Terrorist Perpetrators, Unlike Common Criminals, Do Not See Themselves as Criminal

First, although common criminals vary widely in terms of how they perceive their activities (cf., Reference KatzKatz, 1988; Reference BlackBlack, 1998; Reference AndersonAnderson, 1999), it is safe to say that few criminals see themselves as altruists. By contrast, many terrorist perpetrators see themselves not as criminals but as individuals making sacrifices for a noble cause (Reference Pedahzur, Perliger and WeinbergPedahzur, Perliger & Weinberg, 2003; Reference HafezHafez, 2006). Indeed, many members of the most prominent terrorist groups in the world – including the Islamic State, al Qaeda, Shining Path, Euskadi ta Askatasuna or Basque Homeland and Freedom (ETA), the Irish Republican Army (IRA), the LTTE and the FARC – often conceive of themselves as freedom fighters and positive agents of change.

Lack of Traditional Criminology Data on Terrorism

Second, although data on crime are far from perfect, criminologists have traditionally had three major options for studying criminal behavior, corresponding to the major social roles connected to criminal events: “official” data collected by legal agents, especially the police; “victimization” survey data collected from the general population, which include both crime victims and nonvictims; and “self-report” survey data collected from offenders. However, all three of these sources are problematic when it comes to gathering data on terrorism (Reference LaFree and DuganLaFree & Dugan, 2004). Few countries develop systematic data on terrorism-related crimes and certainly no worldwide official police data on terrorism exists. Indeed, thus far, the United Nations (UN) has been unable to provide a definition of terrorism that is accepted by all member nations. Police departments in most countries do not maintain separate records for terrorism-related offenses, and much primary data collected by intelligence agents are not available to researchers working in an unclassified environment. Nor are data collection challenges confined to the police. Most individuals convicted of behavior that would be widely regarded as terrorism do not show up in court records as terrorist perpetrators because they are convicted not of terrorism but of more common crimes connected to terrorist behavior, like weapons violations and homicide.

Official data on terrorism in the United States provides an example. The UCR does not include statistics on terrorist attacks. Following the passage of a National Defense Authorization Act, starting in 2015, the FBI (in cooperation with the Department of Homeland Security and the Director of National Intelligence) began reporting data on domestic terrorism (FBI, 2021). The annual report also includes intelligence assessments, a discussion of investigative activities and a list of recommendations. However, this reporting system is limited to arrests, uses a different methodology than the UCR data collection for ordinary offenses and is not integrated with the UCR system.

Starting in 1973, each year the National Crime Victimization Survey (NCVS), administered by the US Census Bureau, has interviewed approximately 240,000 persons in 150,000 households about their experience as crime victims, including assault, burglary, larceny, motor vehicle theft, rape and robbery (Bureau of Justice Statistics, 2022). However, victimization surveys have been of little use in providing statistical evidence on the characteristics of terrorist attacks or its perpetrators. Despite the attention it gets in the media, terrorist attacks are much less common than more ordinary types of violent crime and thus, even with extremely large sample sizes, few individuals in most countries will have been victimized by terrorists. Indeed, victims of terrorism often have no direct contact with perpetrators (e.g., in many bombings), and in too many cases, terrorism victims are killed by their attackers, leaving no one to survey.

A final option for crime data in general is the self-report survey, where individuals are asked to describe their participation in past criminal behavior (Reference Hindelang, Hirschi and WeisHindelang, Hirschi & Weis, 1979). Self-report surveys in criminology have been especially useful for either studying minor crimes or reporting on criminal behavior when respondents were juveniles (Reference Huizinga and ElliottHuizinga & Elliott, 1986; Farrington, Ohlin & Wilson, 1998). Self-report surveys based on interviews with current or former terrorist perpetrators have also provided some excellent scholarship (Reference McCauley and StoutMcCauley, 2002; Reference HorganHorgan, 2004). However, most active terrorists are unwilling to participate in interviews, and even when they are willing, doing interviews with them raises obvious challenges. Terrorism researcher Ariel Reference MerariMerari (1991: 90) described the problem succinctly: “The clandestine nature of terrorist organizations and the ways and means by which intelligence can be obtained will rarely enable data collection which meets commonly accepted academic standards.” Hence, self-report surveys have been of little use in providing national let alone worldwide statistics on terrorist attacks or the characteristics of perpetrators.

Common Crimes Are Usually Local; Terrorism Often National or International

Third, for most ordinary crimes, criminal justice decisions are made locally and rarely gather international or even national attention. By contrast, terrorist attacks are frequently reported outside of the local area where they occur and often gather worldwide coverage. For example, incidents like the January 2015 attack on the satirical newspaper Charlie Hebdo in Paris, or the April 2013 attack on the finish line of the Boston Marathon create enormous media attention. A Google search of the latter while this Element was being prepared yielded sixty-four million hits. However, this distinction does not mean that terrorist perpetrators are always unmindful of local targets. For example, as we shall see later, there is evidence that terrorist perpetrators, like ordinary offenders, often choose targets in familiar locations (Reference Hasisi, Perry, Ilan and WolfowiczHasisi et al. 2020b).

Common Criminals Seek Anonymity; Terrorist Perpetrators Seek Media Attention

Fourth, although ordinary criminals are usually struggling to avoid detection much less an audience, a large audience is precisely what many terrorist perpetrators are seeking. A half-century ago, terrorism researcher Brian Jenkins noted that (Reference Jenkins1975: 15), “terrorists want a lot of people watching and a lot of people listening, and not a lot of people dead.” Reference HoffmanHoffman (1998: 131) argues that terrorist groups seek maximum publicity for their actions, and because of this fact, getting picked up by the print, and increasingly, the electronic media, is critical to their perceived success. Because a common goal of terrorism is to gain media attention, terrorist attacks are often carefully orchestrated. This is much less common with ordinary crimes.

Terrorism, Unlike Common Crime, Is Often a Means to Broader Goals

Finally, the goal of most ordinary crimes is to obtain a particular material reward, such as money or valued goods, or to kill or injure a specific victim. By contrast, the overriding objective of terrorism and its ultimate justification is to further a political cause. Thus, criminals often have selfish, personal motives, and their actions are not intended to have consequences or create psychological or political repercussions beyond the criminal act. By contrast, the fundamental aim of terrorist perpetrators is often to overthrow or change the dominant political system. Terrorism expert Martha Reference CrenshawCrenshaw (1983: 2) points out that “the intent of terrorist violence is psychological and symbolic, not material.” This conclusion was well supported in a recent study with my colleagues (Reference Becker, Decker and LaFreeBecker et al., 2022) in which we examined forty-five US gang members and thirty-eight US political extremists. Compared to gang members, extremists were far more likely to be motivated by the perceived moral authority of their actions. By contrast, we found that gang members were more likely to cite material rewards or group prestige as their main motive for criminal involvement. At the same time, it is clear that the motives of political extremists are not universally nonmaterial. For example, in a study of suicide bombers, Reference Perry and HasisiPerry and Hasisi (2015) find that perpetrators frequently consider self-gratifying benefits in making the decision to launch attacks.

Situational Criminology and the Routine Activities Theory

According to Table 1, the most common theoretical perspective in terms of the total times it appeared in the articles we reviewed was situational/routine activities theory. While some criminological theories focus on individual-level variables that contribute to radicalization, a situational approach shifts the attention away from the offender and onto the crime itself (Reference ClarkeClarke, 1980; Reference Clarke1995). Routine activities theory is a situational approach that focuses on both the situational determinants (i.e., the suitability of targets and the strength of guardianship) as well as the motivation of criminal offenders (Reference Cohen and FelsonCohen & Felson, 1979). However, most of the applications of situational perspectives to terrorist perpetrators, like much of the research on crime in general, has emphasized situational determinants more than the motivation of offenders. For example, recent criminology articles that have relied on situational or routine activities perspectives to understand terrorism include studies of terrorist attack “hot spots” (Reference Hasisi, Carmel and WolfowiczHasisi, Carmel & Wolfowicz, 2020a), terrorist target characteristics (Reference Sturup, Gerell and RostamiSturup, Gerell & Rostami, 2020) and situational crime prevention (SCP) applied to terrorist attacks (e.g., Reference Perry, Apel, Newman and ClarkePerry et al., 2017).

In support of earlier studies of common crimes (Reference Sherman, Gartin and BuergerSherman, Gartin & Buerger, 1989), Reference PerryPerry (2020) found that a large proportion of the total number of terrorist attacks in the city of Jerusalem were concentrated in a relatively small number of geographic spaces: one-quarter of all attacks occurred in less than 2 percent of all attack locations. This finding has important policy implications as it suggests that terrorism, like more ordinary crimes, does not occur randomly but rather is highly concentrated in relatively well-defined spaces. This insight raises the possibility that societies can deploy police and other preventative measures more effectively to reduce the number of successful terrorist attacks.

Closely related to the situational concept of hot spots, is the idea of “near repeats” – the finding that once criminal offenders select a particular time and place for a crime, if successful, they are likely to continue offending at the same or nearby locations (Reference Townsley, Homel and ChaselingTownsley, Homel & Chaseling, 2003; Reference Bowers and JohnsonBowers & Johnson, 2004). Building on the possibility that terrorist attacks will also follow near-repeat patterns, Reference Behlendorf, LaFree and LegaultBehlendorf, LaFree and Legault (2012) propose a method for identifying and analyzing what they identify as violent microcycles; groups of events that take place close to each other in both space and time. They use the Global Terrorism Database (GTD) to analyze 3,335 terrorist attacks attributed to the Farabundo Martí National Liberation Front (FMLN) in El Salvador and 1,993 terrorist attacks attributed to ETA in Spain – two terrorist organizations that were both extremely active and violent but differed greatly in terms of history, grievances and motives. They find strong support for the conclusion that many of the terrorist attacks attributed to these two distinctive groups were part of violent microcycles and that the spatiotemporal attack patterns of these two groups exhibit substantial similarities. Their analysis shows that for both the FMLN and ETA, compared to other tactics used by terrorists, bombings and nonlethal attacks were more likely to be part of microcycles, and that compared to attacks that occurred elsewhere, attacks aimed at national or provincial capitals or areas of specific strategic interest to the terrorist organization were more likely to be part of microcycles.

Closely linked to the situational approach, Reference Cohen and FelsonCohen and Felson’s (1979) routine activity theory suggests that crime occurs when suitable targets, motivated offenders and the absence of capable guardians converge in time and space. Several recent terrorism studies have focused on soft and hard target characteristics along with security-related variables. For example, Reference Fahey, LaFree, Dugan and PiqueroFahey et al. (2012) examined whether terrorist hijackings situationally differed from nonideologically motivated aerial hijackings. By examining contextual variables including number of hijackers, day of the week and departure country, the authors found that measures of organizational resources (e.g., number of hijackings, weapon type) differed significantly between terrorist and nonterrorist perpetrators. By focusing on situational characteristics like these, the researchers were able to offer situation-specific policy recommendations.

Reference ClarkeClarke (1980) proposed SCP as a method for reducing crime through managerial and environmental changes. Situational crime prevention rests on the assumption that offenses, regardless of whether they are carefully planned, are heavily influenced by situational opportunities (Reference ClarkeClarke, 1997, Reference Clarke, Wortley and Townsley2016). The goal of SCP is to manipulate the environment in order to raise the costs and reduce the benefits of crimes. Reference Clarke, Homel and LabClarke and Homel (1997) offer sixteen crime opportunity-reducing techniques (e.g., target hardening, natural surveillance). These strategies can also be applied to the prevention of terrorist attacks. For example, based on a situational perspective, Reference Perry, Apel, Newman and ClarkePerry et al. (2017) conclude that the “West Bank barrier” in Jerusalem has been effective in preventing suicide bombings.

Rational Choice and Deterrence

According to Table 1, the second most common criminology perspective applied to terrorism research in the major criminology journals reviewed is rational choice or deterrence theory. Clarke and Reference Clarke and CornishCornish’s (1985) rational choice theory borrows from economics, psychology and sociology to model offenders’ decision-making processes. In criminology, this perspective is used primarily to predict an individual’s initial involvement in crime. Because rational choice theory assumes that all individuals are decision makers who take incentives and risks into account, deterrence is often the focus of those who apply a rational choice perspective. In research on terrorism, criminologists have used rational choice perspectives most frequently to examine the deterrent effects of various counterterrorism policy initiatives.

The rational choice perspective is especially interesting in the study of terrorism because, as we discussed in the previous section, compared to ordinary criminal offenders, terrorist perpetrators are more likely to perceive the benefits of their crimes in terms of political, ethical or moral gains. Terrorism perpetrators may thus be more likely to engage in political extremism if they perceive the rewards of their actions as outweighing the risks or costs for a particular social cause or political movement. Much of the research in this area tries to gauge the extent to which various deterrent measures implemented by governments are successful either at preventing attacks in general or at preventing attacks on specific target types.

Based on a rational choice perspective, Reference Dugan, LaFree and PiqueroDugan, LaFree and Piquero (2005) examined aerial hijackings and found that some measures of target hardening like metal detectors and law enforcement at passenger check points significantly reduced the rate of new hijacking attempts. However, the authors also found that these methods were generally limited to nonterrorist hijackers. Likewise, Reference Carson, Dugan and YangCarson, Dugan and Yang (2020) apply rational choice theory to study the impact of government actions on radical eco-movement attacks and find that when government policies increase the costs of committing this type of extremist behavior, incidents decline.

While most applications of rational choice/deterrence perspectives in criminology have examined whether the threat or imposition of punishment reduces future criminal behavior, the perspective applies equally well to asking whether the threat of punishment instead increases future criminal behavior. Several applications of rational choice/deterrence perspectives in criminology studies of terrorism have looked for possible backlash effects of punishment or the threat of punishment. For example, Reference LaFree, Dugan and KorteLaFree, Dugan and Korte (2009) compare deterrence and backlash outcomes by examining rates of terrorist attacks in Northern Ireland between 1969 and 1992 as the British imposed various types of counterterrorism interventions. In support of backlash interpretations, the authors discovered that three of the six counterterrorist interventions used by the British during this period significantly increased the hazard of future terrorist strikes, two interventions had no significant effect in either direction, and only one found a short-lived deterrence effect. Similarly, Reference Hsu and McDowallHsu and McDowall (2020) found that repressive counterterrorism actions increased violence in Israel, and that backlash effects were dependent on the magnitude of government repression, the target and the lethality of terrorist attacks.

Other Criminological Theories

According to Table 1, other criminological theories were less commonly applied in recent terrorism research. Still, there were examples of most mainstream criminology theories, including four studies based on anomie/strain, criminal subcultures and life-course perspectives, three based on differential association/learning perspectives, and two each based on social control and social disorganization perspectives. Thus far, support for anomie/strain theory with regard to terrorist perpetrators is relatively weak, with one article finding that collective strain does not have a direct effect on violent extremist attitudes (Reference Nivette, Eisner and RibeaudNivette, Eisner & Ribeaud, 2017); one finding an opposite effect (i.e., economically deprived counties were less likely to have far-right perpetrators; Reference Freilich, Adamczyk, Chermak, Boyd and ParkinFreilich et al., 2015); and another finding effects of collective deprivation only for far-left terrorist attacks (Reference VaraineVaraine, 2020).

Three of the four criminological studies that have applied criminal subculture approaches to the study of terrorism have used the perspective to better understand radicalization in prison. While Reference Useem and ClaytonUseem and Clayton (2009) conclude that there are low or modest levels of radicalization in prison, Reference HammHamm (2009) and Reference LaFree, Jiang and PorterLaFree, Jiang and Porter (2020) find that spending time in prison and being exposed to radicalization in prison increases the likelihood of violent political extremism after release. In the fourth study in this group, Reference CotteeCottee (2020) finds support for a criminal subculture perspective in a study of Western Islamists. He argues that a radical Islamist subculture is composed of three characteristics that heighten the likelihood of extremist behavior: (1) violence and machismo; (2) death and martyrdom; and (3) disdain for the temporal world and its earthly concerns and possessions.

Among the life-course applications, Simi, Reference Simi, Sporer and BubolzSporer and Bubolz (2016) use in-depth interviews with former violent extremists to determine how childhood- and adolescent-related variables influence later patterns of radicalization and violent political extremism. They argue for a three-stage risk-factor model such that the individual: (1) experiences different types of adversity in childhood; (2) subsequently has conduct problems in adolescence; and (3) is then motivated to seek out circumstances that lead to participation in extremist groups. While Simi, Sporer and Bubolz focus on psychological processes leading to radicalization, Reference Carlsson, Rostami and MondaniCarlsson et al. (2020) instead emphasize the social processes that lead to terrorism and political extremism. The authors argue that the process of radicalization occurs in three stages: (1) a weakening of informal social controls; (2) association with individuals close to extremist groups; and (3) a process of “meaning-making” in relation to the group.

Life-course theories have also been used to explain terrorism-related recidivism and desistance. Reference Hasisi, Carmel and WolfowiczHasisi et al. (2020a) study offending patterns of imprisoned terrorists and show that many of the risk factors for recidivism are the same for terrorist perpetrators and other types of criminal offenders (e.g., age, prior criminal record). While the life-course perspective typically follows individuals, Reference MillerMiller (2012) examines activities of more than 500 terrorist groups and finds that terrorist organizations that arise more rapidly and attack more frequently also survive longer.

After the original formulation of Sutherland (Reference Sutherland and CresseySutherland & Cressey, 1960), differential association begins with the premise that criminal behavior, like other behavior, is learned. Specifically, individuals interact in small intimate groups to learn and adopt values that are favorable to breaking the law. By this logic, social learning sometimes overlaps with criminal subcultures. Reference LaFree, Jiang and PorterLaFree et al. (2020) argue that imprisonment and prison radicalization increase the probability of post-prison violent extremism because inmates, through association with peers in prison, learn and adopt radical values. This argument is also supported outside of prison as individuals with radical peers are more likely to engage in violent political acts (Reference LaFree, Xie and MatanockLaFree et al., 2018). Further, some scholars argue that learning does not necessarily come from associations with deviant peers but can be facilitated by previous high-profile examples. Thus, Reference Miller and HaywardMiller and Hayward (2019) apply learning theory to understand the contagion-like nature of terrorist attacks where perpetrators purposely ram vehicles into groups of individuals with the intent of killing and injuring as many as possible.

Proponents of the social control perspective argue that crime occurs when an individual’s bonds with prosocial communities or institutions are weakened (Reference HirschiHirschi, 1969). Common prosocial bonds include employment, education, marriage and military experience. Reference LaFree, Xie and MatanockLaFree et al. (2018) apply this perspective to a sample of American political extremists and find that one variable associated with social control theory (lack of stable employment) is a significant predictor of violent political extremism. Further, in a model that connects social control and strain theories to political aggression, Reference Pauwels, Ljujic and de BuckPauwels, Ljujic and de Buck (2020) find that social integration, or an accumulation of social bonds, increases perceived respect for others and commitment to procedural justice.

In their classic formulation of social disorganization theory, Reference Shaw and McKayShaw and McKay (1942) argue that when communities are unable to develop strong social bonds, individuals in those communities will be more likely to engage in a variety of antisocial behavior including crime. This perspective suggests that crime will be more common in areas that lack shared norms and have high population heterogeneity, residential mobility and concentrated disadvantage (Reference BursikBursik, 1988). Reference LaFree and BersaniLaFree and Bersani (2014) apply the social disorganization perspective to terrorist attacks in the United States and find that counties with high population heterogeneity and residential instability experience high rates of terrorist attacks. The authors also find that counties with high concentrated levels of poverty are associated with few attacks. However, in a more general attempt to explore the applicability of various macro-level theories for understanding terrorist attacks, Reference Freilich, Adamczyk, Chermak, Boyd and ParkinFreilich et al. (2015) find little support for measures commonly associated with social disorganization theory.

According to Table 1, five articles in our review looked at theoretical perspectives that appeared only once. Four of these articles (Reference Mythen and WalklateMythen & Walklate, 2006; Reference ArrigoArrigo, 2010; Reference RuggieroRuggiero, 2010; Reference Ilan and SandbergIlan & Sandberg, 2019) take a critical look at mainstream criminology theories and encourage researchers to move beyond traditional criminology theories for understanding radicalization and political violence. Three of these articles appeared in the British Journal of Criminology and the other appeared in the European Journal of Criminology. All four articles recommend some version of a conflict/constructionist perspective as an alternative to a more traditional criminological framework. We classify the fifth article with only a single theoretical mention as psychological. Reference Corner and GillCorner and Gill (2020) analyze ninety case studies to examine the onset of psychological distress across three stages of terrorist involvement (engagement, disengagement and post-disengagement). The authors demonstrate that the relationship between terrorist engagement and psychological distress is mediated by several variables and combinations of variables.

Conclusions: Theoretical Contributions

In general, research on terrorism guided by mainstream criminological theories and published in criminology outlets is still relatively uncommon but clearly growing since 9/11. Based on a review of articles on terrorism and political extremism in twelve leading criminology journals from 2000 to 2020 (Reference LaFree, Yanez, Busher, Malkki and MarsdenLaFree & Yanez, In Press), forty-nine of fifty-four articles (91 percent) that applied a criminological theory to the study of terrorism or political extremism were published after 2010. In other words, the use of criminology theories to explain terrorism and political violence has increased rapidly in the decade since 2010. Nevertheless, mainstream criminology theories like social learning, social control and social disorganization are still rarely applied to understand the etiology of terrorism.

It is important to acknowledge that although the twelve journals reviewed here are among the most influential in the field of criminology, the selection is far from complete. In particular, the journals reviewed exclude articles by criminologists in general social science journals (e.g., American Sociological Review, Social Forces) or journals that specialize in the study of terrorism (e.g., Terrorism and Political Violence, Studies in Conflict and Terrorism). The list reviewed also excludes more specialized criminology journals (e.g., Homicide Studies, Journal of Interpersonal Violence) and most journals connected to specific countries (e.g., Australian and New Zealand Journal of Criminology, Canadian Journal of Criminology and Criminal Justice). In addition, the focus is only on articles published in refereed journals and excludes contributions from monographs and book chapters.

Researchers generally agree that among the purposes of science, description is the most basic. For example, fields like biology began by simply describing flora and fauna. As a field matures, its practitioners are likely to develop increasingly complex scientific methods that move beyond description to explanation, prediction and control. Many of the atheoretical criminology articles reviewed here were efforts to describe specific aspects of extremism, terrorism or the radicalization process. Recent articles, which more often incorporate theoretical perspectives, are also more likely to move beyond description.

The mix of theories represented in the articles reviewed is also of interest. The most common theoretical perspectives that we identified were situational/routine activities and rational choice/deterrence perspectives. By contrast, we found relatively few articles applying popular mainstream criminology theories like social control, strain and learning. These patterns may tell us something about the current state of data on terrorism and political extremism; in particular, the strong reliance on open source data. In the context of terrorism research, open source data are unclassified data drawn from public resources, typically the print and electronic media. As noted earlier, because most common criminological data sources – official police data and victim or offender surveys – are generally not available in the case of political extremism, terrorism researchers have relied extensively on open source data. The fact that nearly all of the theoretical studies we report in this section are based on open source terrorism/extremism-related data has important implications for the theories being tested and the timing of those tests.

In general, open source terrorism data lend themself to situational/routine activities and rational choice/deterrence perspectives. Given that much of the open source data on terrorism includes information on the characteristics of terrorist attacks (Reference LaFree, Muro and WilsonLaFree, 2022), these data usually include information about the situations under which attacks occur (e.g., location, degree of urbanization, time of day). This makes situational perspectives an obvious theoretical choice. Likewise, because much of the open source data being examined focuses on terrorist attacks, it is relatively straightforward to test rational choice/deterrence models by looking at what happens to the rate of terrorist attacks before and after some specific intervention. By contrast, open source event data are often less useful for understanding individual-level decision making. Terrorism event databases like the GTD only include individual information to the extent that it is explicitly linked to attacks (e.g., there were two perpetrators of a mass shooting). Many of the mainstream criminology theories that are underrepresented in our survey of terrorism-related articles (e.g., social control, differential association, situational action theory) require detailed information on the characteristics and behaviors of perpetrators – data that thus far have been less common for those who engage in terrorism.

These considerations may also help explain the timing of theoretical contributions to the literature on political extremism and terrorism. The only three theories that were present before 2010 in the articles reviewed in Table 1 were situational, rational choice and criminal subcultures. Because some open source event databases have been available for decades (Reference MickolusMickolus, 1976; Reference LaFree, Dugan and MillerLaFree, Dugan & Miller, 2015), situational and rational choice perspectives were easier to apply before 2010. Further, the two studies grounded in a criminal subcultures perspective and published before 2010 (Reference Useem and ClaytonUseem & Clayton, 2009; Reference HammHamm, 2009) were both based on separate data collection projects on political extremists in prison and did not depend on perpetrator databases.

The open source data situation has changed considerably in recent years as new individual-level databases on political extremism have started to appear. Three of the most extensive of these are from the United States: (1) the American Terrorism Study (Reference Smith and DamphousseSmith & Damphousse, 2007); (2) Profiles of Individual Radicalization in the United States (PIRUS; Reference LaFree, Xie and MatanockLaFree et al., 2018); and (3) the Extremist Crime Database (Reference Freilich, Chermak, Belli, Gruenewald and ParkinFreilich et al., 2014). Two additional individual-level databases focus on Islamist cases in Europe and the United States (Western Jihadism Project; Reference Klausen, Morrill and LibrettiKlausen, Morrill & Libretti, 2016) and on lone actor perpetrators in the United Kingdom and the United States (Lone Actor Terrorist Database; Reference Gill, Horgan and DeckertGill, Horgan & Deckert, 2014). The availability of these databases makes it increasingly feasible to test criminological theories at the individual level, including life course (Reference Hasisi, Carmel and WolfowiczHasisi, Carmel & Wolfowicz, 2020a), strain (Reference VaraineVaraine, 2020), differential association (Reference LaFree, Jiang and PorterLaFree, Jiang & Porter, 2020) and social control (Reference LaFree, Xie and MatanockLaFree et al., 2018) perspectives. As more individual-level databases become available in the future, we are likely to see a growing number of articles based on criminological theories that have to this point been relatively uncommon.

The increasing interest in individual-level data on terrorism itself likely represents a shift in research away from explaining crime events (i.e., terrorist attacks) toward understanding criminal development (i.e., radicalization). As 9/11 and the threat of radical Islamist attacks recedes and concerns about domestic, especially right-wing, terrorism increases, there appears to be growing interest in understanding radicalization. In general, individual-level theories like social control and situational action theory may be more relevant than rational choice and situational perspectives for explaining pathways to radicalization.

It is clear that criminological interest in terrorism has grown dramatically since the turn of the twenty-first century. Although there are indications that this specialization within criminology has been undertheorized, I also find evidence that applications of criminological theories to understand terrorism are becoming more common over time. A major challenge in terms of applying criminological theories to radicalization and political extremism is that the nature of terrorism makes it difficult to track individual perpetrators through traditional criminological data sources such as police files or victimization or self-report surveys. In response to this challenge, terrorism researchers have turned to open source data that most often counts terrorist attacks, rather than the characteristics and behaviors of perpetrators. As a result, popular criminology theories that focus on predicting individual offender behavior have been less useful than theories that emphasize where terrorism occurs (e.g., situational/routine activities) and whether it increases or decreases following counterterrorism interventions (rational choice/deterrence). Scholars will likely find it easier to do theoretically driven research in the future as more individual-level data become available.

Series Hazard Modeling

According to Reference DuganDugan (2011), the series hazard model is an extension of the Reference CoxCox (1972) proportional hazard model, and is used to estimate the impact of policy interventions on the risk of additional incidents (in this case, terrorist attacks) controlling for a set of relevant variables. The key distinction between the series hazard model and the more traditional Cox proportional hazard model is that the latter estimates the contributions of covariates to the hazard of repeated events (e.g., terrorist attacks) for one unit rather than for a single event (e.g., arrest) across multiple units. This method is most often used as an alternative to time-series analysis, which also estimates the effects of covariates on outcomes for a single unit over time. Reference DuganDugan (2011) notes that a major advantage of the series hazard model over standard time-series analysis is the ability of hazard models to capture variation in the timing between events. By contrast, time-series analysis is arbitrary with respect to time in the sense that it aggregates all events to a single time unit (e.g., month, quarter, year) and then counts their frequency.

In the previous section, I mentioned a study by Reference Dugan, LaFree and PiqueroDugan, LaFree and Piquero (2005) on aerial hijackings in which we used the series hazard model to estimate the deterrent effects of several policies on the hazard of six categories of aerial hijacking, 1931–2003: (1) US hijackings; (2) non-US hijackings; (3) Cuban hijackings; (4) terrorist-motivated hijackings; (5) non-terrorist-motivated hijackings; and (6) all hijackings. The results showed that policies that targeted specific categories of hijackings were most effective – for example, raising criminal penalties in Cuba for Cuba-bound hijackings. However, the results also showed that traditional deterrence strategies like tighter screening methods were ineffective in reducing terrorist-motivated attacks.

A series hazard model was also used in the Reference LaFree, Dugan and KorteLaFree, Dugan and Korte (2009) evaluation of six interventions by the British government on IRA terrorist attacks discussed earlier. We found that three of the interventions led to increases in subsequent attacks, two had no impact and only one (a large military operation) was linked to declines in future terrorist attacks. Reference CarsonCarson’s (2014) use of series hazard models to assess the impact of four laws on the activity of radical eco-groups produced mixed results. The tree-spiking clause of the Anti-Drug Abuse Act (ADA) decreased environment-only attacks but increased the hazard of animal-related attacks. The Animal Enterprise Terrorism Act decreased future animal-related attacks. However, neither the Anti-Terrorism and Effective Death Penalty Act nor the Animal Enterprise Terrorism Act had a significant effect on any of the outcomes. Reference Sharvit, Kruglanski and WangSharvit et al. (2013) used series hazard models to assess the effects of efforts by the Israeli government on different measures of Palestinian terrorism by specific groups and found that the response to Israeli tactics varied across groups. Finally, Reference Argomaniz and Vidal-DiezArgomaniz and Vidal-Diez (2015) used the series hazard model to test the effects of six major efforts by the Spanish government to reduce the risk of terrorism by the Basque group ETA and found evidence that some of these efforts increased the risk of new terrorist attacks by ETA rather than deterring them.

Group-Based Trajectory Analysis

Group-based trajectory analysis was originally developed by criminologists to examine individual patterns of criminal offending beginning early in childhood and extending through adolescence and on into adulthood (Reference Nagin and LandNagin & Land, 1993; Reference Nagin and TremblayNagin & Tremblay, 1999). Group-based trajectory analysis is especially useful for exploring a long-standing issue in criminology: whether crime rates decline as individual offenders age or whether the appearance of declining crime rates with age is instead produced by a mixture of nonoffenders and offenders with varying lengths of criminal careers. Reference LaFree, Yang and CrenshawLaFree, Yang and Crenshaw (2009) used GBTA to study the attack patterns of fifty-three foreign terrorist groups that were identified by the US State Department as posing a special threat to the United States. My colleagues and I produced two sets of analyses that differed in terms of whether the attacks targeted US citizens or non-US citizens and found that each study produced four distinct trajectories. Three of these trajectories distinguished three periods of rapid growth in terrorist attacks: the 1970s, the 1980s and the early twenty-first century. The fourth category included terrorist groups that attacked very infrequently during the four decades included in the analysis. Our analysis also revealed that despite being labeled as a threat to US nationals by the US government, these groups targeted the US homeland in only 3 percent of their attacks. In short, a group of foreign terrorist organizations that were avowedly anti-American in fact rarely attacked US targets!

In 2012, Miller expanded the application of GBTA by using it to examine attack patterns of terrorist groups at the beginning and at the end of their organizational careers. She studied the attacks of 557 terrorist groups in the GTD that were active for at least 365 days between 1970 and 2008. She used GBTA to build a more complete understanding of how the origins of terrorist organizations are related to their decline and demise. She found that terrorist groups that perpetrate attacks at a rapid pace immediately after their onset are two to three times more likely to attack frequently throughout their careers, and compared to other groups, decline at a slower rate.

Criminologists have also applied GBTA to spatial units in order to examine country-level trends in terrorism. Reference LaFree, Morris and DuganLaFree, Morris and Dugan (2010) examined worldwide terrorist attack patterns for 206 countries from 1970 through 2006 and found that countries could be classified into five distinct trajectory categories. Among them was a category composed of only five countries, but accounting for nearly 40 percent of all attacks. The authors point out that, as with more common crime, terrorist attacks are highly concentrated both in time and space.

Reference Morris and SlocumMorris and Slocum (2012) updated the Reference LaFree, Morris and DuganLaFree, Morris and Dugan (2010) study in a methodological paper that compared country-level patterns of domestic and transnational terrorism using two types of group-based analyses, latent class growth analysis (LCGA) and an alternative GBTA modeling approach based on general mixture modeling (GMM). They conclude that for the purpose of identifying hot spots of terrorist activity, LCGA results provide more useful results than GMM.

Self-exciting Point Process Methods

Self-exciting point process methods were originally developed by scholars in other disciplines (Reference HawkesHawkes, 1971; Reference OzakiOzaki, 1979), but were gradually adopted by criminologists (Reference Mohler, Short, Brantingham, Schoenberg and TitaMohler et al., 2011; Reference MohlerMohler, 2014). Self-exciting point process models are generally used to describe the intensity of some outcome (in this case terrorist attacks) based on the timing of previously occurring outcomes. The idea is that the timing of previous events may “excite” the occurrence of future events. Reference White, Porter and MazerolleWhite, Porter and Mazerolle (2013) apply SEP process models to terrorist attacks within Indonesia, the Philippines and Thailand from 2000 to 2010 to measure and compare the fluctuation of terrorist activity over time. The researchers identify three parameters describing trends in terrorist attacks: risk, resilience and volatility. They measure risk by estimating increases in the expected number of daily attacks, resilience as the length of time after an initial attack that it takes the risk of a new attack to return to its pre-event level, and volatility as the total increase in risk caused by each new terrorist attack. Their findings show that the patterns of risk, resilience and volatility differ across countries. For the three countries across the years examined, the Philippines was the riskiest and had the least resilience. Indonesia was the least risky, the least volatile and had the most resilience. Thailand was the most volatile of the three countries.

Reference Porter and WhitePorter and White (2012) present a variation of the SEP process model that jointly estimates the probability of an attack and the number of attacks on a given day in Indonesia as a function of the timing of previous attacks. The authors claim that this variant of the SEP process model improves predictability by rapidly adapting to changes in attack patterns and that the predictive capabilities of the SEP model can help governments more strategically allocate scarce security resources.

Analysis of Spatial and Temporal Clustering

Another set of methodological contributions that were developed in other disciplines but applied to terrorism by criminologists also identifies hot spots in terrorist activity, but in this case, by estimating the spatial and temporal clustering of events. To examine how terrorist attacks cluster spatially and temporally over time, researchers have relied on the Knox test (Reference KnoxKnox, 1964), which uses a 2 × 2 contingency table to assess pairs of events according to how close they are in space and time. In the mid-1960s, researchers started using the Knox test to explore space-time clustering of different forms of cancer. Decades later, Reference Braithwaite and JohnsonBraithwaite and Johnson (2012) used similar methods to investigate the space-time clustering of improvised explosive device (IED) attacks in Iraq using data on 3,775 IED discoveries and explosions between January and June 2005. The researchers used Knox analysis to study the connections between IED events and a wide variety of counterinsurgency activities aimed at stopping these events. They found that IED events were indeed highly clustered in space and time. After an initial IED explosion, more are likely to occur soon (two weeks or less) and nearby (within 1.5 kilometers).

In the earlier discussion of criminology theories applied to terrorism, I introduced an article by Reference Behlendorf, LaFree and LegaultBehlendorf, LaFree and Legault (2012) that identified localized bursts of terrorist attacks (which the authors call “microcycles”) for ETA attacks between 1970 and 2007 and FMLN attacks between 1980 and 1992. Behlendorf and colleagues use the Knox analysis to identify these microcycles. Because the Knox test evaluates distances in space and time between every event and every subsequent event, the authors ultimately examined the distances between 1.5 million pairs of attacks drawn from 1,762 cases for ETA and the distances between almost 3.5 million pairs of attacks drawn from 2,636 cases for the FMLN. The researchers found significant evidence for space and time interaction for five miles and two weeks from the original event. They then modeled the predictors of these “microcycles” and concluded that bombings were the predominant attack mode and that compared to nonfatal attacks, fatal attacks were less likely to be part of microcycles.

Another type of spatial analysis brought to the study of terrorism by criminologists is the use of Local Indicators of Spatial Association (LISA) statistics. They were originally developed by Luc Reference AnselinAnselin (1995) to locate hot spots. Reference Cohen and TitaCohen and Tita (1999) extended the cross-sectional application of LISA statistics to identify patterns of diffusion for youth-gang homicides over time within one city. Their model distinguishes between contagious diffusion, which is diffusion across contiguous units, and hierarchical diffusion that spreads more broadly with common influences.

Reference LaFree, Dugan, Xie and SinghLaFree et al. (2012) apply temporal LISA statistics to terrorist attacks by the Basque terrorist group, ETA. The group was chosen because its campaign between 1970 and 2007 was nearly continuous, allowing for the shifts in spatial patterns to be meaningful. Further, in 1978, the group announced a change in tactics from seeking an outright military victory in the Basque region to undertaking a broader war of attrition throughout Spain. The authors interpret this tactical shift as moving from “control” attacks that take place mostly in the Basque region to “attrition” attacks that are perpetrated throughout Spain and are calculated to wear down the Spanish government. The authors define control attacks as those that spread through direct contact between spatial units in the terrorist organization’s base of operations (i.e., contiguous provinces). They define attrition attacks as those that spread to more distant locations outside of the terrorist organization’s operational base (i.e., to more distant, noncontiguous provinces). The authors find that following the public announcement of a change in tactics by the leadership of ETA in 1978, attacks in Spain indeed shifted from control to attrition forms of diffusion.

Reference PerryPerry (2020) points out that most of the prior literature examining the geospatial distribution of terrorist attacks has focused on macro-level analyses such as countries and regions while few studies have examined the micro-level distribution of attacks. He argues that this contrasts with studies of ordinary crime, where there is a large body of research on micro-level crime hot spots. To bring the advantages of micro-level spatial analysis to the study of terrorism, Perry relies on a database of terror attacks in the city of Jerusalem between 2000 and 2017. The database includes the exact geographic location of attacks involving explosives, shootings, stabbings, attacks with a deadly weapon, and attacks using vehicles. Perry’s analysis shows that there is a high frequency of terror attacks concentrated in specific hot spots in Jerusalem and these hot spots are relatively stable over time. He points out that the results suggest the need for specialized counterterrorism responses equivalent to the hot spots policing strategies that often inform ordinary police work.

Reference Hasisi, Perry, Ilan and WolfowiczHasisi et al. (2020b) analyzed vehicular terrorist attacks (where perpetrators deliberately ram vehicles into crowds, buildings or other vehicles) in relatively small neighborhoods within Jerusalem and the West Bank. Additionally, the researchers analyze the concentration of the travel routes of attackers and compare the relative concentrations of the travel routes with the attack sites. The authors find that terrorist perpetrators, like ordinary offenders, favor targets close to where they live; the number of attacks decays as the distance between the residence of the attacker and the attack increases.

Risk terrain modeling (RTM) is another recent method for analyzing spatial and temporal clustering of events. Reference Marchment, Gill and MorrisonMarchment, Gill and Morrison (2020) use RTM to identify the risk factors for bombings and bomb hoaxes committed by violent dissident Republicans in Belfast, Northern Ireland. Based on their analysis, they conclude that high-risk areas for bombings were associated with previous protests and riots, spatial characteristics of the targets, punishment attacks (i.e., attacks on dissident members of the perpetrator’s own community) and areas dense with pubs and bars. By comparison, bomb hoaxes were associated with punishment attacks, police stations and areas of the city that were dense with shops. Based on the observed differences between bombings and bomb hoaxes, the researchers conclude that perpetrators choose targets that are relevant to their ideology or that have a high chance of success. More generally, their results suggest that offenders assess risk and select targets rationally based on these assessments.

Network Analysis

Reference Semmelbeck and BesawSemmelbeck and Besaw (2020) examine terrorist networks and connections between terrorism and ordinary crime. The authors use random forest methods to predict which terrorist groups will also engage in organized crime. Briefly, random forest methods depend on machine learning techniques that are used to solve regression and classification problems. The random forest algorithm makes predictions by taking the average or mean of the output from various outputs or “trees.” Increasing the number of outputs increases the precision of the prediction. Using random forest methods, Semmelbeck and Besaw show that it is the characteristics of terrorist organizations themselves rather than environmental factors that predict their involvement in organized crime. Even though the authors state that their analyses are not designed to identify causal relations, the findings are noteworthy as such organizational characteristics may easily serve as useful warnings about future behavior. Their analyses highlights how models that incorporate linear associations may miss out on the nonlinear pathways in which terrorist groups end up engaging in organized crime.

Reference McMillan, Felmlee and BrainesMcMillan, Felmlee and Braines (2020) investigate how the structure of terrorist networks develops over time, in different phases of their formation and activities. The authors argue that these changes are guided by balancing the needs for efficiency and security in the period before attacks. The authors use information about individual terrorists and their mutual social relationships connected to eleven prominent Islamist attacks (e.g., the 2002 Bali bombings) since the 1980s. Based on network analysis, the researchers find that in the period before a violent attack, networks become increasingly well-connected and organized around key actors. These kinds of analyses may help counterterrorism efforts by suggesting which actors in networks are the most influential targets.

Agent-Based Modeling

Two decades ago, Reference BankesBankes (2002: 7199) declared that ABM represented “a revolutionary development for social science.” Agent-based modeling has been gradually gaining traction in criminology, and applications to terrorism are beginning to appear. A recent example is Reference Weisburd, Wolfowicz, Hasisi, Paolucci and AndrighettoWeisburd et al. (2022) who use agent-based models to compare the impacts of three different types of interventions targeting recruitment to terrorism – community workers at community centers; community-oriented policing; and an employment program for high-risk agents. The first two programs are social interventions that focus on deradicalization and changing the dispositions of agents in the model, whereas the employment program focuses on “deflection” and represents a situational-/opportunity-reducing approach to prevention. The results show significant impacts of the community worker and community policing interventions on radicalization but no significant impact on recruitment. In contrast, the employment intervention had a strong and significant impact on recruitment, but little impact on radicalization. The ABM analysis underscores the importance of social interventions that focus on risk and protective factors for reducing radicalization in society. The findings also suggest that policy makers should focus less on counter and deradicalization approaches and more on deflection and opportunity reduction.

Meta-analysis

A final method used frequently in criminology that has been applied to the study of terrorism is meta-analysis. A meta-analysis is a statistical analysis that combines the results of multiple scientific studies that address the same question, with each individual study reporting measurements that are expected to have some degree of error (Reference Lipsey and WilsonLipsey & Wilson, 2001). The aim is to use statistical calculations to derive a more accurate pooled estimate of the underlying relationship between some intervention and some outcome. Based on searches in English, German and Dutch, Reference Wolfowicz, Litmanovitz, Weisburd and HasisiWolfowicz and colleagues (2020) identified sixty studies containing over seventy individual models and provide a meta-analysis of risk and protective factors for three radicalization outcomes: radical attitudes, intentions and behaviors. They used random effects meta-analysis to produce pooled estimates to quantify the effects of all risk and protective factors for which they were able to identify rigorous empirical data. They next created a rank order of effect sizes to identify the relative importance of each factor. Across all three outcomes, the researchers found that the most important protective and risk factors are those associated with social control and self-control theories, specifically factors pertaining to peers, school, parenting, social integration, and attitudes towards norms and values such as legitimacy.

Methods Contributions of Criminology: Conclusions

The research methods described in this section demonstrate that criminologists have indeed contributed to the growing methodological sophistication of terrorism research. The increased application of cutting-edge methods commonly used in criminology, including many of those reviewed here, are improving the rigor of scholarship on terrorism. Moreover, the application of criminological methods offers fresh perspectives that will help us better understand terrorism and its effects. I will argue in the next section that these developments not only improve research on terrorism but also increasingly provide insights that may be usefully applied to more common areas of mainstream criminology. I next briefly consider some of the policy advantages of criminal justice responses to terrorism.

Perpetrator Age

The common criminological observation that violent crime is associated with youth (Reference FarringtonFarrington, 2003; Reference Sweeten, Piquero and SteinbergSweeten, Piquero & Steinberg, 2013) is also commonly made in the research literature on terrorist perpetrators. For example, Reference PapePape (2005) found that the average age of offenders in his study of suicide terrorists ranged from a low of twenty-one years for the Lebanese Hezbollah to thirty years for Chechen political extremists. In Table 3, I compare the age of violent, nonviolent and total political extremists in the PIRUS data. From the SPI, I compare all inmates and inmates convicted of either violent or nonviolent offenses.

What is most striking about Table 3 is that both violent and nonviolent political extremists are considerably older than both violent and nonviolent prison inmates. For both groups as a whole, this difference amounts to about ten years.Footnote ³ Similar findings have been reported in several other recent studies. For example, in an analysis of 600 American Islamist extremists, Reference Klausen, Morrill and LibrettiKlausen, Morrill and Libretti (2016) found that the median age for commission of violence was older and occurred across a broader age range than was the case for offenders who had committed more typical violent crimes. Similarly, in a comparison of political extremists and gang members in the United States, Reference Pyrooz, LaFree, Decker and JamesPyrooz et al. (2017) found that the average age of adolescent gang members in the United States from the National Longitudinal Survey of Youth was nineteen years of age while the average age of US political extremists (from the PIRUS data) was thirty-four years old.

Other Demographic Comparisons

In Table 4, I compare a set of common demographic characteristics for violent and nonviolent political extremists from the PIRUS to convicted offenders from the 2016 SPI.

Gender. The overrepresentation of men in both crime (Reference Gendreau, Little and GogginGendreau, Little & Goggin, 1996; Reference DeLisi, Neppl, Lohman, Vaughn and ShookDeLisi et al., 2013) and terrorist attacks (Reference BerrebiBerrebi, 2007; Reference LaFree, Xie and MatanockLaFree et al., 2018) is well established. Nevertheless, Reference BloomBloom (2017) argues that the proportion of women engaging in terrorism is increasing, and a growing literature (Reference Sjoberg and GentrySjoberg & Gentry, 2011; Reference BloomBloom, 2012; Reference Ortbals and Poloni-StaudingerOrtbals & Poloni-Staudinger, 2018) has examined forms of female participation in terrorism.

Our results generally confirm prior research on the disproportionate participation of men in both crime and terrorism. According to Table 4, men are consistently more highly represented among both violent and nonviolent political extremists and prison inmates. However, there are substantial differences across the categories being compared. Men were most common in our sample of violent political extremists, where they accounted for nearly 95 percent of all cases. They were least common among nonviolent prison inmates, where they accounted for less than 70 percent of all cases. In general, men were more highly represented across categories of political violence than ordinary crime. For both political extremists and inmates, the gender gap is greater for violent than nonviolent crimes.

Race/ethnicity. According to Table 4, there are major differences between US political extremists and prison inmates in terms of race/ethnicity. Comparing all extremists to all inmates shows that Blacks are more than three times more likely and Hispanics are nearly eight times more likely to be in the prison inmate sample than in the political extremist sample.

Born US citizen. In Table 4, I next compare whether the political extremists and prison inmates were born in the United States. Although the differences were relatively small, prison inmates were somewhat more likely than political extremists to have been born in the United States. The difference is largest among political extremists and inmates connected to violent crimes, where 92 percent of inmates but only 78 percent of political extremists were born in the United States.

Marital Status. The relationship between marital status and lower crime rates has been supported in criminology research across a variety of methodological approaches (Reference KirkKirk, 2012; Reference Bersani and DohertyBersani & Doherty, 2013), however, its relationship to extremist violence is less clear. Reference Russell and MillerRussell and Miller’s (1977) study of terrorist profiles in eighteen countries concluded that the typical perpetrator is an unmarried male, and Reference BerrebiBerrebi (2007) found evidence suggesting married individuals are less likely to participate in Palestinian extremism, however, other scholars have reached different conclusions. For example, three-quarters of the Islamist terrorist perpetrators in Reference SagemanSageman’s (2008) study of Islamist extremists were married. In fact, Sageman concluded that those joining Islamist organizations are frequently not fully trusted unless their wives or daughters are sisters of other terrorist perpetrators. Reference BakkerBakker (2006) also reported high levels of marriage among Islamist perpetrators. Reference ShapiroShapiro (2013) found that many Islamist groups encourage intermarriage among group members to build intragroup cohesion and trust. In a recent study, Reference Altier, Leonard Boyle and HorganAltier, Leonard Boyle and Horgan (2021) found that marriage did not act as a protective factor against extremism. In fact, in many cases, the spouse was supportive of radical behavior and marriage was found to predict recidivism.

My findings in Table 4 are generally in line with the latter results, showing that political extremists were more likely than prison inmates to be married. I hasten to add that a large part of the difference in marital status for political extremists and prison inmates is likely due to the fact that the political extremists are on average substantially older than the prison inmates and therefore have had more time to get married.

Educational Status. In general, the link between educational status and violent crime has been inconsistent (cf., Reference GlaserGlaser, 1964; Reference UggenUggen, 2000). Reference Tauchen, Witte and GriesingerTauchen, Witte and Griesinger (1994) and Reference Witte and TauchenWitte and Tauchen (2000) found no significant link between educational attainment and crime after controlling for several individual characteristics. Similarly, Reference GroggerGrogger (1998) found no relationship between education and crime after controlling for wages. In contrast, using data from the National Longitudinal Survey of Youth, Reference Lochner and MorettiLochner and Moretti (2004) found that schooling significantly reduced the probability of incarceration and arrest. Other researchers (Reference Witte and TauchenWitte & Tauchen, 2000; Reference GottfredsonGottfredson, 1997) found that time spent in school significantly reduced criminal activity.

Reference Arnold and KennedyArnold and Kennedy (1988) apply similar arguments to their discussion of education as a factor discouraging participation in terrorism. But, thus far, these expectations have received little empirical support. Indeed, the findings from much prior research (Reference Krueger and MalečkováKrueger & Malečková, 2003; Reference PapePape, 2005; for a review, see Reference LaFree and AckermanLaFree & Ackerman, 2009) show that those who participate in terrorist actions are, if anything, somewhat better educated than the general population. For example, Reference SagemanSageman (2004) found that members of Islamist terrorist organizations were generally well educated compared with their compatriots: more than 60 percent had some education beyond high school. Similarly, Reference Russell and MillerRussell and Miller (1977) studied eighteen non-Muslim terrorist groups (including the Japanese Red Army, Germany’s Baader–Meinhof Gang and Italy’s Red Brigades) and found that, overwhelmingly, group members were well educated, with approximately two-thirds having at least some university education. Reference BerrebiBerrebi’s (2007) analysis of Hamas and the Palestinian Islamic Jihad found that both group-affiliated individuals and suicide bombers were far more likely than average Palestinian citizens to have obtained a secondary or postsecondary school education.

The results in Table 4 support those researchers who claim that educational attainment is generally higher for political extremists than ordinary prison inmates. For the full sample, nearly 37 percent of the political extremists have at least a high school education compared to just over 23 percent of the prison inmates. Compared to the prison inmates, the political extremists are nearly three times more likely to have a college degree.

Employment Status. The link between various measures of employment and crime is one of the most comprehensively researched areas in criminology, and many (but not all) studies conclude that the less consistent the work history, the higher the levels of criminal activity (Reference Smith, Devine and SheleySmith, Devine & Sheley, 1992; Reference UggenUggen, 2000). Similar arguments have been applied to discussions of the causes of terrorism (Reference Arnold and KennedyArnold & Kennedy, 1988; Reference LaFree and AckermanLaFree & Ackerman, 2009). However, thus far, empirical results have been less convincing than support for connections between unemployment and more ordinary types of crime. Research specifically examining the relationship between employment status and participation in terrorism has found that many members of organizations that use terrorism have jobs. For instance, Reference HewittHewitt (2003) found that members of the Ku Klux Klan had a diverse range of positions, from blue-collar laborers to business owners. Similarly, Reference SagemanSageman (2004) concluded that, at the time extremists joined Islamic terrorist groups, the majority of them were students, worked as professionals (e.g., doctors, engineers) or performed semiskilled labor.

Other research (Reference KruegerKrueger, 2007; Reference 90SilkeSilke, 2008) shows that those who participate in terrorist acts are frequently not the poorest members of their societies. An early study by Reference Russell and MillerRussell and Miller (1977) compiled profiles of more than 350 individual terrorist cadres and leaders across eighteen different terrorist groups from the years 1966 to 1976. The authors concluded that the majority of these individuals had middle-class backgrounds. Reference SagemanSageman’s (2004) survey of 172 members of Islamist terrorist groups found that about three-quarters came from upper- or middle-class backgrounds. Just over one-quarter (27%) came from working-class or poor backgrounds.

However, some recent research does report a positive relationship between unemployment and terrorism. In a recent US study, Reference LaFree, Xie and MatanockLaFree et al. (2018) found that a lack of stable employment was a strong risk factor for engaging in violent political extremism. My colleagues and I measured lack of stable employment by looking at those who were unemployed as well as those who alternated between periods of employment and unemployment and those who habitually changed careers in the years leading up to their involvement in extremism.

According to Table 4, prison inmates, both those convicted of violent and nonviolent crimes, are considerably more likely than political extremists in the PIRUS data to be employed. Employment data for the inmates refers to those who report working in any job within thirty days of arrest. Employment in the PIRUS data refers to the individual’s employment status at the time of exposure.

Military Background. Theories of informal social control in criminology argue that military service, along with marriage and employment, is a positive turning point that can disrupt criminal trajectories (Reference BouffardBouffard, 2003; Reference Bouffard, Laub, Maruna and ImmarigeonBouffard & Laub, 2004). This conclusion is generally supported in the empirical literature, with studies showing that military service is positively related to a substantial number of positive outcomes, such as economic well-being, job stability and desistance from crime (Reference BouffardBouffard, 2003; Reference Bouffard, Laub, Maruna and ImmarigeonBouffard & Laub, 2004).

By contrast, we could not identify any comparable literature in which the reported results show that military service is associated with lower rates of participation in political violence. In fact, preliminary evidence suggests that military service may be positively associated with extremist behavior. For example, some government reports (FBI, 2008; Department of Homeland Security, Office of Intelligence and Analysis, 2009) have detailed the risk of far-right recruitment activities in the US military. Moreover, there is evidence that military training increases rather than diminishes participation in terrorism among extremist Islamists (Reference CooleyCooley, 2002; Reference 81HafezHafez, 2008). Indeed, Reference MendelsohnMendelsohn (2011) argues that individuals with military training are often specifically recruited by extremist groups because of their useful skillset, whereas those without military training are deemed to be less desirable.

In support of those who argue that military experience may have a different impact on engaging in terrorism rather than ordinary crime, Table 4 shows that political extremists from the PIRUS sample are considerably more likely than prison inmates to have a military background. Interestingly, a military background is also more common for individuals who have engaged in violent crime – both for the political extremists and the prison inmates.

The 1970s and Earlier

The first four databases listed in Table 6 originated during the 1970s and earlier. The ITERATE database (https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/TH4ADJ) began coverage in 1968, includes only international attacks and has been periodically updated through 2020 (Reference MickolusMickolus, 2002; Reference Mickolus, Sandler, Murdock and FlemmingMickolus et al., 2010). The RAND Corporation was an early pioneer in developing terrorism event databases and with the support of the US State Department and the Defense Advanced Research Projects Agency (DARPA), in 1972, Brian Jenkins at RAND began to develop a “Chronology of International Terrorism” dating back to 1968 (www.rand.org/nsrd/projects/terrorism-incidents.html). Like ITERATE, the original RAND data were generally limited to international attacks.Footnote ⁴

The PGIS – a corporate descendant of the famous detective agency started in the mid-1800s by Scottish immigrant to the United States Allan Pinkerton – began collecting unclassified terrorism data in the mid-1970s. The PGIS trained researchers to identify and record terrorism incidents from wire services (including Reuters and the Foreign Broadcast Information Service [FBIS]), US State Department reports, other US and foreign government reporting, and US and foreign newspapers (e.g., the New York Times, the British Financial Times). The most unique aspect of the PGIS data is that from the beginning it included domestic as well as international terrorist attacks – the only early database to do so.

The US State Department began publishing an annual report on international terrorism in 1982 (reporting 1981 incidents), and in 1983, began calling the report “Patterns of Global Terrorism.” The Patterns Report reviews international terrorist events by year, date, region and terrorist group and includes background information on terrorist organizations, US policy on terrorism and counterterrorism efforts. The Patterns Reports were generally issued a few months after each calendar year.

The Late 1990s and Beyond

As shown in Table 6, four of the terrorism event databases include domestic as well as international attacks: the RAND-MIPT/RDWTI database, the GTD (stages 1 and 2) and the WITS data. Although the Data Base on Suicide Attacks (DSAT) includes both international and domestic attacks, it is limited to suicide attacks. In April 2001, the RAND Corporation, which had been collecting terrorism data since 1968, received support from the National Memorial Institute for the Prevention of Terrorism (MIPT) – an organization funded by the US Congress to study terrorism in the wake of the Oklahoma City bombing. With considerably more resources devoted to the database, RAND staff verified much of the earlier data and in 1998 began collecting terrorism data on domestic attacks. Funding for the RAND-MIPT data collection ended in 2008. However, shortly after, RAND received additional support and continued collecting terrorism event data, now referred to as the RAND Database of Worldwide Terrorism Incidents (RDWTI).

In 2001, I was able to secure original hard copies of the PGIS terrorism data – which by then contained more than 67,000 cases from 1970 to 1997.Footnote ⁵ With funding from the National Institute of Justice, our team at the University of Maryland completed the verification and digitization of the original PGIS data in December 2005. This marked the beginning of the GTD (www.start.umd.edu/gtd/). In April 2006, the GTD team received funding from the Department of Homeland Security through the National Consortium for the Study of Terrorism and Responses to Terrorism (START) to extend the GTD beyond 1997. Primary data collection of the GTD for 1998–2011 was completed by two different research teams and then verified and compiled by the GTD team at START (Reference LaFree, Dugan and MillerLaFree, Dugan & Miller, 2015). Based on these procedures, in March 2009, we released an extended version of the GTD through 2007 (Reference LaFree and DuganLaFree & Dugan, 2007). Updates were subsequently released first biennially and then annually.

DSAT was founded in 2004 at the University of Chicago and collects worldwide data on suicide attacks (Reference Pape, Rivas and ChinchillaPape, Rivas & Chinchilla, 2021; https://cpost.uchicago.edu/research/suicide_attacks/important_definitions/). DSAT defines a suicide attack as “an event in which one or more attackers deliberately kill themselves in an effort to harm or kill others.” It does not include attempted suicide attacks, in which a perpetrator fails to kill him or herself, or suicide missions, in which the perpetrator may expect to be killed in an attack. Importantly, DSAT does not require that a given suicide attack qualifies as an act of terrorism but rather includes suicide attacks undertaken for any purpose. The first confirmed suicide attack according to DSAT took place in 1982 and the data have been updated through 2019. According to Table 6, DSAT includes less than 7,000 attacks.

The WITS data collected by the National Counterterrorism Center (NCTC) began in 2004 but did not provide comprehensive annual coverage until 2005. It originated because of congressional dissatisfaction with the quality of the State Department’s Patterns Report. In its 2003 Patterns Report, the State Department concluded that “worldwide terrorism had dropped by 45 percent between 2001 and 2003.” However, when economists Alan Krueger and David Laitin reviewed the data tables at the end of the State Department’s Patterns Report for 2003, they found that the numbers in the tables did not add up and that the conclusion of the report, namely that worldwide terrorism had decreased that year, was in error and that terrorism had actually increased. When they subsequently published this information in an op-ed piece in the Washington Post (Reference Krueger and LaitinKrueger & Laitin, 2004b) and in an article in Foreign Affairs (Reference Krueger and LaitinKrueger & Laitin, 2004a), the State Department admitted that the report was wrong and retracted it. As a result of this criticism, the name of the report was changed to “Country Reports on Terrorism,” the statistical data and chronology of “significant” international terrorist events was dropped, and Congress mandated that, starting in 2004, terrorism data were to be compiled by the newly created NCTC.

The WITS data were collected by NCTC from open sources manually using commercial subscription news services, the US Government’s Open Source Center, local news websites reported in English and, as permitted by the linguistic capabilities of their employees, local news websites in foreign languages (Reference WigleWigle, 2010: 5). Like GTD and RAND-MIPT, WITS collected both international and domestic data. From its inception, a major goal of those administering WITS was to “cast a wider net on what may be considered terrorism” (p. 5). As a result, WITS was extremely inclusive in its coverage. From 2004 to 2011, WITS reported nearly 80,000 terrorist attacks – far more than any other event database for this period. These numbers stand in stark contrast to the earlier US State Department Patterns Reports, which typically reported only several hundred international terrorist attacks per year.

The final data set included in Table 6 is the GTD Stage 2. In recent years, the explosive growth of online media availability ushered in a new wave of innovation in terms of the collection of worldwide open source terrorism data. Early efforts to collect terrorism data quickly started to rely on news aggregators like the wire service Reuters and later online aggregators like LexisNexis, Factiva and OpenSource. However, these efforts have become far more sophisticated and comprehensive over time. At present, it is fair to say that none of the major open source terrorism databases rely only on manual data collection.

Starting in 2012, the GTD team at START began to increase substantially the amount of automation used to generate the data. The team still relies on primary sources including individual news outlets such as Reuters and Agence France-Presse, as well as existing media aggregators such as LexisNexis and Factiva, but these are now continually evaluated in terms of which sources make the most valid contributions to the overall data collection effort. At present, data collection for the GTD begins with a universe of two million articles published daily worldwide in order to identify the small subset of articles that describe terrorist attacks. The GTD team uses customized search strings to isolate an initial pool of potentially relevant articles and then relies on natural language processing methods to automatically identify and remove duplicate source articles by measuring similarities between pairs of documents. In addition, the team has developed a machine learning model using feedback from trained GTD staff that classifies documents identified by the initial automated processes to determine how likely they are to be relevant to terrorism. This model is continually refined using input from the research team regarding the accuracy of the classification results. At present, approximately 15,000–30,000 articles are manually reviewed to identify attacks for each month of data collection.

One of the innovations of GTD Stage 2 is to move from data coding based on area experts (e.g., Southeast Asia, Western Europe) to rely instead on domain-specific research teams organized to collect data on specific characteristics of attacks, including location, perpetrators, targets, weapons, tactics, casualties and consequences. Each domain-specific team records information according to the ever-evolving specifications of the GTD codebook (www.start.umd.edu/gtd/downloads/Codebook.pdf). In short, the GTD team uses automated tools to process millions of documents a day but human coders to digest the information and ensure the quality of the resulting data.

Conclusions: Terrorism Data versus Crime Data

Worldwide terrorism and crime data have very different histories. Cross-national estimates of terrorist attacks get under way in the late 1960s, along with the availability of satellite technology and portable cameras. Over time, they have become more inclusive and increasingly rely on electronic rather than print media. Because open source terrorism databases are not based on official reports, they are able to include information from many countries that are missing statistics on homicide and other common crimes. Compared to these more common crimes, terrorist attacks are unique in the sense that their perpetrators are often seeking media attention and therefore want to see information on their attacks showing up in the print and electronic media. On the other hand, open source databases have all the downsides that we discussed earlier, including whether minor events get reported, whether the press pays more attention to some types of terrorism than other types and, more generally, all of the sources of misinformation associated with the media. Attempts at developing worldwide crime data began in the 1950s and slowly grew more inclusive over time. However, the most successful efforts to produce worldwide crime data have been limited to homicide and still exclude many industrializing countries.

Youth Percentage

The size of the youth population is a frequent measure in cross-national studies of homicide (Reference NivetteNivette, 2011) but is rare in cross-national studies of terrorism (Reference Morris, LaFree, LaFree and FreilichMorris & LaFree, 2017). Most studies of terrorism that include age as a variable (Reference Klausen, Morrill and LibrettiKlausen, Morrill & Libretti, 2016; Reference Pyrooz, LaFree, Decker and JamesPyrooz et al., 2017) are micro-level studies focused on terrorism perpetrators. Nonetheless, as can be seen in Table 8, we find that both terrorism and homicide are significantly more common in countries with a high proportion of individuals aged fifteen to twenty-nine.

Economic Development

Economic development, usually measured as GDP, has been a common measure in both studies of terrorism (Reference Fahey and LaFreeFahey & LaFree, 2015) and homicide (Reference PridemorePridemore, 2008; Reference Messner, Raffalovich and SuttonMessner, Raffalovich & Sutton, 2010), and for both, results have been inconsistent. Findings from several studies show that countries with high economic growth and strong welfare measures experience fewer terrorist attacks (Reference Krieger and MeierrieksKrieger & Meierrieks, 2012; Reference ChoiChoi, 2015). For example, Reference Gries, Krieger and MeierrieksGries, Krieger and Meierrieks (2011) find that economic success significantly reduces terrorist violence in three of the seven Western European countries included in their sample. However, based on a sample of twelve West European countries (1994–2007), Reference Caruso and SchneiderCaruso and Schneider (2011) find that expected future economic growth is associated with an increase in current terrorist activity. Similarly, Reference LaFree and BersaniLaFree and Bersani (2014) find that terrorist attacks are less common in US counties with high levels of concentrated economic disadvantages. In another US study, Reference VaraineVaraine (2020) finds that far-right perpetrators mobilized more frequently under periods of long-term economic deprivation, whereas far-left terrorism was more prevalent under improving economic conditions. Still other studies find either no effect (Reference AbadieAbadie, 2006; Reference Kurrild-Klitgaard, Justesen and KlemmensenKurrild-Klitgaard, Justesen & Klemmensen, 2006) or a curvilinear effect (Reference Enders, Sandler and GaibulloevEnders, Sandler & Gaibulloev, 2011) of economic development on terrorism.

Per capita GDP is likely the single most commonly examined variable in quantitative studies of cross-national homicide (Reference LaFree and KickLaFree & Kick, 1986; Reference Messner, Raffalovich and SuttonMessner, Raffalovich & Sutton, 2010). Several prior studies (Reference BennettBennett, 1991; Reference Ortega, Corzine, Burnett and PoyerOrtega et al., 1992) find that homicide rates increase along with GNP or GDP. However, a meta-analysis by Reference NivetteNivette (2011) finds no effects for GNP. I obtained time-series data for GDP from the Penn World Table 9.1 (Reference Feenstra, Inklaar and TimmerFeenstra, Inklaar & Timmer, 2015). According to Table 8, increases in GDP are associated with significant increases in both the frequency of terrorist attacks and homicides.

Income Inequality

Many studies over several decades support the conclusion that country-level income inequality is associated with high homicide rates (Reference NivetteNivette, 2011; Reference Baumer and WolffBaumer & Wolff, 2014). Reference 83Krieger and MeierrieksKrieger and Meierrieks (2019) show that high levels of income inequality are associated with an increase in domestic terrorism, whereas redistributive efforts reduce terrorist activity. Reference 88PiazzaPiazza (2011) and Reference Enders and HooverEnders and Hoover (2012) both look at the impact of inequality (the GINI index) on terrorist attacks and fatalities and find that it is associated with high rates of attacks and fatalities.

I obtained an inequality measure (the GINI index) from the Standardized World Income Inequality Database (Reference SoltSolt, 2016). The results for inequality shown in Table 8 are in line with prior research on homicide but not terrorism: I find that inequality is associated with a significant increase in the former but no effect on the latter.

Poverty

Poverty is among the most widely studied, and most controversial, macro-level predictors of both homicide and terrorism. A link between area poverty and homicide is a consistent finding in most individual-level criminological research (Reference Pratt and CullenPratt & Cullen, 2005; Reference PridemorePridemore, 2008), however, the poverty–crime link is much less common in cross-national research. Moreover, most cross-national research on homicide that indicates a theoretical interest in poverty often measure it by using GDP, which we have already considered earlier. We also find little agreement about the extent to which economic measures like poverty increase terrorist attacks. The relationship between measures of poverty and terrorist attacks are significantly reduced once country-specific characteristics and measures of good governance and political freedom are taken into account, suggesting that economic stress may have indirect effects on terrorist attacks (Reference Basuchoudhary and ShughartBasuchoudhary & Shughart, 2010; Reference Plümper and NeumayerPlümper & Neumayer, 2010).

For this analysis, I operationalize poverty as the percentage of the population living below the national poverty line, based on population-weighted subgroup estimates from household surveys. According to Table 8, there is no significant relationship between poverty for either terrorist attacks or homicides.

Urban Percentage

Urbanization or urban percentage has been associated with increases in terrorist attack in several prior studies (Reference Campos and GassebnerCampos & Gassebner, 2013; Reference Danzell and ZidekDanzell & Zidek, 2013). By contrast, declines in urbanization have been identified as one of the variables responsible for declining homicide rates across countries (Reference NivetteNivette, 2011; Reference Baumer and WolffBaumer & Wolff, 2014). For my analysis, urban percentage measures the proportion of the population in a given country that resides in urban areas relative to the total resident population (World Bank Open Data, 2020). According to Table 8, as the proportion of the population that is urban increases, homicide counts increase, but I find no significant association between urban percentage and terrorist attacks.

Strength of Democracy

Strength of democracy is used frequently to predict country-level rates of terrorist attacks but is relatively uncommon as an independent variable in the homicide literature. An exception is Reference LaFree and TseloniLaFree and Tseloni (2006), who analyze homicide trends in forty-three nations and find that during the second half of the twentieth century, homicide rates increased for full democracies. Nevertheless, the authors also find that violent crime rates were highest for countries transitioning between autocracy and democracy. A good deal of research has investigated the extent to which regime type and level of democratization of a country correlates with terrorism (Reference PiazzaWilson & Piazza, 2013; Reference Gaibulloev, Piazza and SandlerGaibulloev, Piazza & Sandler, 2017). These studies operationalize country-level democracy in different ways but generally account for some or all of the following elements: electoral processes, civil and political freedoms, levels of political participation and competition. Despite the fact that the connection between democracy and terrorism has been analyzed frequently, the true effect of democracy on terrorist attacks remains uncertain. Research has suggested both positive (Reference PiazzaWilson & Piazza, 2013; Reference BellBell, 2017) and negative relationships (Reference LiLi, 2005; Reference Masters and HoenMasters & Hoen, 2012; Reference SimpsonSimpson, 2014) between measures of democracy and terrorism, with some studies finding no linkage between the two (Reference PiazzaPiazza, 2008a). Still other research (Reference AbadieAbadie, 2006; Reference Gaibulloev, Piazza and SandlerGaibulloev, Piazza & Sandler, 2017) points to a curvilinear, and more specifically to an inverted U-shaped, relationship between different measures of democracy, such as political freedom and terrorism.

My measure of democracy is the “polity score,” which classifies countries on a twenty-one-point scale ranging from full autocracy to consolidated democracy (Reference Jaggers and GurrJaggers & Gurr, 1995). Bivariate results in Table 8 confirm findings by other research (Reference ChenowethChenoweth, 2013; Reference PiazzaPiazza, 2013), showing that as regimes transition toward more democratic systems, they experience high levels of political violence. However, the connection between strength of democracy and homicide is not significant.

Fragile States

A good deal of terrorism research has examined connections between terrorist attacks and fragile or weak states (Reference LaFree, Dugan and FaheyLaFree, Digan & Fahey, 2008; Reference PiazzaPiazza, 2008b), however, I am unaware of any similar studies of cross-national homicide. In general, fragile states are defined as those that experience prolonged periods of civil conflict and war, political crises, and massive human rights violations (Reference Esty, Goldstone, Gurr, Surko and UngerEsty et al., 1995: 1). Most of this research concludes that fragile states provide an ideal environment for terrorist organizations (Reference Savun and PhillipsSavun & Phillips, 2009; Reference BellBell, 2017). Nonetheless, not all research is supportive of this association (Reference Eubank and WeinbergEubank & Weinberg, 2001; Piazza, 2008b).

I examine the Fragile States Index (Reference Messner de Latour, Haken and TaftMessner de Latour et al., 2020), which is based on twelve conflict risk indicators that measure the stability of a state at any given moment, including group grievance, economic decline, human rights, and rule of law and external intervention. According to the bivariate results in Table 8, both terrorist attacks and homicides are significantly more common in states experiencing high fragility rates.

Religious, Ethnic and Linguistic Diversity

Religious motives have frequently been examined as a cause of terrorism (Reference SternStern, 2002; Reference JuergensmeyerJuergensmeyer, 2003), and various measures of religion have also been used in studies of cross-national homicide (Reference Hansmann and QuigleyHansmann & Quigley, 1982; Reference Groves, McCleary and NewmanGroves, McCleary & Newman, 1985). Reference Egger and Magni-BertonEgger and Magni-Berton (2019) use survey data collected in twenty-one European countries to investigate the link between religious beliefs and terrorism justifications among European Muslims. They find that in countries affected by homegrown terrorism, Muslims who adhere to frequent religious practice show high levels of support for terrorism. In a study of worldwide terrorist attacks, Reference Piazza and LaFreePiazza and LaFree (2019) find that compared to attacks by nonreligious groups, attacks carried out by religious groups produce more casualties. Overall, prior research agrees that there are no consistent links between specific types of religions and propensity to engage in terrorism (Reference Kurrild-Klitgaard, Justesen and KlemmensenKurrild-Klitgaard, Justesen & Klemmensen, 2006; Reference Krueger and LaitinKrueger & Laitin, 2008).

Due to the difficulty of distinguishing religious from secular motivations, other researchers have focused on the relationship between measures of ethnic or linguistic fractionalization and homicide (Reference Avison and LoringAvison & Loring, 1986; Reference Krahn, Hartnagel and GartrellKrahn, Hartnagel & Gartrell, 1986) and terrorist attacks (Reference PapePape, 2005; Reference Moghadam and PedhazurMoghadam, 2006). This research reports mixed results. Whereas Reference Kurrild-Klitgaard, Justesen and KlemmensenKurrild-Klitgaard Justesen and Klemmensen (2006) find no significant association for measures of ethnic and linguistic fractionalization and terrorist attacks, Reference 88PiazzaPiazza (2011) and Reference Foster, Braithwaite and SobekFoster, Braithwaite and Sobek (2013) find that countries with a more ethnically homogeneous population generate fewer terrorist attacks. According to Reference PiazzaPiazza (2012), minority socioeconomic discrimination is a particularly important predictor of terrorist activity. Reference Basuchoudhary and ShughartBasuchoudhary and Shughart (2010) show that high levels of ethnic tension are associated with high levels of terrorism. Reference Choi and PiazzaChoi and Piazza (2016) find that domestic terrorism occurs more frequently in countries that exclude certain ethnic groups from political power. Reference Python, Brandsch and TskhayPython, Brandsch and Tskhay (2017) investigate the link between ethnic tensions and terrorist attacks and find that areas with high levels of ethnic polarization experience more terrorist attacks.

Taken together, these studies point to links between identity-related measures and terrorist outcomes. However, as Reference Krieger and MeierrieksKrieger and Meierrieks (2012) note, many of these studies have measurement weaknesses. For example, several studies use time-invariant proxies to measure identity conflict and ethno-demographic inequality (e.g., indices for ethnic and religious fractionalization) or rely on variables indicating simple proportions (e.g., the share of the Muslim population). It is unclear to what extent these measures adequately capture tensions rooted in identity conflict, ethno-demographic diversity or group-specific grievances.

To measure fractionalization, I use data obtained from the Historical Index of Ethnic Fractionalization Dataset (Reference DražanováDražanová, 2019), which defines fractionalization as the likelihood that two randomly selected individuals drawn from the population belong to two different ethnic groups. According to Table 8, high ethnic fractionalization is associated with high frequencies of both terrorist attacks and homicides.

Globalization

Reference Bove and BöhmeltBove and Böhmelt (2016) and others (Reference Blomberg and HessBlomberg & Hess, 2006; Reference Brockhoff, Krieger and MeierrieksBrockhoff, Krieger & Meierrieks, 2015) assess effects of globalization on terrorist attacks through increased migration flows, international cooperation and advances in educational attainment. Several studies (Reference Forrester, Powell, Nowrasteh and LandgraveForrester et al., 2019; Reference McAlexanderMcAlexander, 2020) have analyzed the relationship between immigration and political violence, finding mixed results. Reference Bove and BöhmeltBove and Böhmelt (2016) find that migrant inflows generally reduce the number of terrorist attacks. However, terrorism does diffuse to other countries through migrants leaving from terrorist-prone states. Reference Choi and SalehyanChoi and Salehyan (2013) show that countries with many refugees are more likely to experience terrorist attacks. However, drawing on bilateral migration data for 170 countries from 1990 to 2015, Reference Forrester, Powell, Nowrasteh and LandgraveForrester et al. (2019) find no evidence in support of the thesis that immigrants import terrorism.

With few exceptions (Reference LevchakLevchak, 2015; Reference LaFree and JiangLaFree & Jiang, 2022), globalization has not been examined in cross-national studies of homicide. However, Levchak finds partial support for some forms of globalization and LaFree and Jiang find strong support for the conclusion that as worldwide globalization increases (measured as trade openness), homicide declines. According to Table 8, high levels of globalization (measured here as trade openness) are significantly related to a low frequency of both terrorist attacks and homicides.

Population Size

Most prior studies find that compared to countries with fewer people, more populous countries experience higher levels of terrorist activity (Reference Plümper and NeumayerPlümper & Neumayer, 2010; Reference CocciaCoccia, 2018). Some scholars (Reference Freytag, Krüger, Meierrieks and SchneiderFreytag et al., 2011; Reference Ezcurra and PalaciosEzcurra & Palacios, 2016) argue that large populations correlate with high levels of demographic stress, which in turn fosters conflict and violence. By contrast, Reference Krieger and MeierrieksKrieger and Meierrieks (2012) claim that terrorist attacks may simply be more likely to occur in large countries.

Population size has also been included in many cross-national studies of homicide (Reference MessnerMessner, 1982; Reference ShichorShichor, 1990) and usually does not have a significant effect. For example, this is the conclusion reached in Reference NivetteNivette’s (2011) meta-analysis of important predictors of homicide. However, most cross-national studies of homicide are based on rates rather than counts. Thus, in these studies, population size is already built into the dependent variable. By contrast, Table 8 compares total frequencies of terrorist attacks and homicides. In the analysis of frequencies, population size is significantly associated with both high frequencies of terrorist attacks and homicides.

Conclusions: Determinants of Worldwide Terrorism and Homicide

Taken together, the results in Table 8 suggest that comparisons of the determinants of worldwide terrorism and homicide might be a fertile area for future research. Of the ten variables examined, six were statistically significant and in the same direction in predicting both the frequency of terrorism and homicide: five positive and one negative. Both terrorist attacks and homicides were more common when countries had a high proportion of young persons, high GDP, had a fragile or weak state, had high ethnic fractionalization, and had large populations. Both terrorist attacks and homicides were less common when countries had high levels of globalization. Two variables were significant for predicting homicide but had no significant effect on the frequency of terrorist attacks: countries with high inequality and a high urban percentage had a high frequency of homicides. Strength of democracy had a significant positive effect on terrorist attacks but not on homicides. Only the measure of poverty had no significant effect on either terrorist attacks or homicides.

In general, the ten macro-level variables examined here have been used more frequently in the study of terrorist attacks than homicides. Given this, it is interesting that six of these variables are significantly associated with homicides. As noted in previous sections, a major difference between terrorist attacks and more ordinary types of crime is that the former have a weaker connection to economic well-being than the latter. In the bivariate comparisons, one variable that is related to homicides but not terrorist attacks is inequality. Meanwhile, poverty had no significant effect on either terrorist attacks or homicides. I offer these bivariate comparisons as a first step toward a more exhaustive multivariate comparison of terrorist attacks and homicides.