Deprivation, instability, and propensity to attack: how urbanization influences terrorism

International Interactions Empirical and Theoretical Research in International Relations ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/gini20 Deprivation, instability, and propensity to attack: how urbanization influences terrorism Maxim Slav, Elena Smyslovskikh, Vladimir Novikov, Igor Kolesnikov & Andrey Korotayev To cite this article: Maxim Slav, Elena Smyslovskikh, Vladimir Novikov, Igor Kolesnikov & Andrey Korotayev (2021) Deprivation, instability, and propensity to attack: how urbanization influences terrorism, International Interactions, 47:6, 1100-1130, DOI: 10.1080/03050629.2021.1924703 To link to this article: https://doi.org/10.1080/03050629.2021.1924703 View supplementary material Published online: 17 Aug 2021. Submit your article to this journal Article views: 429 View related articles View Crossmark data Full Terms & Conditions of access and use can be found at https://www.tandfonline.com/action/journalInformation?journalCode=gini20 INTERNATIONAL INTERACTIONS 2021, VOL. 47, NO. 6, 1100–1130 https://doi.org/10.1080/03050629.2021.1924703 NOTE Deprivation, instability, and propensity to attack: how urbanization influences terrorism Maxim Slav a, Elena Smyslovskikh and Andrey Korotayev a,b a , Vladimir Novikov a , Igor Kolesnikov a , a HSE University; bInstitute for African Studies ABSTRACT KEYWORDS The study investigates different ways in which urbanization and its tempo influence terrorist activity. In line with other researchers investigating nonlinear effects on instability, we suggest that the influence of both of them is nonlinear, with quadratic regression being more appropriate for urbanization level impact and interaction between urbanization and its tempo being more appropriate to measure the tempo’s influence. Nonlinearity has been confirmed in the robustness section of the paper, in which an alternative dependent variable distribution and a greater set of control variables were used. The findings are in line with those of other researchers who found that societies, in the process of modernization, demonstrate heavier instability than societies before modernization or those after the modernization period. terrorism; urbanization; instability; modernization; non-linear effects El estudio investiga las diferentes maneras en que la urbanización y su ritmo influyen en la actividad terrorista. En consonancia con otros investigadores que estudian los efectos no lineales en la inestabilidad, sugerimos que la influencia de ambos aspectos es no lineal, con la regresión cuadrática siendo más adecuada para el impacto del nivel de urbanización y la interacción entre la urbanización y su ritmo siendo más conveniente para medir la influencia del ritmo. La no linealidad se ha confirmado en la sección de solidez del artículo, en la que se utilizaron la distribución alternativa de variables dependientes y un mayor conjunto de variables de control. Los hallazgos son coherentes con los de otros investigadores que observaron que las sociedades, en el proceso de modernización, demuestran una inestabilidad más intensa que aquellas antes o después del período de modernización. Cette étude enquête sur les différentes manières dont l’urbanization et son tempo influencent l’activité terroriste. En accord avec d’autres chercheurs étudiant les effets non linéaires sur l’instabilité, nous suggérons que l’urbanization et son tempo ont tous deux une influence non linéaire, une régression CONTACT Andrey Korotayev akorotayev@gmail.com HSE University, 20 Myasnitskaya, Moscow 101000, Russia Supplemental data for this article can be accessed on the publisher’s website © 2021 Taylor & Francis Group, LLC INTERNATIONAL INTERACTIONS 1101 quadratique étant plus appropriée pour mesurer l’impact du niveau d’urbanization, et l’interaction entre l’urbanization et son tempo étant plus appropriée pour mesurer l’influence du tempo. Cette non-linéarité est confirmée dans la section sur la robustesse de l’article, dans laquelle une distribution alternative des variables dépendantes et un plus grand jeu de variables de contrôle sont utilisés. Les conclusions de l’article sont en accord avec celles d’autres chercheurs qui ont constaté que les sociétés en cours de modernization présentaient une plus grande instabilité que celles qui étaient déjà passées ou qui n’étaient pas encore passées par une période de modernization. Introduction Urbanization is an important factor in stimulating instability. Chiefly, two reasons stand behind this connection. First, rural–urban migration is a process connected with structural changes in society and is highly likely to produce multiple dissatisfied individuals without proper social networks (Geifman 2005; Kornhauser 1959; Korotayev et al. 2011). The prevalence of mechanic solidarity among individuals combined with the exposure of multiple groups in a dense urban area sparks anomie and tensions among those with different modes of thinking (Douglas 2012). This leads to an escalation of grievances inside the cluster of newly integrated members of society, who still suffer from various forms of socioeconomic deprivation, and this results in overall destabilization. Second, cities are also important logistically, and they are more attractive among the agents of instability for various reasons (Coward 2009). In the case of terrorist organizations, cities provide the base of human resources which is essential for their persistence; likewise, the conduction of a terrorist attack in a city means a larger contribution to the production of fear, inconvenience, and discomfort in the lives of citizens of the target country (Mccartan et al. 2008), since terrorists usually destroy strategic infrastructure and shatter residents’ everyday life. In line with Mccartan et al. (2008), we define terrorism as “premeditated use of violence by subnational groups to obtain political, religious, or ideological objectives” (Mccartan et al. 2008, 61). Even though we focus, here, on terrorist attacks specifically, the conclusions that we come to can be extrapolated to other manifestations of instability as well. Urbanization is also one of the key indicators of the modernization process. This complicates our analysis. The previously mentioned ways in which urbanization influences terrorism are linear: (a) more cities (hypothetically) mean more terrorist attacks, and (b) more rural–urban migration means greater dissatisfied first-generation proletariat. However, as Huntington (1968) as well as other researchers on economic and political determinants of instability hint, both of the alleged links can be curvilinear (Korotayev, Vaskin, and Bilyuga 2017; Korotayev et al. 2018). Just like any other indicator 1102 M. SLAV ET AL. of modernization, urbanization, per se, is predicted to have a specific inverted U-shaped correlation with instability. It was also stressed by Huntington that the instability coming from the destruction of mechanical solidarity and mismatch between the old ways of living and complex societies must end at some point. It was proposed that new institutions should be developed over time, and social interaction was transformed into so-called organic solidarity. In other words, individuals “learn the art of associating together” (Huntington 1968, 5). At the same time, those enjoying organic solidarity are less sensitive to terrorists’ aims of division as well as their methods. The resulting mistrust of the residents hampers terrorist activity. This means that problems, caused by increased rural–urban migration, can be at least partially neutralized; at the same time, higher rural–urban migration hastens modernity as a relatively stable haven for society. We, hence, consider in the current study the effects of both urbanization and its tempo, since the exclusion of any of these would make the research incomplete. We test for both linear and nonlinear types of relationships as well as for interaction between the level of urbanization and its tempo. The rest of the paper is organized as follows. First, in the literature review, we display evidence and theories at hand and develop hypotheses to be tested afterward. We then describe the data we use and the modifications we make. In the exploratory part, we split the sample into several subsamples and find an upward trend in the first half and a downward trend afterward. In the rest of the paper, we use quadratic regression on the whole sample to test for curvilinearity. Following one of our reviewers’ advice, we consider the number of terrorists’ victims as well as well as several forms of the main independent variables. Then, we show the results of the models we estimate in the main part of the work, where we conduct negative binomial regressions. We then present our Robustness section, presented in online supplementary materials. To test the robustness of our models, we use both alternative distribution (namely, quasi-Poisson) and alternative theories to validate the results from our main part. Overall, we find nonlinear relation to be more appropriate, although linearity reveals itself sometimes. Finally, in the discussion section, we present cases that can be explained with the hypotheses that are developed and supported empirically in this paper. Theoretical Approach In this section, we cover three major links between urbanization and terrorist attacks. First, there are several reasons, cited in the literature, for terrorists to choose cities and dense urban areas as their targets. Second, the process of urbanization, per se, might increase the propensity of an average citizen to become a terrorist. Third, urbanization might be intrinsically linked to other parts of modernization, which, in turn, provoke terrorist attacks. We also INTERNATIONAL INTERACTIONS 1103 mention, in this section, control variables that are used in the main modeling part. Fourth, we cite alternative hypotheses that we test in the robustness part. The literature review, accordingly, comes in the following subsections: first, we explain the effect of cities’ attractiveness to terrorists, second, we cover the effect of urbanization speed on terrorism. Then, in the third subsection, we focus on the effect of other phenomena and variables correlated with urbanization, and finally, we provide alternative explanations. Overall, we find that urbanization has been observed as both a positive and a negative correlation of terrorism. Due to this uncertainty, we question the linearity of the relationship between urbanization and terrorism. Our hypotheses in this research come from the two main explanations; so, the links we test are devoted to two major predictors, which are urbanization level, on the one hand, and tempo of urbanization, on the other. Hypothesis 1: 1.1. The level of urbanization has a positive linear relationship with the number of terror attacks and the number of terrorist killings. 1.2. The relationship between urbanization level and the number of terror attacks and the number of terrorist killings is inverted U-shaped. Hypothesis 2: 2.1. Tempo of urbanization is positively related to the number of terror attacks and the number of terrorist killings. 2.2. The effect of tempo of urbanization is conditional on its tempo: we expect inverted U-shaped relationship. Cities’ Attractiveness Today, the major school of thoughts on terrorism considers terrorists as rational actors. A terrorist attack, per se, is considered as a means to “bargain” with the other party, by making the average citizens fear hypothetical direct physical harm and to push the government to let the organizations achieve their aims. They also try to get more supporters, which sometimes brings ambiguity about their choices: the victim countries’ citizens may sympathize with possible terrorists’ supporters (Adelaja and George 2019; Mccartan et al. 2008). As for the choice of recruits, knowledgeable and skilled potential candidates are concentrated in cities, and this is important for terrorist organizations which screen the volunteers for quality, rather than accept the most likely volunteers who often lack education or ability (Bueno de Mesquita 2005). The choice of a target is calculated as well, and there are several reasons why cities are more popular as targets. First, a terrorist attack, there, could gain more attention compared to a rural 1104 M. SLAV ET AL. terrorist attack (Campos and Gassebner 2013; Mccartan et al. 2008; Tavares 2004), and proximity to cities sometimes proves to be a positively significant predictor (Python et al. 2019a). Second, as some scholars stated, although terror originates in places with high social breakdown and resource mobilization, it is often transmitted to globally oriented, target-prone cities with high potential of conveying a global “message” (Savitch and Ardashev 2001). This might create what some researchers call “vicious cycles” of terrorism, which happen as a result of continuous terror and counter-terror actions (Beall 2006). Third, some actions of terrorists are aimed not at the creation of fear of direct physical threat, but at targeting critical infrastructure: communication, logistics systems, etc. – to hamper the victim country’s usual way of life (Coward 2009). These targets are most usually concentrated in cities. All in all, cities are more preferable as targets than rural areas; hence, the conduction of a terrorist attack is less costly in an urbanized country compared with a rural country. Since the cited evidence does not suggest that the effect is different on different levels of urbanization, the first prediction, Hypothesis 1.1, is as follows: the influence of urban population on the number of terrorist attacks is positive for all levels of urbanization. The positive link proved itself in several studies (Tavares 2004) and the urban population as a predictor sometimes has a stronger effect than GDP per capita (Campos and Gassebner 2013). However, it also manifests ambiguity in some veiled way in research on latent anger in Africa (Adelaja and George 2019) which, in turn, stimulates unrest and, specifically, terrorism. Latent anger was found to have a significant negative relationship with urbanization, but the researchers explained this relation by the specificities of the organization of the terrorists and the execution of territorial control. Finally, one of the lines of research shows significant negative coefficients for urbanization as regards more urbanized countries (Korotayev, Vaskin, and Tsirel 2019; Vaskin, Korotayev, and Tsirel 2018). This brings us to the conclusion that the relation between the level of urbanization and the number of terrorist attacks is actually curvilinear (Hypothesis 1.2). There were attempts to test the curvilinearity of the influence of more common determinants of terrorism. For instance, there is evidence that supports the existence of the curvilinear relationship between domestic terrorism and economic development (Enders and Hoover 2012; Enders, Hoover, and Sandler 2016; Gassebner and Luechinger 2011; Ghatak and Gold 2017; Lai 2007). Still, inspections have not yet been made into nonlinear relationships between urbanization – either its overall level or its tempo – and terrorism. Some studies used OLS methods (Tavares 2004) which may be inappropriate for the nature of the examined data (Kis-Katos, Liebert, and Schulze 2011) or inspected correlations and found no clear results for urbanization as a predictor (Newman 2006). That said, investigation of a nonlinear relationship, via more suitable statistical methods, may deserve more attention. INTERNATIONAL INTERACTIONS 1105 The nature of this curvilinearity was hinted by Huntington (1968): modernity breeds stability, but modernization breeds instability (Huntington 1968, 41; see also Korotayev et al. 2011, 2018). Concerning urbanization, we expect that instability increases at the beginning of the process (the period when modernization begins as well) and starts to decrease from a certain point. This gives us Hypothesis 1.2: the influence of the urban population on the number of terrorist attacks is positive at the beginning of the urbanization process and negative at the end of the urbanization process. Note, however, that in Hypothesis 1.1, the reason for the effect arises from the relative ease of undertaking terrorist attacks in cities. In the case of Hypothesis 1.2, though, the reason for the effect is rooted in social changes. Some other reasons for the decrease of terrorism likelihood at high urbanization levels are related to the overall level of development in the society. As was stated, terrorist attacks serve as a form of expressing social grievances, and it is obvious that the degree of the grievances, and well as motivations of the poor, differ between agrarian and urbanized societies. We may see this for the case of Turkey in the 1970s when the country’s urban centers became arenas of organized political violence. The rise of terrorism came as a result of rapid urbanization and social transformations, leading to a dramatic decrease of rural–urban migrants’ quality of life compared to urban dwellers: namely, migrants formed an entire precarious class living predominantly in squatter settlement districts (Sayari and Hoffman 1994). The situation changed during the 1980s after the Turkish state had gained control over urban centers, and the number of terrorist incidents went down, which proves the claim of Charles Tilly: one of the major reasons why modern states were successful is that the object of their control had undergone changes in the process of urbanization. The concentration of population in cities became much higher, which means that it became easier for states to accumulate resources in urban centers, and to build technologies of control, too (Tilly 1992). One potential reason for such decrease is that, starting from a certain point, it becomes hard for terrorists to gain access to recruitment and human resources in cities due to lack of trust among residents. This means that the decline in terrorism at higher levels of urbanization may be driven by parallel mechanisms: social changes provide the state with additional resources to develop coercive apparatus and other mechanisms of control, and at the same time, urban economies grow, which softens the deprivation of the unemployed and socially excluded migrants. If we look at education as a process associated with urbanization, there is evidence that it promotes terrorism in societies with less favorable conditions while reducing the number of terror attacks in more stable and developed countries (Brockhoff, Krieger, and Meierrieks 2015; Korotayev, Vaskin, and Tsirel 2019). Thus, over time, cities become more secure both because of social stabilization and the reinforcement of the state. 1106 M. SLAV ET AL. Tempo of Urbanization As Kornhauser (1959) argues, “The rapid influx of large numbers of people into newly developing urban areas invites mass movements” (Kornhauser 1959, 145). Also, as Grinin and Korotayev (2009) note, political instability tends to spread at the moments of high rural–urban influx. Both Kronhauser and others, in their research, show that the problem is twofold. First, individuals, who move from home villages to cities, lack proper social networks and coping mechanisms appropriate for urban life, which makes them more vulnerable psychologically. The first wave of rural–urban migrants is also characterized by mechanical solidarity, which implies solidarity by commonality (Durkheim 1964; Huntington 1968). In rural areas, unity of individuals is observed. In urban areas, where different types of people can be observed, the conflict is imminent. It may not result in an immediate attack, but the certainly adds to frustration (Douglas 2012). As a study on terrorists at the beginning of the twentieth-century shows (Geifman 2005), terrorists – both the migrants from the villages and the others – suffered from severe psychological problems, and terrorism became one of the psyche’s defense mechanisms. In the research, it was also argued that at least half of the Socialist Revolutionary terrorists were first-generation proletariat, who came from the countryside. Second, rapid urban growth usually means that many citizens lives do not match decent living standards (Ibimilua 2011; Korotayev et al. 2011; William and Piyu 2008). They can also sometimes be discriminated against, for example, with the restrictive institution of registration (Ding and Ding 2012). This makes the first-generation urban residents specifically fertile for various social diseases. The lethality of terrorism and the frequency of lethal attacks are also proven to be driven by antagonistic mechanisms (Python et al. 2019b). It is claimed, for example, that the likelihood of an attack being lethal is higher in poorer areas that are prone to instability and conflict. Hence, the tempo of urbanization might be a useful factor for a terrorism researcher. To our knowledge, there were a few attempts to test such a relationship directly; however, for different reasons, we consider them to be insufficient. There is no straightforward evidence that urban population growth may be the cause of the rise in terrorist attacks, as long as the mediating effect of social inequality and the inequitable access to such resources as education and basic public services take place (Østby 2016). Therefore, individual and collective inequality, as well as relative deprivation, give motivation for political violence among rural–urban migrants. Thus, the effect of such a migration is indirect. In an article by Buhaug and Urdal (2013), city population growth is tested as a predictor of lethal and nonlethal urban violence events. Although their findings show that the direct link is negligible and urban violence is unrelated to city population growth, their estimations still demonstrate that, under some conditions, urban growth may increase INTERNATIONAL INTERACTIONS 1107 the number of lethal urban violence events, which is, despite being of low significance, still observable with a positive coefficient. For example, the interaction effect of economic shock and city population growth happened to be negative. We suppose that the absence of a direct relationship may be the result of a poor choice of indicators. And it might be more apt to concentrate on the level of urbanization, instead of the city population, so that different urbanization periods are considered. Apart from the results mentioned above, the possible link revealed itself in other qualitative research. In their research, Korotayev, Malkov, and Grinin (2014) mention that civil unrest can characterize societies at the end of the first-phase demographic transition. However, this research was not aimed at terrorism, per se, but at political violence as a whole. Focusing on purely terror events, as a subject of study, is essential, due to the difference in the nature of violence. Korotayev et al. (2019), also tried to explain the negative relationship between the urbanization rate and terrorism by a negative relationship between the urbanization rate, per se, and the tempo of urbanization. However, although noted by preceding research, the link between urbanization tempo and terrorism was not tested directly. All in all, this brings us to the second hypothesis. On the one hand, the cited evidence does not imply the need for curvilinearity. Thus, Hypothesis 2.1 comes as follows: the tempo of urbanization has a positive influence on the number of terrorist attacks on every level of urbanization. However, if Huntington (1968) is right, it could be that the main problem is not in the changes alone. As he proposes, “instability . . . derives precisely from the failure to meet this condition: equality of political participation is growing much more rapidly than ’the art of associating together’” (Huntington 1968, 5). If social and political institutions become more solid as modernization continues, the frustration of the first-generation urban residents could be absorbed more easily, and hence proceeding with modernization does not increase, but decreases instability. Even though the societies with strong organic solidarity (Durkheim 1964) may still produce terrorists, their activity may be seriously hampered by it due to the residents’ opposition to terrorists’ methods. Hence, Hypothesis 2.2 is as follows: the tempo of urbanization has a positive influence on the number of terrorist attacks at the beginning of urbanization, and a negative influence on the number of terrorist attacks at the end of urbanization. Other Variables’ Effect Following the line of reasoning by (Huntington 1968), other components of modernization could be blamed for instability and, consequently, terror attacks, and, hence, they should be taken into account. 1108 M. SLAV ET AL. Economic factors Much research has been done on the influence of economic development on terrorism, and scholars have shown interest in evaluating the propensity of a country to experience more terrorist attacks, depending on either its wealth level or the tempo of development. There has been much evidence of poverty provoking terror attacks as a part of more general consequences of poverty such as political instability and the rise of civil unrest. Poverty has been linked to terrorism by governments and has had its implications in establishing policies on reducing poverty levels worldwide (Kahn and Weiner 2002). With respect to those conventional views, researchers looked for a causality between economic development and terrorism. The findings are quite diverse though. Some studies demonstrate that it is not actually the poorest countries that breed terrorism, but intermediate-level economies, where the need for economic improvement is highly likely (Daniel and Thomas 2013). There is evidence showing that, indeed, low opportunity costs for terror, in countries with slow economic growth, lead to more terrorist organizations there. However, this effect is only relevant after a certain level of development has been reached (Freytag et al. 2011). The authors also conclude that the causality between terrorism and economic growth does not remain stable over time and varies in different countries, which indicates the impact of shifting geographical and ideological patterns in terrorism that are associated with the end of the Cold War. The ambivalent effect of growth also proves itself via separating agriculture from industry and making a distinction between different kinds of terrorism. Seung-Whan (2015) infers that, despite the positive effect of stable economic growth in preventing domestic and international terrorism, some kinds of terrorism (suicide attacks) are still more likely to happen in prosperous countries with a steady industrial growth. Other studies have found no evidence that economic development promotes terror. For instance, they claim that such predictors as population, ethno-religious diversity, increased state repression, and the structure of party politics are more significant predictors of terrorism compared to growth, which validates that social distinctions should be taken as explanatory factors primarily (Abadie 2006; Piazza 2006). Still, there is proof that there is a positive relationship between more countries having more favorable economic conditions or obtaining foreign aid, and the decrease in the number of terror attacks there (Azam 2012; Gassebner and Luechinger 2011). Finally, as mentioned above, some studies argue that economic development is positively correlated with terrorism at lower levels of development and is negatively correlated at higher ones (Enders and Hoover 2012, Enders, Hoover, and Sandler 2016; Gassebner and Luechinger 2011; Ghatak and Gold 2017; Korotayev, Vaskin, and Tsirel 2019; Lai 2007; Vaskin, Korotayev, and Tsirel 2018). INTERNATIONAL INTERACTIONS 1109 Our assumptions about the link between urbanization levels and terrorist attacks derive from the fact that more stable conditions are more common for the lowest and the highest levels of urbanization. Therefore, the relationship is highly likely to be positive in the first urban deciles and become negative after reaching the peak of the highest instability level at average urban deciles. Political Factors Another direction of research concentrates on political predictors of terrorism, and the main explanations boil down to three causal mechanisms. First, democracies prove to be less successful in dealing with terror attacks due to the lack of coercive forces (Eubank and Weinberg 1994, 2001; Schmid 1992). They impose constraints on the executive that make law enforcement more difficult and thus make it easier for terrorists to operate (Kis-Katos, Liebert, and Schulze 2011). Second, domestic political instability and poorly managed internal political conflicts promote terror (Campos and Gassebner 2013; Piazza 2009; Tavares 2004), whereas regime durability reduces the number of terror events. This goes in line with Erica Chenoweth’s research, in which she finds that cases of terrorism are most likely to happen in weak and transitioning democracies (Chenoweth 2013). In contrast, advanced democracies or full autocracies experience terrorism in a lesser degree. That said, we may suggest that terrorism is associated predominantly with the inconsistency of political institutions in intermediate regimes (Korotayev, Vaskin, and Romanov 2019). According to the results of a research by Fahey and Gary (2015), what underpins terrorism is social disorganization produced by unstable states. In the presence of political instability, such as ethnic war or regime transition, existing social organization is no longer capable of providing prosocial behavior of individuals via institutional or informal mechanisms. Their findings support Durkheim’s claim: societies experiencing rapid social change cannot integrate their members properly, which leads to the proliferation of antisocial behavior of all forms, including political violence. Therefore, political instability increases the likelihood of terrorism cases due to lack of social organization (Fahey and LaFree 2015). The third approach considers democracies to be better at reducing terrorism at the level of society via more open systems which make it possible for social groups to express their interests at fewer costs, instead of holding hard constraints and coercive power as tools of control for already existing conflicts. As some scholars state, alleviating existing conflicts and establishing social welfare policies can diminish domestic and international terrorism alike by reducing the conditions for terror as economic insecurity, inequality, poverty, and religious-political extremism (Burgoon 2006). Thus, democratic regimes provide more opportunities for nonviolent political expression, and thereby the grievances are less likely to turn into terrorism and more often result in 1110 M. SLAV ET AL. more peaceful forms of interest articulation (Eyerman 1998). While “terrorism is an attractive strategy for small organizations of diverse ideological persuasions who want to attract attention for their cause, provoke the government, intimidate opponents, appeal for sympathy, impress an audience, or promote the adherence of the faithful” (Crenshaw 1981, 396), giving more ways to articulate and lessen social tension may help reduce terror attacks. Some findings show that regime type may affect terrorism differently. A recent study shows a robust inverted U-shaped impact on various terrorism measures (Gaibulloev, Piazza, and Sandler 2017), whereas “strict autocracies and full-fledged democracies are much less plagued by terrorism than anocracies” (Gaibulloev, Piazza, and Sandler 2017, 519; see also Slinko et al. 2017; Korotayev, Vaskin, and Romanov 2019). Some other political explanations stem from the characteristics of potential targets and go further than the condition of internal policies and the degree of violence in society. When it comes to transnational terrorism, the likelihood of an attack rises under recent local experience with civil war battles, actual high levels of civil violence, and overall low security levels (Marineau et al. 2020). Education Despite the former prevailing arguments for foreign aid and education promotion on the part of developed countries as the necessary steps for fighting terrorism, more recent works demonstrate a positive link (Berrebi 2007) and claim that secondary or higher education is positively associated with participation in terrorist organizations. The explanations generally build on rational choice theory (Azam 2012) and the fact that terrorist organizations, as mentioned, recruit well-educated members strategically (Bueno de Mesquita 2005). Contrary to the proposed positive correlations, there is statistical evidence showing that education itself has an insignificant effect on terrorism; rather, education levels accelerate the effect of poor political and socioeconomic conditions as the main correlates for terror attacks (Brockhoff, Krieger, and Meierrieks 2010; Danzell, Yeh, and Pfannenstiel 2020; Krueger and Maleckova 2003). A curvilinear relationship between education levels and terrorism seems to be relevant: there is evidence that the intensity of terror attacks increases up to the level of approximately 3 to 6 years of schooling and declines afterward (Korotayev, Vaskin, and Tsirel 2019). Based on these results, we expect a positive relationship between education levels and terrorist attack intensity for lower urban deciles and a negative relationship at the highest levels of urbanization. Group Discrimination and Civil Liberties Any form of group discrimination is a powerful cause of frustration, anger, and, hence, extremism and terrorism (Geifman 2005; Skoczylis and Andrews INTERNATIONAL INTERACTIONS 1111 2020). In this research, we consider gender, religious, and ethnic discrimination as the most common types. Women comprise half the population, yet they are frequently institutionally discriminated against, but it seems that they rarely reveal their anger via terrorism: compared to male terrorism, female participation is low (Ness 2007). Previously, it was noted that their labor force participation rate negatively affects the number of terrorist attacks. We suppose, the lower the gender discrimination in a country, the lower the number of terrorist attacks. The overall level of civil liberties is measured in this work based on Polity5 index, as well as an alternative measurement by Freedom House. Other Control Variables The size of the population has proven itself in many works concerning terrorism (Gassebner and Luechinger 2011). Its predictive value is quite intuitive: larger countries will have more terrorist attacks than other countries, both because greater opportunities are available, and as some note, “larger countries are heterogeneous and subject to more intergroup tension” (Burgoon 2006, 189). It was also hypothesized that larger countries’ terrorist attacks attract greater media attention and that it is harder to establish firm surveillance and control there; a terrorist attack in a large state is also more likely to yield a greater return (Dreher and Justina 2010; Eyerman 1998). Finally, urbanization and demographic transition are both components of modernization and sometimes coincide; it is, hence, important to consider population growth, per se, as a potential cause of the increase in terrorist activity. In the case of transnational terrorism, the set of factors that make cities potential targets of terror attacks include accessibility, symbolism, material harm, and exclusion (Marineau et al. 2020). For instance, a smaller distance to the international border makes it possible for more terrorist organizations to access a country, while proximity to capital areas serves the purpose of message translation and symbolism, as well as higher potential material harm. Data and Methods We construct a time-series cross-sectional data set for 240 unique country entries from 1970 to 2018. As a response variable, we use two different measures from the Global Terrorism Database (LaFree and Dugan 2007): the number of terrorist attacks and the number of fatalities resulting from terrorist attacks in a given country in a particular year. As key predictor variables for the first hypothesis, regarding the direct impact of urbanization, we use the proportion of the urban population (UN data) and 5- and 10-year rolling average of the proportion of the urban population. Regarding the second hypothesis, we use growth rates of the share of the urban 1112 M. SLAV ET AL. population (UN data) and 5- and 10-year rolling average for the robustness. Note that we exclude several observations from the sample, namely the ones related to the Vatican city-state, which has a low registered population but a high number of tourists, thus leading to the bias. Table 1 contains a description of our data set in terms of measurements, data sources, time span, etc., and Table 2 presents descriptive statistics for the data. Our dependent variables represent instances of terrorist attacks or killings in a given country-year which is essentially count data. In this case, the default option to draw inference is a Poisson regression. However, the Poisson distribution assumes that the mean and variance are the same, and is, hence, vulnerable to overdispersion when the latter is higher than the former. The maximum number of terrorist attacks in one country in 1 year is 3774, the mean number is 17.7 and the standard deviation equals 110.6.1 To tackle the issue of overdispersion, for the baseline empirical modeling (results presented in the next section), we use negative binomial regression from the new fixest R package (Berge 2018). Even though previous researchers used different R packages for negative binomial models (for example, Zeileis, Kleiber, and Jackman 2008), we found them seriously inconvenient and inconsistent when approaching two-way fixed effects. In the robustness section that is presented as an online supplementary material, we use quasi-Poisson estimators from glm package (Croissant and Millo 2008). Regarding the proposed nonlinear relationship, during the preliminary analysis, we divide the data into 20 vigintiles2 to test whether the relationship is similar for different subsamples. Figure 1 shows the mean number of normalized terror attacks for each vvigintile, as we observe from the plot there is a clear upward trend for the first half of the sample and fluctuations at the higher levels of urbanization. Table 3 presents the results of negative binomial regressions with the number of terror attacks as a response variable and the proportion of the urban population as a predictor. The first model is estimated for the whole sample, the second for the lower level of urbanization, the third for the medium level, and the last one for the higher level of urbanization. For the whole sample, the share of the urban population is a negative yet insignificant predictor, for the lower levels of urbanization it is positive and significant. In the middle, the effect is insignificant and at higher levels of urbanization it is negative and significant at a 0.05 significance level. Thus, we assume that the relationship between urbanization and terrorist activity varies at different stages of urbanization. We also model this through a quadratic term of the proportion of urban population or through an interaction effect between the growth of the share of the urban population and the proportion of urban population itself. 1 For homicides maximum is 13965, mean 42.9, and SD 350.3. Based on the share of the urban population, where the first vigintile stands for the lowest levels of urbanization and the twentieth for the highest. 2 Table 1. Variables used in the results and robustness sections. Variable Terrorist attacks Years Description Measure 1970–2018 The number of terror attacks in Units country-year Terrorist killings Measurement level 1970–2018 The number of victims of terror Units attacks in country-year 0 – lower level Global Terrorism Database Proportion of urban population 1950–2018 Share of urban population In percentage points 0 – lower level United Nations Population Division Growth of urban population (proportion) 1950–2018 Change of share of urban population In percentage points 0 – lower level United Nations Population Division Population 1960–2019 Size of population in country in Units particular year 0 – lower level United Nations Population Division GDP per capita (PPP) 1960–2019 PPP per capita at constant 2017 prices Dollars 0 – lower level World Bank Polity score 1919–2017 Polity5 score Discrete variable −10 – full autocracy, 10 – full democracy Polity5 Annual Time-Series Occupation 1919–2017 Polity5 score being −66 or −77 Binary variable 0 – lack of occupation of any sort and size, Polity5 Annual Time-Series 1 – presence of occupation of any sort and size War 1816–2014 Being in a state of war Binary variable 0 – being in a state of war, 1 – lack of state of war Correlates of War Project State capacity 1960–2009 Index (Hanson and Sigman 2013) Discrete variable −4 – lower level Hanson, Sigman, 2013 The number of refugees 1960–2017 WDI score Population count population count World Development Indicators Female labor force participation 1960–2019 Share of economically active In percentage women in the female points population ages 15 or older 0 – lower level World Bank Political freedom Units 1972–2017 Aggregated freedom index mean score for civil liberties and political rights 1 – higher level of freedom, 10 – lower level of freedom Freedom House Proportion of discriminated population 1946–2017 Proportion of population that In percentage points experiences active and intentional discrimination to exclude it from political power 0 – lower level Ethnic Power Relations Data Set Family 1113 Source Global Terrorism Database INTERNATIONAL INTERACTIONS 0 – lower level 1114 Statistic Terrorist attacks N 9,564 Mean 17.715 St. Dev. 110.581 Min 0.000 Terrorist kills 9,564 42.888 350.269 0.000 Proportion of urban population 13,916 46.529 24.983 1.700 Population 13,916 25,290.400 102,649.600 2.950 GDP per capita PPP 8,612 14,166.670 17,754.320 Polity score 8,301 0.511 7.497 Proportion of discriminated population 10,073 0.050 0.140 Freedom house score 7,858 7.438 4.108 Number of refugees in the country 4,510 92,412.700 State capacity (Hanson index) 7,601 −0.013 Female Labor Force (%) 4,413 Moving average of urban population (10-year) Pctl(25) 0.000 Pctl(75) 2.000 Max 3,774.000 0.000 1.000 13,965.000 26.018 66.773 100.000 964.665 14,827.550 1,424,548.000 256.166 2,874.420 18,311.240 187,942.900 −10.000 −7.000 8.000 10.000 0.000 0.000 0.016 0.980 2.000 3.000 11.000 14.000 354,395.300 1.000 243.250 36,871.750 4,404,995.000 0.994 −3.512 −0.773 0.648 2.862 51.334 15.991 8.026 42.242 61.228 90.770 12,143 46.748 24.665 1.882 26.508 66.567 100.000 Moving average of urban population (5-year) 13,128 46.628 24.845 1.796 26.273 66.750 100.000 Growth of urban population (proportion) 13,719 0.421 0.494 −24.580 0.120 0.640 3.900 Moving average of urban population growth (10-year) 11,946 0.423 0.411 −1.543 0.142 0.650 3.032 Moving average of urban population growth (5-year) 12,931 0.422 0.435 −4.612 0.130 0.644 3.292 M. SLAV ET AL. Table 2. Descriptive statistics of the variables. INTERNATIONAL INTERACTIONS 1115 Figure 1. Number of terrorist attacks (per 1 million people) per urbanization vigintile. We propose a sort of “competing for” pairs of hypotheses for both main predictors assuming a different (linear or nonlinear) relationship between urbanization and terrorist activity. To test them, we implement the following identification strategy. For the first hypothesis, for all combinations of response and predictor variables, we estimate, first, “linear3” relationship between an independent and dependent variable and a quadratic effect of the proportion of the urban population which is a way to directly model nonlinear relationship. For the second hypothesis, we estimate at first “linear” effect of the delta share of the urban population (proportion of the urban population included as a control) and then implement the interaction effect Table 3. Preliminary models on subsamples. Dependent Variable: Model: Vigintiles: Terrorist attacks (1) All (2) (1–9) (3) (10–14) (4) (15–20) −0.0095(0.0106) 0.0680***(0.0180) −0.0424(0.0419) −0.0586**(0.0242) Country Yes Yes Yes Yes Year Yes Yes Yes Yes 4,304 2,088 1,016 1,171 Variables Proportion of urban population Additional controls included Fixed-effects Fit statistics Observations Squared Correlation 0.44046 0.42870 0.22584 0.57513 Pseudo R2 0.18336 0.21007 0.21127 0.21683 BIC 20,358.1 8,632.0 5,140.1 6,557.3 Over-dispersion 0.61324 0.65918 0.81345 1.1561 Heteroskedasticity-robust standard-errors in parentheses. Signif. Codes: ***: 0.01, **: 0.05, *: 0.1. 3 As we use negative binomial regression, “linear” stands for the formula in the regression without quadratic or interaction effects. 1116 M. SLAV ET AL. between the delta share of the urban population and the share of the urban population itself. This strategy allows us to model different effects of urbanization growth at different levels of urbanization. Results This section presents the results of the baseline negative binomial regression models for all the proposed hypotheses. We also present online supplementary materials with robust checks that provide additional evidence. Urbanization and Terrorism Table 4 presents the results for Hypothesis 1.1 considering a simple linear relationship between the urbanization level (measured as the proportion of the urban population in the total population of the country) and terrorist activity. All three predictors: a simple share of urban population and five- as well as 10year rolling average – appear to be either negative insignificant or negative marginally significant (p-value between 0.05 and 0.1) predictors of the number of terrorist attacks. In addition, the already mentioned measures of urbanization are statistically insignificantly (positively) related to the number of terrorist killings. Thus, in five out of six tests the correlation has turned out to be insignificant and in the sixth test (with the number of terror attacks regressed against the 10-year rolling average) the main independent variable appears to be marginally significant, but in the direction opposite to the predicted one. Thus, our tests unequivocally reject Hypothesis 1.1. Note that all the models include two-way fixed effects on country and year and use heteroskedasticity robust standard errors to determine the coefficient significance. The next Table 5 shows the results for Hypothesis 1.2. The set of models uses a quadratic term for all the predictors for both response variables. Coefficients for the share of urban population and the squared share of urban population are positive and significant and negative significant, respectively, at 0.01 significance level. The trend remains for all models (that is for all combinations of predictor and response variable). This means that the relationship between the share of urban population and terror attacks and killings follows a reversed U-shaped curve: the sign is positive at lower and becomes negative at higher stages of urbanization. Figure 2 presents the fitted values of the model normalized on the total population of the country dependent on the proportion of urban population. The trend line in the plot was drawn by LOESS estimate of the predicted values of terror attacks on the proportion of urban population. The visualization illustrates a quadratic relationship with peak value of predicted terror attacks around the proportion of urban population equal 55% of total population of the country. The result proves the hypothesis and, thus, we Table 4. Proportion of urban population. Dependent Variables: Model: Variables Terrorist attacks (1) Terrorist killings (2) (3) (4) (5) (6) Proportion of urban population −0.0095(0.0106) Log of population 3.667***(0.3657) 3.759***(0.3730) 3.827***(0.3813) 4.553***(0.5939) 4.675***(0.6024) 4.822***(0.6098) Log of GDP per capita (PPP) 0.5016***(0.1712) 0.5142***(0.1714) 0.5123***(0.1715) 0.2762(0.2597) 0.3067(0.2590) 0.3253(0.2581) Polity score 0.0259***(0.0077) 0.0266***(0.0077) 0.0267***(0.0077) 0.0004(0.0110) 0.0016(0.0110) 0.0029(0.0109) Occupied country 1.128***(0.2943) 1.133***(0.2942) 1.130***(0.2938) 1.872***(0.5463) 1.887***(0.5465) 1.894***(0.5468) Proportion of discriminated population 0.8862**(0.3459) 0.8862**(0.3443) 0.8930***(0.3437) 0.6835(0.5668) 0.6825(0.5662) 0.6790(0.5641) War 1.069***(0.0852) 1.073***(0.0851) 1.073***(0.0852) 1.901***(0.1377) 1.913***(0.1376) 1.919***(0.1376) Moving average of urban population proportion (5-year) 0.0222(0.0167) −0.0150(0.0107) Moving average of urban population proportion (10-year) 0.0122(0.0170) −0.0186*(0.0109) 0.0005(0.0172) Fixed-effects Yes Yes Yes Yes Yes Yes Year Yes Yes Yes Yes Yes Yes 4,304 4,302 4,297 4,203 4,201 4,196 Fit statistics Observations Squared Correlation 0.44046 0.45902 0.46744 0.15914 0.19189 0.18746 Pseudo R2 0.18336 0.18361 0.18375 0.12350 0.12355 0.12356 BIC 20,358.1 20,332.5 20,274.5 19,415.4 19,389.1 19,318.5 Over-dispersion 0.61324 0.61412 0.61339 0.20891 0.20866 0.20787 Heteroskedasticity-robust standard-errors in parentheses Signif. Codes: ***: 0.01, **: 0.05, *: 0.1 INTERNATIONAL INTERACTIONS Country 1117 1118 Table 5. Quadratic effect of proportion of urban population. Dependent Variables: Terrorist attacks (1) (2) Terrorist killings (3) (4) Proportion of urban population 0.2454*** (0.0225) 0.2980***(0.0369) Proportion of urban population2 −0.0025*** (0.0002) −0.0027*** (0.0003) (5) (6) 3.383***(0.6508) Log of population 2.558***(0.3617) 2.632***(0.3708) 2.663***(0.3823) 3.070***(0.6271) 3.220***(0.6394) Log of GDP per capita (PPP) 0.2680*(0.1585) 0.2894*(0.1593) 0.3188**(0.1606) 0.0222(0.2360) 0.0802(0.2388) 0.1517(0.2431) Polity score 0.0243***(0.0076) 0.0244***(0.0076) 0.0241***(0.0076) −0.0013(0.0113) −0.0004(0.0113) 0.0010(0.0112) Occupied country 0.9379***(0.2773) 0.9364***(0.2804) 0.9339***(0.2831) 1.689***(0.5427) 1.722***(0.5484) 1.738***(0.5523) Proportion of discriminated population 0.9677***(0.3490) 1.009***(0.3547) 1.032***(0.3552) 0.7066(0.5827) 0.7645(0.5971) 0.7909(0.5972) War 1.112***(0.0828) 1.112***(0.0836) 1.108***(0.0843) 1.936***(0.1373) 1.944***(0.1384) 1.950***(0.1395) Moving average of urban population proportion (5-year) Moving average of urban population proportion (5-year) 0.2218***(0.0231) 2 0.2627***(0.0364) −0.0023***(0.0002) −0.0025***(0.0003) Moving average of urban population proportion (10-year) 0.2008***(0.0242) 0.2286***(0.0365) Moving average of urban population proportion (10-year)2 −0.0021***(0.0002) −0.0023***(0.0003) Fixed effects: Country Yes Yes Yes Yes Yes Yes Year Yes Yes Yes Yes Yes Yes 4,304 4,302 4,297 4,203 4,201 4,196 Fit statistics Observations Squared Correlation 0.45992 0.47285 0.47759 0.15214 0.16464 0.15013 Pseudo R2 0.19215 0.19127 0.19033 0.12788 0.12732 0.12683 BIC 20,164.5 20,164.9 20,132.1 19,334.3 19,320.5 19,260.6 Over-dispersion 0.67207 0.66617 0.65859 0.21729 0.21585 0.21415 Heteroskedasticity-robust standard-errors in parentheses Signif. Codes: ***: 0.01, **: 0.05, *: 0.1 M. SLAV ET AL. Model: Variables INTERNATIONAL INTERACTIONS 1119 Figure 2. Predicted number of terrorist attacks (normalized on total population). conclude that the relationship between urbanization and terrorist activity should be described as quadratic rather than a simple linear relationship. Urbanization Growth and Terrorism The results for Hypothesis 2.1 are shown in Table 6. Growth of urban population in a given country-year is a significant positive predictor at 0.01 significance level for both number of terrorist attacks and terrorist killings. However, the significance drops as the interval for average widens to 10 years (yet the signs remain positive). Also note that the share of urban population is mostly insignificant and even becomes negative for number of terrorist attacks. Concerning the decline in significance of wider rolling average measures a possible explanation is that the terrorist activity is dispersed with time; thus, the 10-year rolling average is measuring too general trend to predict such varying phenomenon. Finally, Table 7 presents the results for the last hypothesis (Hypothesis 2.2) assuming the joint effect of urbanization growth and level. Both the urbanization growth (including rolling averages) and the interaction term are significant at the 0.01 significance level. While the coefficient for growth is positive, the interaction effect has a negative sign which means that the effect of the share of urban population growth becomes negative as the share of urban population increases and the strength of the relationship weakens. Next Figure 3 performs fitted values for the interaction term model. To plot the effect of urbanization growth conditional on the level of urbanization, we divided the sample into two parts. The breaking point was chosen such that after this level of urbanization effect of growth becomes negative according 1120 Table 6. Growth of proportion of urban population. Dependent Variables: Growth of urban population (proportion) Terrorist attacks (1) (2) Terrorist killings (3) 0.3903***(0.1151) Proportion of urban population −0.0135(0.0106) Log of population 3.858***(0.3806) (4) (5) (6) 4.902***(0.6251) 4.887***(0.6210) 0.6659***(0.1990) 0.0149(0.0163) 3.895***(0.3866) 3.860***(0.3872) 4.785***(0.6190) log of GDP per capita (PPP) 0.4833***(0.1703) 0.4821***(0.1705) 0.4935***(0.1707) 0.1763(0.2569) 0.1364(0.2564) 0.1665(0.2558) Polity score 0.0270***(0.0078) 0.0263***(0.0078) 0.0260***(0.0078) 0.0018(0.0111) −2.45 × 10−5(0.0111) −0.0010(0.0111) 1.827***(0.5550) Occupied country 1.109***(0.2920) 1.105***(0.2939) 1.112***(0.2943) 1.803***(0.5498) 1.805***(0.5536) Proportion of discriminated population 0.8953***(0.3352) 0.9071***(0.3390) 0.9204***(0.3449) 0.6183(0.5457) 0.6533(0.5511) 0.7620(0.5675) War 1.058***(0.0845) 1.061***(0.0847) 1.067***(0.0853) 1.874***(0.1365) 1.866***(0.1369) 1.882***(0.1376) Moving average of urban population growth (5-year) 0.3349***(0.1233) Moving average of urban population proportion (5-year) 0.8358***(0.2025) −0.0173(0.0107) Moving average of urban population growth (10-year) Moving average of urban population proportion (10-year) 0.0080(0.0164) 0.1601(0.1323) 0.6804***(0.2250) −0.0192*(0.0109) −0.0005(0.0167) Fixed-effects Country Yes Yes Yes Yes Yes Yes Year Yes Yes Yes Yes Yes Yes Fit statistics Observations 4,304 4,301 4,296 4,203 4,200 4,195 Squared Correlation 0.44100 0.47048 0.47082 0.14441 0.15552 0.15757 Pseudo R2 0.18389 0.18401 0.18379 0.12410 0.12439 0.12402 BIC 20,354.3 20,320.7 20,268.9 19,411.5 19,366.6 19,302.6 Over-dispersion 0.61653 0.61668 0.61392 0.21008 0.20997 0.20854 Heteroskedasticity-robust standard-errors in parentheses Signif. Codes: ***: 0.01, **: 0.05, *: 0.1 M. SLAV ET AL. Model: Variables Table 7. Interaction between proportion of urban population and urban population growth. Dependent Variables: Model: Terrorist attacks -1 -2 Terrorist killings -3 -4 -5 -6 4.918*** (0.6208) Variables Growth of urban population (proportion) Proportion of urban population 1.561*** (0.2680) 1.998*** (0.4370) 0.0044 (0.0112) 0.0425* (0.0188) Log of population 3.891*** (0.3691) 3.943*** (0.3745) 3.959*** (0.3810) 4.902*** (0.6143) 4.951*** (0.6198) log of GDP per capita (PPP) 0.4782** (0.1663) 0.4431** (0.1662) 0.4262* (0.1666) 0.1989 (0.2490) 0.0877 (0.2422) 0.0772 (0.2445) Polity score 0.0267*** (0.0078) 0.0262*** (0.0077) 0.0257*** (0.0077) 0.0007 (0.0111) -0.0011 (0.0111) -0.0025 (0.0111) Occupied country 1.162*** (0.2953) 1.191*** (0.3042) 1.197*** (0.3083) 1.796*** (0.5324) 1.843*** (0.5353) 1.878*** (0.5432) Proportion of discriminated population 0.7802* (0.3248) 0.7363* (0.3305) 0.8625* (0.3418) 0.4048 (0.5151) 0.4342 (0.5229) 0.7251 (0.5609) War 1.051*** (0.0827) 1.055*** (0.0827) 1.064*** (0.0840) 1.861*** (0.1345) 1.855*** (0.1344) 1.871*** (0.1362) Growth of urban population (proportion) × Proportion of urban population -0.0245*** (0.0049) Moving average of urban population growth (5-year) -0.0289** (0.0089) 1.816*** (0.2862) Moving average of urban population proportion (5-year) Moving average of urban population growth (5-year) × Moving average of urban population proportion (5-year) 2.630*** (0.4669) 0.0052 (0.0114) 0.0414* (0.0189) -0.0309*** (0.0052) -0.0382*** (0.0093) Moving average of urban population growth (10-year) 1.608*** (0.3298) Moving average of urban population growth (10-year) × Moving average of urban population proportion (10-year) 2.200*** (0.5468) 0.0019 (0.0116) 0.0270 (0.0195) -0.0293*** (0.0058) -0.0313** (0.0104) Fixed-effects Country Yes Yes Yes Yes Yes Yes Year Yes Yes Yes Yes Yes Yes Fit statistics Observations 4304 4301 4296 4203 4,2 4195 Squared Correlation 0.43786 0.46273 0.45887 0.15391 0.18151 0.17599 Pseudo R2 0.18493 0.1854 0.18478 0.12464 0.12515 0.12446 BIC 20338.9 20297.2 20254.5 19408.7 19359.4 19302 Over-dispersion 0.62187 0.6241 0.6196 0.2111 0.2112 0.20916 INTERNATIONAL INTERACTIONS Moving average of urban population proportion (10-year) 1121 1122 M. SLAV ET AL. Figure 3. Predicted number of terrorist attacks (normalized on total population). to the model estimates4 (see model 1 from Table 7). The left subplot stands for the trend before the breaking point while the right does for the one after. The trend on the plot was drawn by LOESS estimate of fitted values from a regression model on the growth of proportion of the urban population. The graph indicates that the effect of the growth of the share of urban population varies with the proportion of urban population itself, as it is positive at the lower stages and negative at the higher stages of urbanization. Thus, we have gained evidence in favor of this hypothesis and assume that the effect of urbanization growth is conditional on urbanization itself. In other words, at lower levels of urbanization, its growth is associated with a higher impact on terrorist activity than at higher levels. Discussion In this section, the results from the previous two sections are interpreted in a broader context and with more specific evidence. We illustrate our findings with several case-study examples. The majority of cases supporting the observed relationship for lower urban vigintiles predominantly include countries in sub-Saharan Africa and South and Southeast Asia (see the full list of observations in Table 4 of supplementary online materials). As the Results section shows, we see a significant positive relationship between urbanization and terror attacks in countries where less than half of the population lives in cities.5 This evidence is consistent with both historical examples and the theoretical framework. It corresponds with the baseline 4 The breaking point could be calculated from partial derivative of regression equation on growth of proportion of urban population, it is at the 63,8% urbanization rate. 5 See Table 5. INTERNATIONAL INTERACTIONS 1123 Huntington hypothesis that the early stages of modernization are marked by destabilization and an increase in violence. Latin American countries within the first 10 vigintiles of urbanization are in line with the drastic overall population growth in Latin America (Bongaarts 2009) from 0.17 to 0.56 billion. We observe an increase in the likelihood of terror attacks for countries in Central America and the Caribbean region in the second half of the twentiet century. El Salvador demonstrates a significant increase in urbanization from the end of the 1970s up to the early 1990s, with the share of urban population growing from 43% to 50%. These years coincide with the rise in terrorism and civil war in El Salvador – a marker of an overall destabilization process. This case corresponds with our expectations of an increase in terrorism at the end of the first phase of demographic transition, as long as fertility decline for El Salvador is estimated to have started in the mid1960s (Reher 2004). The same link is observed for Nicaragua in 1978–1983, and it corresponds with the end of the Nicaraguan revolution in 1979 and precedes the period of armed conflict in the 1980s, which, again, confirms our hypothesis on the relationship between urbanization and destabilization. The same is true for Honduras in the last 20 years of the twentieth century, with an increase of urban population ratio from 36% to 43% and a significant rise in terrorism. However, it is important to mention that to account for instability in Central America in the 1980s and 1990s one should take into consideration some other factors other than urbanization, which, of course, requires further analysis. Concerning the Middle East and North Africa rregions we see evidence for the established relationship for Turkey in the 1970s, Iran and Syria around the start of the 1980s, and contemporary Egypt and Yemen. In Egypt, we observe a sharp increase in urban population ratio after 2009. This trend is concurrent with Hezbollah, Khan el-Khalili, and Cairo terror attacks in 2009 together with numerous terror attacks in 2012–2017 following the Arab Spring. Concerning the latter half of the observed link between urbanization and terrorism, we propose that, starting from a certain level of urbanization, the government becomes able to regulate the society, as long as the concentration of population is high enough for officials to provide public goods effectively by using previously accumulated resources, such as education, social support programs, etc. It is important to note here that at this phase of modernization societies are more stable, and therefore urbanization leads to concentration of state control and safety. In contrast, at earlier stages of urbanization, the rise of urban population converts into “proletarianization” of cities, and the state, being itself in process of transformation, cannot keep up with the ongoing changes. This results in major social shocks and destabilization, leading to terrorism as a form of expression of social grievances. This was the case in Columbia in the 1990s, when the state apparatus was reformed in order to increase its effectiveness and capacity. Our data shows 1124 M. SLAV ET AL. that after 2000 the number of terrorist attacks per year did not exceed five, while in the previous two decades, starting from 1980s, most years the number of terror attacks varied between ten and fifteen attacks a year, which proves the effectiveness of state coercive powers. It can be noticed that this relationship is relevant for other Latin American countries, such as Brazil and Chile, too, since these countries experienced rather few terror attacks after state reforms in the late 1990s. Finally, we would like to draw the reader’s attention to a work by Sciubba (2012) as she points out that cities have become different from those of the nineteenth and twentieth centuries. Theoretically, this could change the link between urbanization and terrorism. She argues urbanization would intensify and urban density would increase, which in turn could exacerbate terrorist activity. However, due to the progress in communication technologies, urban residents would be able to sustain their family ties. We suppose this might bring dual results. Potentially, it could decrease the frustration of the firstgeneration urban resident. Alternatively, this might not force them to learn the art of association. In other words, even though we have confidence in the presented results, we advocate for renewed research on this topic in a decade. Conclusion In this paper, we have carried out the first extensive analysis of the influence of urbanization on the number of terrorist attack incidents and terrorist killings. We test different hypotheses on the influence. First, the literature on terrorists as rational actors suggests that there is a greater propensity to conduct a terrorist attack in a city compared to a rural area, both for media and logistical reasons. Based on this evidence, we test the hypothesis on the existence of the linear positive effect of urbanization on terror attacks (Hypothesis 1.1). Second, we observe that problems may arise from rapid modernization, specifically, from intensive rural–urban migration, which brings many individuals to live in social and living conditions they are not used to and do not fit in. This suggestion follows the more general prediction that nations in the process of modernization are less stable than modernized or non-modernized ones and a more complex supposition that modernization has a negative influence on stability in its beginning and positive influence in its end. In that connection, we also test the hypothesis of an increase of terrorist activity at lower levels of urbanization and a decrease at higher urbanization levels (Hypothesis 1.2). We have not found evidence to support Hypothesis 1.1 that the level of urbanization has a simple linear positive relationship with the terrorist activity for the whole range of societies at all phases of urbanization transition. INTERNATIONAL INTERACTIONS 1125 Consistently with the Hypothesis 1.2, we have found that an influx of rural residents to cities does have a consistent positive influence on the number of terrorist attacks at lower urbanization levels. We suppose that this is generally a result of frustration and lack of proper social networks of first-generation urban residents and a lack of proper living conditions in newly urbanized areas. As predicted by two other hypotheses, we have found that the influence of both urbanization and its tempo is inverted U-shaped, with the point of inflection being slightly higher than the median (the inflection point is estimated to be at 55% urbanization rate for the model with simple urbanization level influence and 63.8% urbanization rate for the model with the joint influence of urbanization level and its growth). Following theory and case studies, an increase in the number of first-generation urban residents increases the number of terrorist attacks in the first stages of urbanization and has the opposite effect in the late stages. We explain this curvilinearity with the Huntington hypothesis on the true cause of instability: the issue is not modernization, per se, but modernization combined with a lack of political institutions. This explanation is supported by the negative or declining effects of both the urbanization level and urbanization tempo in the second half of the urbanization process. The evidence we have found can be used extensively to further deepen our understanding of terrorism. We feel that the following research topics would be worthwhile for discussion on factors of terrorism. (1) Since the evidence for the negative correlation between terrorism and urbanization stems predominantly from the Latin American countries, they should be studied separately. Such an investigation would help to comprehend how modernization processes alter the nature of political violence, its actors, and determinants. (2) The nonlinear relationship we obtained suggests that modernization processes and terrorism share a complex relationship. To disentangle the effect of urbanization from other factors, it would be fruitful to look at the exogenous shocks in the urban population caused by crises in the agricultural sector. Acknowledgments This paper is an output of a research project implemented as part of the Basic Research Program at the HSE University in 2021 with support by the Russian Science Foundation (Project No. 18-1800254). We would like to thank Jeff Pickering, the International Interactions editorial team, and two anonymous referees for their valuable comments and suggestions. 1126 M. SLAV ET AL. ORCID Maxim Slav http://orcid.org/0000-0002-3061-1692 Elena Smyslovskikh http://orcid.org/0000-0003-2333-9413 Vladimir Novikov http://orcid.org/0000-0002-2636-043X Igor Kolesnikov http://orcid.org/0000-0002-9459-7359 Andrey Korotayev http://orcid.org/0000-0003-3014-2037 References Abadie, Alberto. 2006. “Poverty, Political Freedom, and the Roots of Terrorism.” American Economic Review 96 (2): 50–56. doi:10.1257/000282806777211847. Adelaja, Adesoji, and Justin George. 2019. “Grievances, Latent Anger and Unrest in Africa.” African Security 12 (1): 111–40. doi:10.1080/19392206.2019.1599641. Azam, Jean-Paul. 2012. “Why Suicide-Terrorists Get Educated, and What to Do about It.” Public Choice 153 (3–4): 357–73. doi:10.1007/s11127-011-9798-7. Beall, Jo. 2006. “Cities, Terrorism and Development.” Journal of International Development 18 (1): 105–20. doi:10.1002/jid.1265. Berge, Laurent. 2018. “Efficient Estimation of Maximum Likelihood Models with Multiple Fixed-effects: The R Package FENmlm.” CREA Discussion Papers, no. 13. Berrebi, Claude. 2007. “Evidence about the Link between Education, Poverty and Terrorism among Palestinians.” Peace Economics, Peace Science and Public Policy (Berlin, Boston) 13 (1): 18–53. doi:10.2202/1554-8597.1101. Bongaarts, John. 2009. “Human Population Growth and the Demographic Transition.” Philosophical Transactions of the Royal Society B: Biological Sciences 364 (1532): 2985–90. doi:10.1098/rstb.2009.0137. Brockhoff, Sarah, Tim Krieger, and Daniel Meierrieks . 2010. “Ties that Do Not Bind (Directly): The Education-Terrorism Nexus Revisited.” Working Papers CIE 26, Paderborn University, CIE Center for International Economics. Brockhoff, Sarah, Tim Krieger, and Daniel Meierrieks. 2015. “Great Expectations and Hard Times: The (Nontrivial) Impact of Education on Domestic Terrorism.” The Journal of Conflict Resolution 59 (7): 1186–215. doi:10.1177/0022002713520589. Buhaug, Halvard, and Henrik Urdal. 2013. “An Urbanization Bomb? Population Growth and Social Disorder in Cities.” Global Environmental Change 23 (1): 1–10. doi:10.1016/j. gloenvcha.2012.10.016. Burgoon, Brian. 2006. “On Welfare and Terror: Social Welfare Policies and Political-Economic Roots of Terrorism.” Journal of Conflict Resolution 50 (2): 176–203. doi:10.1177/ 0022002705284829. Campos, Nauro, and Martin Gassebner. 2013. “International Terrorism, Domestic Political Instability, and the Escalation Effect.” Economics & Politics 25 (1): 27–47. doi:10.1111/ ecpo.12002. Chenoweth, Erica. 2013. “Terrorism and Democracy.” Annual Review of Political Science 16:355–78. doi: 10.1146/annurev-polisci-032211-221825 Coward, Martin. 2009. “Network-Centric Violence, Critical Infrastructure and the Urbanization of Security.” Security Dialogue 40 (4–5): 399–418. doi:10.1177/ 0967010609342879. Crenshaw, Martha. 1981. “The Causes of Terrorism.” Comparative Politics 13 (4): 379–99. doi:10.2307/421717. INTERNATIONAL INTERACTIONS 1127 Croissant, Yves, and Giovanni Millo. 2008. “Panel Data Econometrics in R: The Plm Package.” Journal of Statistical Software 27 (2): 1–43. doi:10.18637/jss.v027.i02. Daniel, Meierrieks, and Gries Thomas. 2013. “Causality between Terrorism and Economic Growth.” Journal of Peace Research 50 (1): 91–104. doi:10.1177/0022343312445650. Danzell, Orlandrew E., Yao-Yuan Yeh, and Melia Pfannenstiel. 2020. “Does Education Mitigate Terrorism? Examining the Effects of Educated Youth Cohorts on Domestic Terror in Africa.” Terrorism and Political Violence 32 (8): 1731–52. doi:10.1080/ 09546553.2018.1506336. Bueno de Mesquita, Ethan Bueno. 2005. “The Quality of Terror.” American Journal of Political Science 49 (3): 515–30. doi:10.1111/j.1540-5907.2005.00139.x. Ding, M., and L. Ding. 2012. “Rural-urban Migration: Seeds of Strife in the People’s Republic of China.” In An Anthology of Contending Views on International Security, 57–71. Hauppauge, New York: Nova Science Douglas, Mary. 2012. How Institutions Think. Syracuse, New York: Routledge. 162. Dreher, Axel, and A.V. Fischer Justina. 2010. “Government Decentralization as a Disincentive for Transnational Terror? An Empirical Analysis.” International Economic Review 51 (4): 981–1002. doi:10.1111/j.1468-2354.2010.00608.x. Durkheim, Emile. 1964. The Division of Labor in Society. Translated by George Simpson. London: Free´ Press, Collier Macmillan. 439. Enders, Walter, and Gary A. Hoover. 2012. “The Nonlinear Relationship between Terrorism and Poverty.” American Economic Review 102 (3): 267–72. doi:10.1257/aer.102.3.267. Enders, Walter, Gary A. Hoover, and Todd Sandler. 2016. “The Changing Nonlinear Relationship between Income and Terrorism.” Journal of Conflict Resolution 60 (2): 195–225. doi:10.1177/0022002714535252. Eubank, William L., and Leonard Weinberg. 1994. “Does Democracy Encourage Terrorism?” Terrorism and Political Violence 6 (4): 417–35. doi:10.1080/09546559408427271. Eubank, William L., and Leonard Weinberg. 2001. “Terrorism and Democracy: Perpetrators and Victims.” Terrorism and Political Violence 13 (1): 155–64. doi:10.1080/ 09546550109609674. Eyerman, Joe. 1998. “Terrorism and Democratic States: Soft Targets or Accessible Systems.” International Interactions 24 (2): 151–70. doi:10.1080/03050629808434924. Fahey, Susan, and LaFree. Gary. 2015. “Does Country-Level Social Disorganization Increase Terrorist Attacks?” Terrorism and Political Violence 27 (1): 81–111. doi:10.1080/ 09546553.2014.972156. Freytag, Andreas, J. J. Krüger, D. Meierrieks, F. Schneider, et al. 2011. “The Origins of Terrorism: Cross-Country Estimates on Socio-economic Determinants of Terrorism.” European Journal of Political Economy 27:S5–S16. doi:10.1016/j.ejpoleco.2011.06.009. Gaibulloev, Khusrav, James A. Piazza, and Todd Sandler. 2017. “Regime Types and Terrorism.” International Organization 71 (3): 491–522. doi:10.1017/S0020818317000169. Gassebner, Martin, and Simon Luechinger. 2011. “Lock, Stock, and Barrel: A Comprehensive Assessment of the Determinants of Terror.” Public Choice 149 (3): 235–61. doi:10.1007/ s11127-011-9873-0. Geifman, Anna. 2005. “Psychohistorical Approaches to 1905 Radicalism.” In The Russian Revolution of 1905: Centenary Perspectives, edited by Jonathan D. Smele and Anthony Heywood (pp. 14–27). London: Routledge. Ghatak, Sambuddha, and Aaron Gold. 2017. “Development, Discrimination, and Domestic Terrorism: Looking beyond a Linear Relationship.” Conflict Management and Peace Science 34 (6): 618–39. doi:10.1177/0738894215608511. 1128 M. SLAV ET AL. Grinin, Leonid, and Andrey Korotayev. 2009. “Urbanization and Political Instability: To the Development of Mathematical Models of Political Processes.” Polis-Politicheskiye Issledovaniya. 1026–9487. 4:34–52. Hanson, Jonathan K, and Rachel Sigman. 2013. “Leviathan’s Latent Dimensions: Measuring State Capacity for Comparative Political Research.” In APSA 2011 Annual meeting paper. Washington. Huntington, Samuel P. 1968. Political Order in Changing Societies. Vol. xi. New Haven: Yale University Press. 488. ISBN: 03000117170300005849. Ibimilua, A.F. 2011. “The Nigerian National Housing Policy in Perspective: A Critical Analysis.” Journal of Social Development in Africa 26 (2): 165–88. Kahn, Josef, and Tim Weiner. 2002. “World Leaders Rethinking Strategy on Aid to Poor.” The NewYork Times, March 18. https://www.nytimes.com/2002/03/18/world/world-leadersrethinkingstrategy-on-aid-to-poor.html Kis-Katos, Krisztina, Helge Liebert, and Günther G. Schulze. 2011. “On the Origin of Domestic and International Terrorism.” European Journal of Political Economy 27:S17–S36. doi: 10.1016/j.ejpoleco.2011.02.002. Kornhauser, William. 1959. The Politics of Mass Society. Glencoe: The Free Press. 1st edited 256. ISBN: 9781412807722. Korotayev, Andrey, Sergey Malkov, and Leonid Grinin. 2014. “A Trap at the Escape from the Trap? Some Demographic Structural Factors of Political Instability in Modernizing Social Systems.” In History Mathematics: Trends and Cycles, edited by Grinin Leonid and Korotayev Andrey, 201–67, Uchitel. ISBN: 978-5-7057-4223-3. Volgograd Korotayev, Andrey, Ilya Vaskin, and Stanislav Bilyuga. 2017. “Olson-Huntington Hypothesis on A Bell-Shaped Relationship between the Level of Economic Development and Sociopolitical Destabilization: A Quantitative Analysis.” Sociologiceskoe Obozrenie 16 (1): 9–49. doi:10.17323/1728-192X2017-1-9-49. Korotayev, Andrey, Ilya Vaskin, Stanislav Bilyuga, and Ilya Ilyin. 2018. “Economic Development and Sociopolitical Destabilization: A Re-Analysis.” Cliodynamics 9 (1): 59–118. doi:10.21237/C7clio9137314. Korotayev, Andrey, Ilya Vaskin, and Daniil Romanov. 2019. “Democracy and Terrorism: A Re-analysis.” Sotsiologicheskoe Obozrenie / Russian Sociological Review 18 (3): 9–48. doi:10.17323/1728-192X-2019-3-9-48. Korotayev, Andrey, Ilya Vaskin, and Sergey Tsirel. 2019. “Economic Growth, Education, and Terrorism: A Re-Analysis.” Terrorism and Political Violence 33 (3): 572–595. doi: 10.1080/ 09546553.2018.1559835. Korotayev, Andrey, Julia Zinkina, Svetlana Kobzeva, Justislav Bozhevolnov, Daria Khaltourina, Artemy Malkov, and Sergey Malkov. 2011. “A Trap at the Escape from the Trap? Demographic-structural Factors of Political Instability in Modern Africa and West Asia.” Cliodynamics 2 (2): 276–303. doi:10.21237/C7clio22217. Krueger, Alan B., and Jitka Maleckova. 2003. “Education, Poverty and Terrorism: Is There a Causal Connection?” Journal of Economic Perspectives 17 (4): 119–44. doi:10.1257/ 089533003772034925 LaFree, Gary, and Laura Dugan. 2007. “Introducing the Global Terrorism Data Base.” Terrorism and Political Violence 19 (2): 181–204. doi:10.1080/09546550701246817. Lai, Brian. 2007. ““Draining the Swamp”: An Empirical Examination of the Production of International Terrorism, 1968—1998.” Conflict Management and Peace Science 24 (4): 297–310. doi:10.1080/07388940701643649. Marineau, Josiah, H. Pascoe, A. Braithwaite, M. Findley, J. Young, et al. 2020. “The Local Geography of Transnational Terrorism.” Conflict Management and Peace Science 37 (3): 350–81. doi:10.1177/0738894218789356. INTERNATIONAL INTERACTIONS 1129 Mccartan, Lisa M., A. Masselli, M. Rey, D. Rusnak, et al. 2008. “The Logic of Terrorist Target Choice: An Examination of Chechen Rebel Bombings from 1997–2003.” Studies in Conflict & Terrorism 31 (1): 60–79. doi:10.1080/10576100701767122. Ness, Cindy D., editor. 2007. Female Terrorism and Militancy. Agency, Utility, and Organization. London: Routledge. ISBN: 978–0–415–77347–8. Newman, Edward. 2006. “Exploring the “Root Causes” of Terrorism.” Studies in Conflict & Terrorism 29 (8): 749–72. doi:10.1080/10576100600704069. Okafor, Godwin, and Jenifer Piesse. 2018. “Empirical Investigation into the Determinants of Terrorism: Evidence from Fragile States.” Defence and Peace Economics 29 (6): 697–711. Østby, Gudrun. 2016. “Rural–urban Migration, Inequality and Urban Social Disorder: Evidence from African and Asian Cities.” Conflict Management and Peace Science 33 (5): 491–515. 07388942, 15499219. doi:10.1177/0738894215581315. Piazza, James A. 2006. “Rooted in Poverty?: Terrorism, Poor Economic Development, and Social Cleavages 1.” Terrorism and Political Violence 18 (1): 159–177. doi:10.1080/ 095465590944578. Piazza, James A. 2009. “Economic Development, Poorly Managed Political Conflict and Terrorism in India.” Studies in Conflict & Terrorism 32 (5): 406–19. doi:10.1080/ 10576100902831552. Python, Andr´e, Janine B. Illian, Charlotte M. Jones-Todd, and Marta Blangiardo. 2019a. “A Bayesian Approach to Modelling Subnational Spatial Dynamics of Worldwide Nonstate Terrorism, 2010–2016.” Journal of the Royal Statistical Society. Series A: Statistics in Society 182 (1): 323–44. doi:10.1111/rssa.12384. Python, Andr´e, Janine B. Illian, Charlotte M. Jones-Todd, and Marta Blangiardo. 2019b. “The Deadly Facets of Terrorism.” Significance 16 (4): 28–31. doi:10.1111/j.17409713.2019.01300.x Reher, David S. 2004. “The Demographic Transition Revisited as a Global Process.” Population, Space and Place 10 (1): 19–41. doi:10.1002/psp.313. Sanso-Navarro, Marcos, and Marıa Vera-Cabello. 2020. “The Socioeconomic Determinants of Terrorism: A Bayesian Model Averaging Approach.” Defence and Peace Economics 31 (3): 269–88. doi:10.1080/10242694.2018.1525935. Savitch, H.V., and Grigoriy Ardashev. 2001. “Does Terror Have an Urban Future?” Urban Studies 38 (13): 2515–33. doi:10.1080/00420980120107529. Sayari, Sabri, and Bruce Hoffman. 1994. “Urbanisation and Insurgency: The Turkish Case, 1976–1980.” Small Wars & Insurgencies 5 (2): 162–79. doi:10.1080/09592319408423077. Schmid, Alex P. 1992. “Terrorism and Democracy.” Terrorism and Political Violence 4 (4): 14–25. doi:10.1080/09546559208427173. Sciubba, Jennifer Dabbs. 2012. “Demography and Instability in the Developing World.” Orbis 56 (2): 267–77. doi:10.1016/j.orbis.2012.01.009. Seung-Whan, Choi. 2015. “Economic Growth and Terrorism: Domestic, International, and Suicide.” Oxford Economic Papers 67 (1): 157. doi:10.1093/oep/gpu039. Skoczylis, Joshua, and Sam Andrews. 2020. “A Conceptual Critique of Prevent: Can Prevent Be Saved? No, But. . ..” Critical Social Policy 40 (3): 350–69. doi:10.1177/0261018319840145. Slinko, Elena, Stanislav Bilyuga, Julia Zinkina, and Andrey Korotayev. 2017. “Regime Type and Political Destabilization in Cross-National Perspective: A Re-Analysis.” Cross-Cultural Research 51 (1): 26–50. doi:10.1177/1069397116676485. Tavares, Jos´e. 2004. “The Open Society Assesses Its Enemies: Shocks, Disasters and Terrorist Attacks.” Journal of Monetary Economics 51 (5): 1039–70. doi:10.1016/j. jmoneco.2004.04.009. Tilly, Charles. 1992. Coercion, Capital, and European States, AD 990–1992. Oxford: Blackwell Oxford. 1130 M. SLAV ET AL. Vaskin, Ilya, Andrey Korotayev, and Sergey Tsirel. 2018. “Economic Development, Education, and Terrorism: A Quantitative Analysis.” Sotsiologicheskiy Zhurnal 24 (2): 28–65. doi:10.19181/socjour.2018.24.2.5844. William, Cooper, and Yue Piyu. 2008. “Challenges of the Muslim World: Rapid Urbanization and Its Problems.” International Symposia in Economic Theory and Econometrics 19:161–71. Zeileis, Achim, Christian Kleiber, and Simon Jackman. 2008. “Regression Models for Count Data in R.” Journal of Statistical Software 27 (8): 1–25. doi:10.18637/jss.v027.i08.