the prominent attention paid to ethnic group geographies in the civil war literature. Many studies of internal and transnational geographic distribution of ethnic groups

4 B. J. ARVA AND J. A. PIAZZA have found such features to be important mediators for civil war onset, duration, and intensity. (Cederman, Buhaug, and Rod 2009; Collier and Hoeffler 2001; Cunningham and Weidmann 2011; Davis and Moore 1997; Forsberg 2008; Saideman 2002; Slack and Doyon 2001; Toft 2003; Weidmann, Rod, and Cederman 2010). Though terrorism differs from civil war as a phenomenon 3 the former is a tactical behavior engaged in by weak actors operating in small groups or as individuals that is focused on grabbing attention and communicating a political message or prompting government concessions rather than securing a more traditional military victory it is reasonable for scholars of terrorism to look to the civil war literature for insight into how ethnic political geography might affect terrorist activity. Findley and Young (2010) point out that from 1970 to 2004 between 44 and 63% of all terrorist attacks occurred within civil conflict zones. Additionally, 73% of all civil wars featured at least one terrorist event: This demonstrates how closely these two phenomena are related. Thus, it is our assumption that the relationships witnessed among diasporas and transnationally dispersed ethnic or religious minority communities and the likelihood of conflict should be similar to the relationships they have in regard to terrorism. In many ways, this study is breaking new ground for terrorism research. However, by looking at scholarly work done in other fields, we can feel confident that our theory and hypotheses have solid foundations. There seems to be agreement that conflict is more likely when ethnic groups have a majority of their members living in a single location within a country. This is perhaps most dramatically illustrated in the findings of Cunningham and Weidmann (2011) who determine that violence is increased when an administrative region of a country or geographic enclave is dominated by a particular ethnic group that is in competition with other ethnicities nationally. Conversely, the more dispersed an ethnic group is the less likely they are to become involved in a conflict. The reasoning behind this is the more members an ethnic group has in an area, the easier it is to recruit people to join their cause. Dispersion of an ethnic group dilutes group identity and raises organization costs, making it more difficult for extremist and terrorist organizations to garner support. Concentration also might raise policing and intelligence costs for states. Often concentrated pockets of ethnic communities are quite insular, have high levels of interpersonal trust and community self-reliance, may contain larger numbers of residents who speak local languages and communicate only with group members, and are resistant to outside interference. 4 All of these would make counterinsurgency and counterterrorism activities more difficult and terrorist activity more likely. In addition to facilitating social mobilization that might boost the possibility of terrorist activity, and imposing policing and intelligence costs on states, concentration of ethnic minority groups into a particular region within a country might also reduce its sense of vulnerability. Scholars examining related forms of political violence, such as deadly ethnic riots, have found ethnic group geographic concentration to increase group feelings of invulnerability, which facilitates outbursts of violence against individuals from rival ethnic groups (see Horowitz 2003). In contrast, when ethnic groups are intermixed with other groups they are more inhibited and less likely to engage in political violence. A similar phenomenon could affect patterns of terrorism in countries. Terrorist activity may be more likely in instances where ethnic groups are concentrated, uninhibited, and feeling invulnerable vis-à-vis rivals. 3 Note, in all model estimations, we include a control for civil war using PRIO data. 4 Though in principle, terrorists from an ethnic minority community concentrated into a particular region or enclave of a country might also find it more difficult to conduct activities outside of that enclave, thereby narrowing the geographic range within which state policing assets need to be allocated. This could lower counterterrorism policing costs overall for the country.

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 5 We furthermore consider the possibility that heightened minority group identity is likely to also produce more mainstream political vehicles, such as ethnic political parties, pressure groups and social movements, for articulating group grievances. While some would suggest that when this occurs within the context of opportunities for political participation, it should have a dampening effect on terrorism and political violence because more lucrative routes of political influence are available empirical research by Chenoweth (2010) demonstrates the opposite. She finds that when political competition and interest group activity increases in democracies, the number of emergent terrorist groups increases and terrorist attacks go up. If geographic concentration of minority groups helps to stimulate and foster minority group engagement in the political process, then we should observe increased terrorist activity as well. These arguments lead us to our first hypothesis: H1. Countries containing geographically concentrated significant minority communities are more likely to experience terrorist attacks. The location and density of minority groups within countries is not the only factor that could potentially affect the likelihood of violent conflict. We also expect that transnational dispersion of minority communities has an important impact on patterns of political violence in countries. We are aided in this expectation by the civil war literature. Though the effects of transnational minority community kin networks on terrorist attacks have not been empirically evaluated, scholars have found them to be important factors in fostering civil war onset and continuance (Moore and Davis 1998; Saideman 2001; Woodwell 2004). Seeking to explain why civil wars tend to geographically cluster and how internal armed conflicts in one country are apt to prompt similar conflicts in neighbors the contagion effect Buhaug and Gleditsch (2008: 215) find that transnational ethnic linkages are central mechanism of conflict contagion. We argue that there are several ways in which transnational minority communities might buttress terrorist activity. First, they form crucial networks of support for terrorist movements that may help to boost terrorist activity. A wide range of scholars observe that terrorist movements assiduously rely upon monetary and non-monetary support from kindred diaspora communities (Adamson and Demetriou 2007; Carter 2005; Clark 2007; Esman 2009; Hess 2007; Levitt 2007; Wayland 2004). Diaspora communities serve as important sources of financial support for terrorist movements by providing donations and holding fundraisers. Perhaps even more importantly, kin communities outside of the home country can help terrorist movements safely or clandestinely bank their assets, launder revenues, and can help transfer money internationally. In terms of non-monetary support, ethnic terrorist movements find kindred diaspora communities to be useful for acquiring weapons and equipment or by providing safe houses or assistance in obtaining legal or illegal entry into foreign countries. Moreover, kin groups can serve as lobbying or pressure-group agents on behalf of terrorist groups, or can be useful for launching public relations campaigns to favorably portray terrorist group interests. In some instances, terrorists may use kin groups in other countries as sources of information and intelligence. Finally, Clark (2007) notes that diasporas are fertile recruiting grounds for terrorists. Examples of terrorist movements that rely heavily on these types of support from their ethnic kin in other countries abound: the LTTE, the Irish Republican Army, Hezbollah, the Mujahideen-e Khalq, various anti- Castro movements, Hamas, Fatah, and Al Qaeda. Many experts maintain that these groups would simply not have been able to function as effectively, or carry out the same level of

6 B. J. ARVA AND J. A. PIAZZA attacks, if not for the support from their kindred diaspora communities throughout the world (Clark 2007; Hess 2007; Levitt 2007; Wayland 2004). Second, in addition to receiving direct support from kin in other countries, terrorist movements embedded in aggrieved minority groups in countries can rely upon kin to help secure state assistance for their cause. Saideman (2002) convincingly explains the reasons why minority groups in some countries may be more or less likely to receive external assistance, which then might translate into the onset or widening of armed conflict in their home country. The motivation for states to provide support to ethnic insurgencies and terrorist movements in other countries which can take the forms of monetary resources, arms, training, safe haven or political and diplomatic support is essentially modulated by domestic politics within the supporting state. The consensus in the field is that all regimes need to maintain a certain level of popular support and one of the ways that states bolster their legitimacy is to keep key domestic (majority or minority) ethnic communities content. In addition to traditional patronage techniques, regimes may opt to please domestic constituents by pledging to support the interests of their ethnic kin in other countries (Bueno de Mesquita and Lalman 1992; Fearon 1994; Morgan and Bickers 1992; Saideman 2002). Though there is always the possibility of undesirable costs with this strategy in the form of blowback, regional instability, or international recrimination regimes may relatively cheaply win the support of strategic domestic constituencies by playing upon their national pride and by constructing narratives of outrage and sympathy for the treatment of ethnic minorities in other countries. Also, leaders may use pledges of support for foreign kin of domestic ethnic communities as a means of mobilizing support for foreign policy objectives vis-à-vis other states. Davis and Moore (1997) illustrate the wide appeal of this strategy on the part of states, explaining that it applies to both democracies and to authoritarian regimes. Dictatorships may also try to bolster support for themselves among domestic ethnic groups by pledging to aid kin in other countries. We expect to uncover a similar process for state support of terrorist activity. Terrorist movements based in aggrieved minority countries are more likely to acquire direct financial, political or military support, or to receive what Byman (2005) terms passive support, from foreign states that are strongly influenced or dominated by their ethnic kin. We can point to several examples of this. The Greek Cypriot independence movement EOKA (National Organization of Cypriot Fighters) that engaged in terrorism against British and Turkish targets in the 1950s derived active and passive support from the Greek government, while their counterpart the Turkish Resistance Organization was more overtly supported by Ankara. Throughout the 1980s, Saddam Hussein s Arab nationalist regime in Iraq provided weapons to Arab extremists based in the Iranian Khuzestan province, while starting in 1979 and continuing to this day, Iran provides support to Shi i extremist groups, such as the Mahdi Army, in Iraq. All factions of the Palestine Liberation Organization, as well as Hamas, receive financial support from various Arab regimes, while the i government provides aid to Muslim Kashmiri terrorist movements that operate within. Third, Salehyan and Gleditsch (2006) demonstrate that refugee inflows often stimulate civil wars and internal armed conflicts by providing an opportunity for rebel groups to construct and exploit cross-border networks. Similar processes could enhance terrorist activity. Countries that share ethnic minority groups are more likely to have higher levels of transborder migration involving members of these communities. These migrations raise security and intelligence costs for the counterterrorism officials in countries experiencing these flows, while terrorist movements embedded in the transnationally dispersed communities can exploit the logistical and other advantages the flows provide.

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 7 Finally, in addition to these direct material effects financial and political support, recruiting opportunities, logistical benefits, etc. transnationally dispersed minority communities also raise the possibility that terrorist activity in one country might be stimulated by public awareness of political turmoil and violence occurring in another country via demonstration effects. Beissinger (2002) and Kuran (1998) have determined that the demonstration of internal armed conflict by actors in one country often prompts political activists and dissidents in neighboring countries to engage in similar activities against their own governments. We suggest that demonstration effects should be particularly strong when witnessed by co-ethnics in countries and likely have the same stimulating effects for lower intensity types of political violence like terrorism. Given all of these theoretical pathways, we test a second hypothesis: H2. Countries containing significant minority communities with kin in other countries are more likely to experience terrorist attacks. ANALYSIS To test the above hypotheses, the study conducts a set of negative binomial regression analyses on a dataset, compiled and shaped by the authors, of the annual count of terrorist incidents occurring within 165 or 170, depending on the specification, countries for the period 1981 2006. The main dependent variable for the study is an annual count of the number of all domestic terrorist attacks occurring within a country-year, though we conduct robustness estimations on counts of total domestic and transnational attacks in a country per observation. The source for the dependent variable is the Enders, Sandler, and Gaibulloev (2011) decomposed database derived from the Global Terrorism Database (GTD), a publicly available database housed at the University of Maryland s Center for the Study of Terrorism and Responses to Terrorism. We opt to use the Enders, Sandler, and Gaibulloev (2011) domestic terrorism measure for the main dependent variable because we expect that the impact of geographic patterns of minority community distribution within a country for example, concentration in one region or enclave of the country is most likely to manifest itself in local terrorism rather than in transnational terrorism. Moreover, the motivation or ability to conduct domestic terrorist attacks by a minority community might be enhanced by financial or other support by kin in the Diaspora, making local attacks more likely. However, we acknowledge that at times, domestic terrorist movements might opt to select an international target within their home country, for example, a foreign embassy, foreign corporation, or group of foreign tourists, in order to garner more international media attention, or ethnic groups with kin in other countries might attack from safe havens abroad. We therefore conduct robustness models using the total count of all terrorist attacks domestic and transnational attacks within a country provided by GTD. The results of these estimations produce the same results as those in the main models, so we are confident that our findings are not dependent on how terrorism is operationalized. 5 Some, though not all, of our theoretical argumentation about how patterns of domestic concentration and transnational dispersion of ethnic minority groups prompts terrorism focuses on ethnic group-level processes. To test the hypotheses derived from these 5 Results in Appendix 1, Table A and Table C.

8 B. J. ARVA AND J. A. PIAZZA processes, one might argue that it would be more theoretically consistent to use an ethnicgroup-year unit of analysis for the study rather than country-years. However, we opt to use an aggregated, country-year count of attacks for the study, and this decision is informed by the quality of data on individual terrorist attacks available to researchers. The source of data we use for the study is based upon the GTD, an event database that includes information on characteristics of terrorist attacks. The GTD does include a variable for the perpetrating group name which in theory could be used to allocate counts of attacks to ethnic minority groups on a yearly basis. However, data on terrorist attack perpetrators are missing for around half of all observations in the GTD data-set. Patterns of missingness for perpetrator identification in GTD vary widely by country and world region for some countries in Asia, for example, information on perpetrating group is absent in 80% of the observations and data are not missing at random (Arva and Beieler 2014). Moreover, the perpetrator failure to claim attacks can be a component of individual group strategy, raising a further complication. Any attempt to construct a group-level database of terrorist events using GTD necessarily injects significant bias into the analysis, eroding confidence in the results presented. In contrast to the name of the perpetrator, every incident in GTD is reliably identified with a country location of the event, allowing researchers to construct a reliable count of attacks that occur by country per year, eschewing selection issues. Though an aggregated country-year analysis comes with its own limitations, a country-year framework minimizes selection problems and enables us to identify patterns of ethnic group spatial distribution as a structural feature that may help to explain why certain countries experience more terrorism than others. Estimation Technique In terms of the modeling technique, the study opts to use a negative binomial estimation technique due to significant overdispersion in the distribution of the dependent variable: It is a count variable where nearly 56% of the observations are zero values; around 17% of the observations are counts of ten terrorist attacks or more; almost 5% of the observations are counts of 100 or more terrorist incidents; and no observation can, of course, have negative values. This makes a negative binomial estimation technique more efficient than a standard, ordinary least squares technique (see Brandt et al. 2000; Cameron and Trivedi 1998; King 1988). To correct for spatial clustering in the dependent variable, we also calculate robust standard errors clustered on country and set the variance dispersion at constant. As a check, we also ran the models using several alternate techniques, including zero-inflated negative binomial estimations, country-fixed effects negative binomial estimations and rare events logistical regression estimations. These reproduce the same overall findings as those in the same model, though each suffers from limitations that make the negative binomial estimation technique the most advantageous in the study. 6 6 Results in Appendix 1, Table F. Note, though Vuong tests indicate that use of a zero-inflated estimation is efficient, we crucially lack a theoretical reason for using such a technique as we do not theorize that separate processes determine whether or not a country will experience a zero count of terrorism (captured by the inflated logit model in the zero-inflated equation) versus the number of terrorist attacks above 1 a country will experience (captured by the negative binomial model in the zero-inflated equation). Also, because the dependent variable does not temporally vary within some country cases it stays at zero this complicates use of a country-fixed effect technique because such observations are automatically dumped. These limitations support the practice of using the zero-inflated and country-fixed effects techniques for robustness checks only.

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 9 Independent Variables The main independent variables used in the study are derived from the Minorities at Risk (MAR) database, which compiles indicators on the political, economic, social, and security statuses of 282 significant meaning sizeable minority communities within countries, globally 7 (Minorities at Risk Project 2009). The project defines a minority at risk group as communities that, collectively suffer or benefit from systematic discriminatory treatment vis-à-vis other groups in society (Minorities at Risk Project 2009, 1). Around 68% of the countries in the sample contain at least one minority at risk group, while about 47% of countries are home to more than one minority at risk group. This actually poses a dilemma given that the MAR database is structured as a group-year database while all of the other indicators in the study are country-year in format. We have, therefore, reshaped the MAR data into a country-year format by truncating the MAR variables into singular measurements for each observation. We have several reasons for doing this. First, our dependent variable is best measured at the national level and cannot be reliably coded for specific MAR groups, as explained above. Second, data for our controls which have been selected based upon their inclusion in previous empirical studies are aggregated to the national level and are not available at the group level. Third, we wish to adhere to the convention established by Piazza (2011, 2012), Lai (2007) and Caprioli and Trumbore (2003) who code aggregate country-level measures of MAR group features for countries containing more than one MAR group by adopting the highest value exhibited by any one MAR group. Finally, using a country-level analysis, we can capture both ethnic minority group perpetrated terrorist events along with majority group perpetrated attacks, which may be prompted by resentments associated with MAR group features (see Piazza 2012). This has the added advantage of introducing a conservative bias into our analysis, making us more confident of the results. So, for the purposes of this study, our observations for MAR variables are the highest ordinal value for MAR group concentration or transnational dispersion, rather than the average or median. 8 We operationalize ethnic group spatial distribution using two variables we develop from MAR. The first is a recoded version of the MAR variable GROUPCON Group Spatial Distribution, or the degree to which a minority at risk group is geographically concentrated or dispersed in a country. In its original form, this ordinal variable takes the following values: 0, indicating that the group is widely dispersed throughout the country; 1, indicating that the group is either concentrated in urban areas only or is concentrated as a minority population in specific regions in the country; 2, indicating that the group is concentrated into a specific region as a majority population in that region, but that the bulk of its community is dispersed as a minority in other regions; 3, indicating that the group is concentrated as a majority in a region with little presence dispersed in other regions of the country (Minorities at Risk Project 2009, 7). We recode it by adding a new ordinal level for the absence of any MAR communities in the country, changing it from a three to a four point ordinal scale. This transformation is depicted in Table I. 7 The MAR database is the only existing source for the independent variables we use in the study: ethnic group concentration and transnational dispersion. The Ethnic Power Relations database, another frequently used data source for scholars conducting cross-sectional time series research on ethnic minority groups in countries, does not code variables measuring geographic patterns of ethnic group residence. 8 An added barrier to creating and analyzing minority-group-year units in addition to missing data and the attendant selection bias problems this presents is that the MAR database codes data on minority groups that are of a minimum size at least 1% of the national population and omits smaller groups. This produces selection bias issues that recommend that we aggregate MAR features like spatial distribution to the national level, again following the protocol established by Piazza, Lai, and others.

10 B. J. ARVA AND J. A. PIAZZA TABLE I Transformation of MAR Group Concentration Indicator Original MAR Indicator Recoded MAR Indicator Group Spatial Distribution (GROUPCON) Transnational Dispersion of Kindred Groups = No MAR Group Present = 0 = 0 Group is widely dispersed = 1 = 1 Group is primarily urban or is a minority in one region = 2 = 2 Group is a majority in one region but dispersed in others = 3 = 3 Group is concentrated in only one region = 4 We transform more significantly the other MAR indicator for our other independent variable. The transformation process is outlined in Table II. To measure the effects of transnational dispersion as opposed to domestic concentration captured in the MAR Group Concentration variable previously described we create an ordinal scale indicator of MAR group dispersion across national borders. This indicator, which we name MAR Group Transnational Dispersed, is a three-point scale whereby 0 indicates that the country lacks any MAR groups, 1 indicates that the country s minority at risk groups do not have kin across borders i.e. they are unique to the country and 2 indicates that the country s minority at risk groups do have kin, either close by right across the border or further away in non-adjacent countries. We derive this indicator by transforming the original MAR indicator GC10, Transnational Dispersion Kindred Groups. However, because of the objective of the study and the limited nature of our hypotheses, we do not make use of the nuanced codes in GC10 listed in Table I and instead truncate them to ease interpretation. These two ordinal measures of MAR group spatial distribution are the main independent variables we use to assess the effects that domestic concentration and transnational dispersion have on terrorism. 9 However, to further validate our findings we also use dummy variables coded 1 for countries containing MAR groups that are not domestically concentrated or that are not transnationally dispersed. We, of course, expect these to be either not significant or negative predictors of terrorism. We furthermore break down into separate dummies the original ordinal values of the MAR concentration and dispersion indicators and regress them against counts of terrorism. Our expectation for these is that the dummies for the higher codes indicating higher degrees of concentration or dispersion will be significant predictors of terrorism. The study also includes in all models a set of controls that previous studies have found to be robust predictors of terrorism (see for example Eyerman 1998; Krieger and Meierrieks 2011; Li 2005; Wade and Reiter 2007). We hold constant some basic indicators that may affect patterns of terrorism including logged measures of gross national income per capita, national population, and surface area. We expect larger and more populous countries to experience more terrorism because they have higher policing costs (Eyerman 1998). We also control for level of economic development in a country, but have mixed assumptions reflecting the divisions in the field about poverty and terrorism: Gross national income may be a positive predictor of terrorism (Piazza 2011); a negative predictor (Freytag et al., 9 Note, as a robustness check, we also created dummy versions of our ordinal independent variables measuring MAR group concentration and dispersion. These were simply coded 1 for observations in which our ordinal scale MAR indicators are 1 and higher. These provide a check of our ordinal transformation and reproduce the same results as our main findings. These results can be found in Appendix 1, Table B. (Note also that running the original MAR indicators produces the same core results as well.)

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 11 TABLE II Transformation of MAR Dispersion Indicators Original MAR Indicator Recoded MAR Indicator Transnational Dispersion Kindred Groups (GC 10) Transnational Dispersion of Kindred Groups = No MAR Group Present = 0 = 1 Group has no close kindred across an international border = 1 = 2 Group has close kindred across a border which does not adjoin its =2 regional base = 3 Group has no close kindred in countries which adjoin its regional base = 4 Group has close kindred in one country which adjoins its regional base = 5 Group has close kindred in more than one country which adjoins its regional base 2011); or may have no effect at all (Krueger and Malečková 2003). Li (2005) found income inequality, political participation, and degree of constraints placed upon the executive branch of countries to be significant predictors of terrorist attacks. 10 We therefore include in all models the Gini coefficient, an index measuring income inequality, level of political participation, from the Vanhanen Index, and a measure of executive constraints from Polity. Because Eyerman (1998) determined that older, established (democratic) regimes were less likely to experience terrorism, we also control for the number of years the current regime has been in place. We use the durable indicator from Polity to control for age of current regime. Li also controlled in his study for pre- and post-cold War temporal effects, under the assumption that state support of terrorism during the Cold War period boosted terrorist activity. We do the same, including a dummy variable for all Cold War years, 1945 1991. Our expectations are, given findings of other scholars, that Gini, executive constraints countries with constrained executives have a limited ability to get tough on terrorists and the Cold War dummy variable are positive predictors of terrorism while political participation dampens terrorist attacks. Walsh and Piazza (2010) determined that states that protect the physical integrity rights of their citizens rights against torture, political imprisonment, disappearances, and extrajudicial killings experience significantly less terrorism. We therefore include the indicator that they used the Physical Integrity Index from CIRI as a control in all of our models. Also, in line with Li (2005), we insert a one-period lag of the dependent variable in all models to account for previous terrorist attacks. Crucially, because there is a literature demonstrating the important contributions made by ethnic group diasporas to the onset and duration of civil wars 11 mostly by supplying resources and lending material support to rebel groups (see, for example, Collier 2003; Collier and Hoeffler 2004; Collier, Hoeffler, and Söderbom 2004; Fearon and Laitin 2003) and because terrorist activity may rise during periods of civil wars or within civil war zones in countries (Findley and Young 2010), we control for the presence of an active civil 10 Though, interestingly, research by Østby (2008) finds that general levels of economic inequality in countries are not a predictor of violent conflict in countries, but horizontal inequalities ethnic or other social-group-based inequalities rather than inequalities between individuals do drive violence. This dovetails with the study s focus on ethnic minority communities as an important factor in episodes of violent conflict such as terrorism. 11 We actually produce a similar, though tentative, finding to this using our own data. When run separately in models, MAR group transnational dispersion predicts civil wars in our sample of countries. However, when included in the same model with group concentration, this relationship disappears.

12 B. J. ARVA AND J. A. PIAZZA TABLE III Descriptive Statistics Variable Obs. Mean Std. deviation Min Max Terrorist Incidents (GTD Domestic, Enders, Sandler, 5893 7.4 32.9 0 524 and Gaibulloev 2011) MAR Group Transnationally Dispersed 4107 1.1.9 0 2 MAR Group Concentration 4785 2.3 1.5 0 4 MAR Group No Trans. Dispersion 5808.01.07 0 1 MAR Group No Concentration 5808.04.19 0 1 MAR GROUPCON = 0 4269.39.48 0 1 MAR GROUPCON = 1 4269.07.25 0 1 MAR GROUPCON = 2 4269.07.26 0 1 MAR GROUPCON = 3 4269.45.49 0 1 MAR GC10 (Dispersion) = 0 3785.35.47 0 1 MAR GC10 (Dispersion) = 1 3785.08.27 0 1 MAR GC10 (Dispersion) = 2 3785.14.35 0 1 MAR GC10 (Dispersion) = 3 3785.03.19 0 1 MAR GC10 (Dispersion) = 4 3785.15.35 0 1 MAR GC10 (Dispersion) = 5 3785.23.42 0 1 Log Gross National Income per cap 5629 7.1 1.6 3.5 14.1 Log Population 5800 1.8 1.7 2.8 7.1 Log Area 6364 11.8 2.1 5.7 16.6 GINI Coefficient 5813 42.8 8.9 17.8 84.8 Durable 5808 21.9 27.4 0 197 Political Participation (Vanhannen) 5719 29.8 22.0 0 71 Executive Constraints (Polity) 5813 3.4 1.5 0 6 Civil War (PRIO) Physical Integrity Index (CIRI) 3837 4.8 2.3 0 8 Cold War Era 6364.5.4 0 1 war in all models. To do this, we use a dummy variable coded 1 for observations in which a civil war is occurring, using data from PRIO. To provide a further check for spuriousness, we also ran two other analyses. We conducted robustness models that control for the incidence and magnitude of civil wars, civil violence and ethnic conflict and war within countries using data from the Major Episodes of Political Violence database and evaluate. We also reran the main models dropping all cases experiencing a civil war. All of these models produce the same core findings as those presented in the main results table of the study. 12 Table III summarizes all of the variables in the study by reporting their descriptive statistics. DIRECTION OF CAUSATION All independent variables in the models are lagged one temporal period. This helps to capture lagged effects of change in them on the dependent variable and also helps to provide some clarity about direction of causation. We note that running the models using current, rather than lagged, versions of the independent variables produces no change in the main 12 Results in Appendix 1, Table G.

SPATIAL DISTRIBUTION OF MINORITY COMMUNITIES AND TERRORISM 13 results. This is an empirical validation of our theoretical story that geographic distribution of minority communities within and among countries is in most cases a product of longterm and dramatic political, social, and economic processes, whereas terrorist activity is a more short-term, sporadic, and low-intensity phenomenon. Some evidence from the data helps to bear this out. In the overwhelming majority of countries in our sample, geographic living patterns of minority communities were established prior to the time period we examine and are remarkably more static than patterns of terrorist attacks. For example, nearly every country in the sample experiences fluctuation in terrorist attacks, some dramatically: 72% of countries saw fluctuations of more than 20 more attacks during the 37 years of observations and the average standard deviation of attacks for countries over the time series was 18.03. However, only 60 of 172 countries saw their level of minority group concentration change at all from 1970 to 2006. Change in transnational distribution of minority communities is even rarer in the time series. Only 9% of countries experienced a change in transnational dispersion of their minority groups. We are therefore dubious at the outset about the likelihood of a reversed relationship, that terrorist attacks in t 1 might cause an ethnic group to cluster into one region of a country or to become transnationally dispersed or to see their kin come into or leave power in other countries in t 2 during the time period examined. However, to provide a further check, we also conducted some Granger causality tests using one, two, and three-year lags of both the independent and dependent variables. These reveal that while temporally lagged measures of terrorist attacks and MAR group concentration and dispersion Granger-cause terrorism, lags of terrorism do not Granger-cause the MAR group indicators. 13 We furthermore ran sets of estimations switching the dependent and independent variables. In none of these models are counts of terrorism predictive of MAR group concentration or dispersion; indeed, both concentration and dispersion have different country-level significant predictors than terrorism in these models. 14 This gives us further confidence that reverse causation does not complicate interpretations of the models. Results The results of the analyses are summarized in Tables IV and V and provide consistent support for the hypotheses: Countries containing ethnic minority groups that are geographically concentrated into one part of the country and ethnic groups with kin in other countries are more likely to experience terrorism. Model 1 finds countries with more geographically concentrated minority at risk groups to be more likely to experience terrorist attacks while Model 2 finds MAR group transnational dispersion to positively predict terrorism. These findings provide preliminary support for Hypotheses 1 and 2 and are robust to the inclusion of strong covariates. Models 3 and 4 solidify this support by including an indicator coded 1 for the presence of a minority ethnic group that is either not geographically concentrated model 3 or transnationally dispersed model 4. In model 3, countries with geographically concentrated ethnic minority groups experience more terrorism, while countries merely possessing an ethnic minority group that is not concentrated, that is, internally dispersed or integrated within the majority ethnic population, do not. In model 4, countries with ethnic minority groups that have kin in other countries experience more terrorism, but actually experience less terrorism if they possess an ethnic minority community that does not. 13 Results available from authors. 14 Results available from authors.

14 B. J. ARVA AND J. A. PIAZZA TABLE IV Main Results, Domestic Concentration and Transnational Dispersion of Minority Groups and Domestic Terrorist Attacks Model 1 Model 2 Model 3 Model 4 MAR Group Concentration t 1.157***.160*** (.040) (.042) MAR Group Transnationally Dispersed t 1.351***.350*** (.081) (.081) MAR Group, Not Concentrated t 1.101 (.212) MAR Group, Not Dispersed t 1 11.461*** (1.019) (ln)gross Nat. Income t 1.142**.113*.141**.112* (.044) (.047) (.045) (.047) (ln)population t 1.350***.329***.348***.329*** (.043) (.046) (.043) (.046) (ln)geog. Area t 1.147***.115**.145**.114** (.041) (.043) (.042) (.043) Gini Coefficient t 1.011*.012*.011*.012* (.006) (.006) (.006) (.006) Durable t 1.003.003.003.003 (.002) (.002) (.006) (.002) Executive Constraints (Polity) t 1.251***.305***.253***.305*** (.061) (.060) (.061) (.060) Political Participation (Van.) t 1.003.002.003.002 (.003) (.003) (.003) (.003) Civil War (PRIO) t 1.925***.813***.923***.811*** (.152) (.160) (.152) (.160) Physical Integrity (CIRI) t 1.123***.127***.124***.127*** (.030) (.031) (.030) (.031) Cold War Era t 1.501***.414***.503***.413*** (.092) (.092) (.092) (.092) Terrorist Attacks t 1.005***.005***.005***.005 (.000) (.000) (.000) (.000) Constant.083.394.119.384 (.602) (.663) (.611) (.663) Wald Chi 2 1046.54*** 853.98*** 1051.51*** 1002.66*** No. of Countries 170 165 170 165 Observations 3285 2730 3285 2730 Notes: All models are negative binomial estimations. Robust standard errors, clustered by country, in parentheses. Dispersion set at constant. *p <.05,**p <.01, ***p <.000, one-tailed tests. Table V disaggregates the MAR concentration and dispersion variables into a set of dummies measuring the level of concentration and dispersion of minority groups in countries separately. In all of these models, the reference category is the zero value for concentration or dispersion, indicating that the country contains a minority group that is either not geographically concentrated or dispersed.