Analysis of the impact of political isolation of ethnic groups on the growth of terrorism

Novikov Andrey Vadimovich Assistant, Plekhanov Russian University of Economics 117997, Russia, Moskva, g. Moscow, per. Stremyannyi, 36, kab. 339 Camouflage@yandex.ru Other publications by this author

Introduction

Ethnicity can be a powerful motivator for political violence. While peaceful relations between ethnic groups are the norm, when conflicts arise, the warring parties often unite along ethnic lines. Previous estimates based on PRIO data on deaths in military conflicts are that 64% of all civil wars are based on ethnicity ^{[32, p. 194]}. Similarly, M.D. Toft claims that "almost two thirds of all armed conflicts involve an ethnic component" ^{[54, p. 3]}. Paying special attention to terrorism, more than a third of terrorist organizations seek to promote the interests of ethnic groups ^{[21, p. 250]}. There is also some evidence that ethnic organizations may have advantages in recruitment, resource support, goal setting and operational strategies compared to other types of terrorist groups ^[8]. Therefore, it is not surprising that ethno-national and ethno-religious groups may be responsible for some of the deadliest terrorist attacks ^[40].

Therefore, it is worth studying the specific impact of ethnic political isolation on terrorist violence. While a number of studies reveal a link between ethnic political isolation and terrorism at the state level, it is necessary to investigate the differences in individual territories of the country ^[28]. Such a local focus will make it possible to better capture the dynamics of terrorist violence. Since terrorism itself is extremely localized, grouped in certain national regions ^[45]. Terrorism often develops in geographically large countries with significant regional differences in terms of ethnic composition, economic development and local politics. This pattern can be seen on the example of 10 countries with the largest number of terrorist attacks in the presented data set (Colombia, Afghanistan, Philippines, Pakistan, Algeria, Syria, Nigeria, Somalia, India, Iraq, and Russia on the 11th place in the list) ^[44]. This means that the aggregation of the analyzed data at the national level loses an important and theoretically interesting local variance. Finally, terrorism is often supported by marginal actors, many of whom do not have political representation or the ability to influence policy at the national level ^[49]. This parallel with civil war studies demonstrates the advantage of analyzing local conditions, especially since it allows for a better comparison of theory and empirical data ^[23].

The analysis of local dynamics has an additional advantage. Allowing us to explore how ethnic political isolation can actually increase the influence of other factors contributing to terrorist violence. In the more extensive literature on conflicts, it has been established that poverty, inequality and competition for resources become much more relevant when they are intensified by the ethnic division of power ^{[17, 24]}. Therefore, it is necessary to analyze whether ethnic exclusion has a similar stimulating effect on the local dynamics of terrorism.

Connection of ethnicity, political exclusion and terrorism

Terrorism refers to the use or threat of violence by non-State actors to incite fear among the population in order to achieve political or social goals ^{[33, p. 4]}. While some studies define ethnicity as one of the many factors predicting the risk of terrorist violence, the link between ethnic diversity and terrorism in general has weak empirical evidence. For example, D. Piazza finds no evidence that the level of terrorism is higher in countries with a large ethnic diversity ^[49]. Despite the fact that L. Heger even claims that higher ethnic factionalism actually reduces the likelihood of terrorism ^[37]. Similarly, A. Basuchudhari and V. Shugart believe that perceived ethnic tension is less important for predicting terrorism than the country's wealth or economic freedom ^{[18, p. 66]}. Leading these authors to claim that "it might be wrong to think that terrorism is rooted in ethnic differences."

At the same time, in previous works on civil conflicts, a general consensus was reached that there is no theoretically justified reason why ethnic diversity in itself could provoke violence ^[27]. In fact, ethnically diverse societies may "have the highest incentives to create transparent and inclusive political institutions, which, in turn, reduce the likelihood of terrorism" ^{[37, p. 40]}. Most ethnic relations are usually peaceful, and violence is only a deviation. In addition, an ethnic "conflict" does not necessarily have to be violent. At the same time, certain institutions can redirect ethnic activity into traditional politics and constructive activism ^[9].

However, there are a number of studies that emphasize how ethnicity can become a powerful motivator of terrorism under certain conditions. For example, in the work of O. Denzell, it was found that demographic polarization, rather than diversity and factionalism, increases the risk of terrorism ^[31]. Borrowing provisions from studies on civil wars and the work of D.L. Horowitz on ethnic conflicts, we can say that terrorism becomes more likely in states where the ethnic majority faces a large minority group that can actively compete for power and resources, forcing members of the group to perceive "others" as a more serious threat ^{[3; 41]}. Similarly, L. Heger's research shows that terrorism is more common among minority groups, which make up a large proportion of the country's population, and that these groups are better able to protect their population from enemy retaliation ^[37].

Another branch of the literature on terrorism focuses not so much on demographic competition as on how ethnic discrimination and political isolation fuel discontent and can provoke individual radicalization ^[1]. M. Crenshaw defines discrimination of an ethnic minority as a condition that can directly lead to terrorism ^[30]. And in a number of case studies of ethnic terrorist organizations, it is noted that discrimination based on ethnicity affects the formation and recruitment of organizations ^[10]. More recent work studying interethnic data shows that terrorism is more likely among ethnic groups with stronger political claims ^[21], as well as in countries where ethnic groups that have been subjected to significant economic discrimination or political isolation live ^[28].

The latter group of studies is largely based on T.R. Garr's theory of relative deprivation, first developed to explain uprisings and other forms of mass political violence, and then applied to ethnopolitical conflicts ^[2]. According to this theory, when people face deprivation of political or economic opportunities due to social identity (for example, ethnicity), this experience can sharpen their awareness of their social identity and alienate them from other members of society. This combination of an increasing sense of identity and alienation can make group members more susceptible to manipulation by extremists. Especially if the efforts of non-radical leaders to solve these problems encounter indifference or repression from the state ^[28]. As relations between groups become increasingly polarized, this serves to "empty the field of constructive centrists", increasing extremism and increasing the risk of political violence ^{[13, p. 22]}. Political isolation or lack of access to decision-making in the national Government, rather than general complaints, is the driving force that encourages members of ethnic groups to resort to violence instead of more peaceful forms of political mobilization.

The local component of ethnic isolation and terrorism

Although past work suggests a link between ethnic discrimination or political exclusion and terrorism, most interethnic terrorism research is mainly focused on the country level. While State-level factors do play a role in creating a polarized ethnic environment, it can be argued that the local environment is particularly important for understanding ethnic dynamics and its potential impact on terrorism. Previous case studies have linked the formation of a terrorist group to specific local factors or events ^[5], and have shown that the risk of terrorism often varies depending on the geographical territory of the State ^[46].

Even if ethnic relations are based on national factors, the consequences are felt at the local level. Since terrorist groups operate with fewer resources than many insurgent groups and are more limited in the choice of attack targets ^[30]. Consequently, they can focus their activities on more easily achievable local goals. In addition, even if the discontent is caused by national policy, terrorist groups can focus on undermining local authorities ^[4]. In addition, local authorities may themselves cause discontent, exacerbating discontent at the national level. Local governments, especially in developing countries, are more susceptible to corruption and more open to capture by competing groups ^[7]. They may also be more prone to budget and management problems, which leads to increased conflicts between local groups due to limited regional resources ^[6].

Ethnic groups deprived of State power do not have any means to solve their national or local problems, and the resulting frustration and "feelings of disenfranchisement" may prompt them to turn to violence ^[4]. It was found that countries that exclude a significant part of their population are at a higher risk of civil war and terrorism ^[28]. Although the link between ethnic isolation and the local risk of terrorism has yet to be analyzed, some scientists conclude that the presence of an isolated group in the region more than triples the risk of armed conflict in this region ^[17]. Extending this logic, this leads to assumption 1: "if an ethnic group removed from power lives in a region, it will be more susceptible to terrorist attacks."

Understanding local ethnicity and hostile environment

While political isolation can play an important and independent role in motivating terrorist activity, other environmental conditions are likely to lead to further escalation of group discontent and polarization of interethnic relations. In fact, although the presence of a politically isolated minority within a region may provide a basic motivation for violence, the risk of terrorism becomes even higher when this exclusion coincides with other economic, demographic and political conditions. This argument is related to the work of V. Asal, who argues that: "although political alienation is a powerful factor determining ethnic conflict, it can be mitigated by the economic and geographical conditions of an ethnic group" ^{[17, p. 1345]}. Their research was mainly focused on the effects of the "resource curse". Having found that regions with politically excluded ethnic groups and oil fields are significantly more prone to armed conflicts than regions with only one of these two factors present. This logic may extend to terrorism. Therefore, it is worth investigating other local conditions that may affect the exclusion and propensity of groups to this kind of violence.

If we return to the original arguments about complaints and discontent ^[2], discrimination and powerlessness combined with awareness of the status of the group in relation to others fuel ethnic conflict. Consequently, the feeling of frustration due to ethnic isolation should be most intensified in areas where intergroup competition is more typical. Because such an environment is more likely to increase awareness of intergroup comparisons. Economic conditions can play a significant role in the formation of group competition and conflicts. M. Sejman found that disadvantaged ethnic groups are more likely to participate in both protests and uprisings, and that groups in richer countries are less prone to violence than groups in poor ^[11]. However, the link between economic development and terrorism is less obvious ^[16]. The lack of such a consistent relationship may be due to fundamental differences in how historical grievances affect the activities of insurgent and terrorist groups. Insurgents need a wider spread of discontent in order to recruit ^[14], while terrorist groups function even when discontent is limited ^[30].

On the other hand, it is possible that the regional economy plays a more important role in the formation of terrorist violence than the wealth of the country. Low-income regions are experiencing a great shortage of resources, which increases competition between ethnic groups and increases the likelihood that different ethnic groups will view each other as opponents ^[3]. The outcome of this struggle will also cause hostility to the winners, increasing the desire to take revenge on more successful offenders ^[24]. On the contrary, accelerated economic growth reduces the risk of forced mobilization ^[20]. If we combine regions with low income and horizontal inequality (based on a group split, not on individual experience), then the risk of terrorism increases ^{[42, p. 302]}. This leads to hypothesis 2: "poorer regions where an isolated ethnic group lives will be more susceptible to terrorist attacks."

Finally, the urban environment is associated with both an increased risk of terrorism and greater ethnic polarization. Urban facilities are a very attractive target for terrorist organizations. Cities are centers of regional or even national importance, economy, media and infrastructure, and, consequently, a single attack in cities can be more destructive and cause more panic than operations in rural areas ^[51]. Cities also have great symbolic power, often combining local, national and even global identities and structures ^[15].

More importantly, the urban environment may be more susceptible to ethnic polarization and at greater risk of violence due to the intergroup dynamics arising from the interaction of many people. In the works on ethnic competition, it is argued that frequent contacts typical of urban conditions can contribute to both intra-group unity and intergroup conflict ^[8]. Localities are ideal places to create networks between ethnic groups, thereby increasing the mobilization potential. However, they can also serve as arenas of economic and social competition between different groups. Even if certain forms of urban civil society create the capital of ethnic groups and reduce the level of violence, too often contacts that arise in densely populated and industrialized countries increase the perception of ethnic differences and contribute to interethnic violence ^[50].

In fact, although settlements serve to strengthen intergroup contacts, it is the presence or absence of group inequality that determines whether this contact will become polarized and violent. Some studies associate radical Islam with the growing pace of urbanization in Muslim countries and believe that support for terrorism is highest among the urban poor ^{[42, p. 46]}. According to him, densely populated urban areas can become a "zero-sum intergroup environment" that terrorist leaders use to convince potential recruits that "terrorizing members of other groups is a cost-effective strategy." Settlements with isolated ethnic groups are not just areas of intensive intergroup contacts. They are also at higher risk of ethnic polarization and alienation. Essentially providing terrorist groups with ready-made recruitment networks. For these reasons, hypothesis 3 is that: "more populated regions where an excluded ethnic group lives will be more susceptible to terrorist attacks."

Despite the fact that it can be expected that terrorism is more driven by more local factors, but national policy plays a role in shaping these local interactions. There are good reasons to suspect that ethnic isolation will be most important in democratic societies where participation in the political process makes sense. The ability of the public to participate in free and fair elections is considered as the most important driving force affecting the risk of terrorism ^[38]. When access to the political process is restricted on the basis of ethnic origin, the consequences are even more significant. While S. Choi and D. Piazza question the influence of political participation on terrorism ^[28]. They believe that executive restrictions and political isolation of ethnic minorities significantly increase the risk of terrorism.

However, instead of considering the indicator of democracy as completely independent of other independent variables, democracy acts as an intermediate factor shaping the influence of ethnic isolation, population density and wealth. In particular, if the excluded ethnic group is located in an area whose population and economic dynamics have already increased the perception of group grievances and intergroup competition, then democracy can contribute to increasing frustration. An excluded ethnic group living in a democratic State may be more aware of the gap between democratic ideals of inclusivity and the reality of power relations. While some may find the argument that democracy can exacerbate existing ethnic tensions illogical, M. Sejman points out that democratic regimes are inherently systems of competition, and this competition can mobilize existing ethnic tensions ^{[11, pp. 108-109]}. In addition, local population and well-being are significant predictors of terrorist activity only under democratic regimes, assuming that many of these factors manifest themselves differently in democratic and non-democratic countries ^[46]. In other words, it can be expected that the type of political regime will enhance the effect of the interactions discussed above according to hypothesis 4: "the effects of hypotheses 1, 2 and 3 will increase in democratic countries."

Methodology and methods

To test these hypotheses, information from the Global Terrorism Database (GTD) is used. The data set covering 185 States from 1970 to 2019 includes domestic and international terrorist incidents ^[44]. The analysis presented in the paper focuses exclusively on incidents of domestic terrorism. Internal and transnational terrorism are caused by fundamentally different processes ^[35]. At the same time, social and ethnic dynamics are most important for understanding domestic terrorism. To encode internal attacks, the GTD Int_Any variable is used, which indicates whether the logistics, ideology, or target of the attack contains any international characteristics. Attacks containing an international dimension in any of the above components are excluded; thus, we have 85991 cases of domestic terrorist attacks.

It is important to note that all acts of domestic terrorism are analyzed, and not only those committed by terrorist organizations identified as ethno-nationalist. Given the absence of group ideology identifiers in GTD, this analysis has certain methodological and theoretical advantages. Firstly, coding the motivation of an attack based on the stated ideology of a group can be problematic: a number of terrorist organizations mix ideological types, and terrorists may adopt ideologies more for reasons of expediency than from true beliefs ^[40]. For example, Boko Haram bases its ideology on Islamic fundamentalism, but recruits its fighters overwhelmingly from the Kanuri ethnic group ^[29]. In addition, factors provoking the emergence of ethnically based terrorist groups can cause a ripple effect in society as a whole, since a common "culture of violence" can legitimize violence among all participants, increasing the overall risk of terrorism ^[43].

Dependent variable. To analyze the impact of local conditions, a sample of internal terrorist incidents is combined into a PRIO-GRID cellular structure ^[55]. PRIO-GRID divides the world into a set of 64818 cells-cells measuring 0.5 decimal degrees (or approximately 5555 km.) on each side, which cover all areas of land. These cells divide countries into separate units, each of which has a set of variables unique to the cell, which makes it possible to analyze how local and time-varying covariates affect the likelihood of terrorist attacks in the cell ^[46].

In order to compare internal terrorist incidents over the years, data on geographical specifics included in the GTD version are used ^[44]. Terrorist attacks are included in the analysis only if they can be identified by the specific latitude and longitude where the attack occurred, or if the GTD determines the location of the attack in the smallest administrative unit of the country. After these cases are identified, geocoding changes (the process of assigning an observation to a point on the map) and each attack is placed in the PRIO-GRID. This reverse geocoding was performed using ArcMap 10.8. The result is a final dataset that covers 191 countries from 1991 to 2009, which gives 1231542 cell-years in which 0 to 282 domestic terrorist attacks occurred.

While the range of data may suggest the use of ordinal regression, the use of a dichotomous dependent variable has theoretical and statistical advantages. Firstly, this work focuses on the general spread of terrorism, not on its severity. Therefore, it is best to use logistic regression for modeling. Secondly, while country-level studies have the advantage of "aggregating" events across the entire territory, which leads to a higher spread in the calculation, a more local analysis is dominated by a dichotomous process. In the dataset used, 99.02% of the cell-years did not experience any attacks, 0.5% suffered one attack, and 0.48% had more than one attack. Since very little information is lost during the transition to logit regression, and since the dichotomous variable reduces the impact of outliers, it is necessary to follow the path of a number of other terrorism researchers and use a simple dichotomous measure for the dependent variable ^[46]. The dependent variable is encoded as 1 if at least one case of domestic terrorism occurred in the PRIO-GRID cell in that particular year, and 0 if no attacks occurred in it during that year.

Independent variables. The main independent variable of interest is the indicator of the status of the ethnic group of the grid cell obtained from EPR data ^{[58, 59]}. Providing annual data from 157 countries on 733 ethnic groups at various levels of government, EPR uses expert surveys to identify all politically significant ethnic groups in the state, and groups are defined as significant if "at least one significant political organization has declared that it represents its interests at the national level, or if its members are exposed to political discrimination by the state" ^{[58, p. 1329]}. The EPR then ranks these groups based on the degree to which they have access to executive power in a given year. The EPR database is more comprehensive than many comparable datasets on discriminated ethnic groups, and is also time-dependent suggesting that a group's power status may change over time. Approximately 76% of the PRIO-GRID cells in this analysis were encoded using EPR. The missing cells for the ethnic groups in question were coded as "power sharing" (and accepted as the base category).

Three broad categories of ethnic power status from the EPR were used to encode each PRIO-GRID year cell based on the lowest-rated ethnic group residing in each cell (i.e., the ethnic group in that cell with the least access to power). Using this measure, two dichotomous variables were generated based on whether a given group has absolute power ("dominance") or this group is excluded from power ("exclusion"). Groups in power-sharing relationships were considered as a basic category. The use of the lowest-rated group is based on the work of L. S?derman, who found that ethnic groups with a high degree of isolation are most likely to rebel ^[25]. It is important to note that although the EPR is an indicator of involvement in the policy of the national government, the geographical distribution of ethnic groups varies across the country. Therefore, EPR values at the cell level may also vary.

To test hypothesis 2 on the relationship between regional wealth and terrorism, the "gross cellular product per capita" indicator (GCP indicator from PRIO-GRID), adapted ^[47], is used. Similarly to the gross domestic product (GDP) of a country, the CPI represents "gross value added in a certain geographical region; and is calculated as the total production of market goods and services in the region minus purchases from other enterprises" ^{[47, p. 3511]}. This indicator allows us to investigate differences in wealth and economic productivity within the borders of countries that would be inaccessible with traditional measurement, such as GDP ^[47]. Since the WCP is only available at five-year intervals since 1990, the missing values are interpolated and then this measure is recorded to account for emissions.

To test the role of population density from hypothesis 3, a population density indicator in cells was used. To calculate it, the population of PRIO-GRID cells was taken, which was adapted from the Gridded Population of the World Dataset, and divided by the total area of the cell ^[26]. In addition, given that terrorist activity depends on the population to provide targets and an audience, cells with zero population are excluded from the analysis. Since this variable depends on the population indicator from the PRIO-GRID, which is also available only for five-year intervals since 1990, the data were interpolated for the missing years.

Since both hypotheses 2 and 3 predict that the influence of regional wealth and population density is enhanced by the presence of an excluded ethnic group, each of the above variables is multiplied by an exclusion variable to create two interaction conditions. In addition, since hypothesis 4 states that the predicted relations should be especially relevant in democratic countries, the presented analysis is divided based on the type of regime and a separate model is given only for democracies. To do this, the Polity2 indicator from the Polity IV dataset is used. At the same time, countries that scored 6 points or higher are classified as democratic ^[39].

Control variables. Following previous studies on the relationship between geography and violence, subnational factors from the PRIO-GRID were included in the analysis ^{[34, 46]}. The study focuses on those variables that affect the ability of the state to govern at the local level (such as terrain, distance to the border and distance to the capital). The potential impact of civil conflict is also introduced, as well as spatial and temporal correlation.

Difficult terrain can reduce the ability of the central government to expand its power and gives non-State actors a stronger position from which they can launch their attacks ^[34]. To take this into account, the PRIO-GRID variable is used for the proportion of mountainous terrain in each cell, originally adapted from the report of the World Environmental Monitoring Center of the United Nations Environment Programme (UNEP-WCMEA) "Mountain Watch Report" ^[55]. The mountain terrain is determined in the report using both the height and slope of the slopes. This leads to six categories of locations exceeding 300 meters above sea level, as well as to the seventh category, which includes basins and plateaus located in mountainous areas ^[57]. Together, all seven categories account for almost 27% of the Earth's surface.

The distance to the state border and to the capital of the country from the PRIO-GRID is also taken into account. Border areas can provide safe haven for terrorists and transport weapons, especially if the state borders with a rival country. These locations also make it easier for groups to avoid the reach of government authorities after carrying out an attack. The boundary distance is measured as the distance in a straight line in kilometers from the center of the cell (or centroid) to the nearest adjacent state. Capitals are also at higher risk of attacks because they represent a range of symbolic and valuable targets ^[45]. The distance to the capital is measured from the center of each grid cell to the capital of the country.

Since the ongoing civil war increases the risk of terrorism, the presence of an ongoing civil conflict is also taken into account. As the Government's security resources are redirected to fighting local rivals, it may be easier for terrorist groups to operate in areas engulfed by civil war. In addition, the rebels may use terrorism as part of their general repertoire of violence ^[35] or to position themselves for subsequent negotiations ^[53]. Data on whether a cell is located in a conflict zone were taken from the UCDP/PRIO data set on armed conflicts ^[52], which was adapted for use in the PRIO-GRID ^[56].

Two controls are used to solve problems arising from the cross-section and the time-varying nature of the data. Firstly, in order to take into account the time dependence, a variable was included that counts the number of years since the last terrorist attack in this cell, and cubic splines were used ^[19]. Secondly, to cope with potential spatial autocorrelation, according to the analysis of civil wars in Africa X. Bukhoga and D. Roda ^[22], a dependent variable with a spatial lag is created. This variable, encoded at the cell-year level, measures the number of years during which a country from a cell was subjected to an act of domestic terrorism (in any part of its territory), divided by the number of years until the moment when the cell appears in the dataset. The values of this variable range from 0 to 1, where 0 indicates the absence of previous attacks throughout the country, and 1 indicates that the country was subjected to a terrorist attack for each year that appears in this analysis.

Logit regression with random effects is also used, since this model specification allows effective control of a number of additional factors at the country level ^[12]. This approach is chosen instead of fixed effects, since a number of variables (terrain, distance to the border and the capital) do not depend on time. To control the independence between cells of the same state, standard errors are grouped by country. Finally, additionally independent variables are taken with a one-year lag to account for endogeneity.

Results

Table 1 shows the results of logit regression. The first model provides a basic test of hypothesis 1. According to this hypothesis, it is found that cells with excluded ethnic groups are significantly more likely to be subjected to terrorist attacks. As shown in Table 2, the limiting effects confirm this. A cell with an excluded group has a 45.3% higher probability of a terrorist attack inside the country than in basic cases. These findings confirm that ethnic political isolation is more likely to cause polarization and violence ^[48]. Corresponding to hypothesis 4 in model 2, it is seen that these effects are amplified in democratic countries. When the analysis is limited to democratic countries, an increase in the risk of domestic terrorism by 417.4% is found in areas with a politically excluded ethnic group. In fact, only one other variable in our model – population density – exceeds the impact of ethnic isolation on the risk of terrorism.

Table 1. Models for cases of domestic terrorism.

Note: *p < 0.50; **p < 0.01; ***p < 0.001. Robust standard errors are given in parentheses. All explanatory variables are taken with a lag of one year. CPI per capita, population density, percentage of mountainous terrain, distance to the border and distance to the capital are logarithmically transformed.

Table 2. Initial probability and marginal effects for Table 1.

Note: Probability changes are given only for significant variables.

However, most of the research focuses on how other environmental factors reinforce the importance of ethnic group isolation. In particular, hypotheses 2 and 3 predict a conditional relationship in which the influence of local wealth and urbanization on terrorism is enhanced by the presence of an excluded ethnic group. To test these hypotheses, interaction conditions are created and then included in Models 3 and 4 in Table 1. Since these indicators are created by multiplying two variables, it is difficult to interpret the results based on coefficients alone. Therefore, in order to better interpret these relationships, a graph of marginal effects is constructed. Figures 1 and 2 show the results for these variables in Model 3 (all States) and model 4 (only democracies). These graphs show the change in the probability of domestic terrorism as the registered CPI per capita increases (Figure 1) and the registered population density (Figure 2), keeping all other variables in the model in their average value (for continuous variables) or modes (for dichotomous variables). Both of these figures present the results of Model 3 at the top and model 4 at the bottom.

Figure 1. The impact of political isolation on the CPSU per capita.

Note: Predicted probabilities for models 3 and 4 with vertical lines representing 95% confidence interval. Other variables are set to their average value or mode.

For the interpretation of each graph, it is important to note both the relationship between the constituent factor (regional wealth or population) and the risk of terrorism, and the difference between regions with at least one excluded group and without it. The two lines on each graph compare these varying probabilities for the cases, and the vertical lines represent a 95% confidence interval. The relationship is positive and significant if the line and its 95% confidence interval exceed 0. If all of them are below 0, then the relationship is negative and significant. If the confidence intervals intersect 0, the results are not significant with this value of the interaction indicator. As for the comparison of cells with excluded groups and without them, we can note a significant difference between these locations when the confidence intervals around each line do not overlap.

Figure 1 shows the impact of the registered VKP per capita and ethnic isolation on terrorism. Although it does not act in the direction predicted in hypothesis 2, it still reveals an interesting relationship between wealth, political isolation and terrorism. In local areas where there are no groups excluded from political power, wealth has less influence on terrorism. In localities where there is at least one excluded group, the registered CPI per capita has an increasing impact. This is most noticeable in the model of democracies (Figure 1). Since it is clear that the wealthiest areas are significantly more likely to be subjected to terrorist attacks if an excluded ethnic group also lives in them than in comparable areas without it. While this contradicts arguments about a lack of resources, this conclusion may be consistent with some previous work on terrorism, which argues that wealthier areas offer more targets for a terrorist group and may provide better conditions for recruitment ^[46].

In addition, the different effects of ethnic isolation on well-being does suggest some arguments about relative deprivation. Proponents of this approach argue that group frustrations are more likely to occur when members of an ethnic group realize their more disadvantaged position compared to other members of society ^[2]. Thus, disadvantaged groups who also live in affluent areas may be more aware of the existing horizontal inequality than people living in poorer regions, which contributes to inflaming ethnic tensions and increases the risk of political violence.

Turning to hypothesis 3, there is a positive and significant relationship between population density and terrorism. In fact, the graphs of marginal effects resemble an exponential growth curve. At the same time, the most populated cells are exposed to a significantly higher risk of terrorism. As can be seen from the graph, the relationship does not reach significance in the least populated cells of the model. In the full model, there is no significant difference between areas with excluded groups and areas without them. However, in democratic countries, densely populated cells with an excluded ethnic group are at significantly greater risk of a terrorist attack than comparable cells without excluded groups. Using the logic of ethnic unrest, densely populated urban areas can contribute to the creation of networks within various ethnic communities, thereby increasing competition and interethnic tensions, potentially provoking terrorism.

Figure 2. The influence of political isolation on population density.

Note: Predicted probabilities for models 3 and 4 with vertical lines representing 95% confidence interval. Other variables are set to their average value or mode.

With respect to control variables, these results are consistent in all models and confirm S. Nemet's earlier work on the influence of geography on terrorism ^[46]. All models show that terrorist attacks are more likely in cells with more mountainous terrain, closer to the border and where the civil war continues. As for the distance to the capital, a significant negative relationship is found in the full model. This means that the risk of terrorism is highest in the capitals of States and the areas closest to them. However, these relations are not observed in sub-elections with democratic countries.

Conclusion

This study analyzed extensive literature examining how ethnic isolation can motivate political violence. For ethnic groups, political isolation creates discontent and limits the possibilities for their consideration within the existing system. Political violence, be it civil war or terrorism, is considered a natural extension of this frustration. Confirming these assumptions, countries with politically isolated or economically discriminated groups are at higher risk of terrorism ^[28]. At the same time, ethnic groups that face discrimination are more likely to be involved in terrorism ^[21]. This logic extends to the subnational level, revealing that the presence of isolated groups significantly increases the risk of local terrorism.

Although a correlation has been found between the presence of isolated groups and terrorism, the question arises as to whether these isolated groups are the perpetrators of violence or only victims of these attacks. Is terrorism a "weapon of the weak" used to attack politically dominant groups, or is it a means by which more powerful groups maintain a social hierarchy? On the other hand, these motives can be much more complex, since the same group can direct violence to different objects depending on organizational needs or environmental factors.

Although the link between group exclusion and terrorism certainly requires further study, the main contribution of the current study is the assertion that ethnic discrimination itself depends on the context. Discrimination is formed depending on local and national factors. While other works link local wealth and localities with increased risks of terrorism, the analysis shows that the influence of these factors is enhanced by the presence of isolated ethnic groups. However, one of the most interesting results is the three-way interaction between ethnic isolation, type of terrain, population size, regional wealth, and the type of regime in the country. At the same time, the statistical trends identified in the full models are amplified in democratic countries.

In a sense , this conclusion is a confirmation of the results of K. 's work . Gledich and S. Polo, who found that modern trends towards strengthening democracy and ethnic reconciliation have significantly reduced the risk of domestic terrorism ^[36]. In general, the findings show that the benefits of democracy can be achieved only if the country also fights ethnic discrimination. Democracies that continue to exclude certain ethnic groups from political power are likely to face an increased risk of terrorism, not a decrease. This means that for democratic countries, ethnic integration and representation is more than just an abstract democratic value, it is also a security issue.