HLSS522Wk2

Extremists and unconventional weapons: examining the pursuit of chemical and biological agents Thomas R. Guarrieria and Collin J. Meiselb

aNational Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland, College Park, MD, USA; bFrederick S. Pardee Center for International Futures, Josef Korbel School of International Studies, University of Denver, Denver, CO, USA

ABSTRACT In this study, we examine the individual-level characteristics of extremists’ pursuit of chemical/biological (CB) agents. Using three different maximum likelihood estimation techniques, we identify three key findings. First, older extremists are more likely to pursue CB than younger extremists. Second, extremists who are jobless or students are more likely to pursue CB than employed extremists. Third, Islamist, far-right, and far-left extremists are less likely to pursue CB than single-issue extremists. We do not find any evidence that gender or education have an effect on whether an extremist will pursue CB agents. Since there has been little quantitative examination of unconventional weapon choices among violent extremists, this study makes an important contribution to the literature on CB adversaries.

ARTICLE HISTORY Received 27 February 2019 Accepted 25 November 2019

KEYWORDS Terrorism; extremism; CBRN; unconventional weapons

Introduction

On 9 March 2011, in the quiet town of Addy, Washington, population 265, Kevin William Harpham drove past a road construction crew in rural Stevens Country. The construction crew, an undercover SWAT team with the Federal Bureau of Investigation, quickly commenced their operation and apprehended the 36-year-old Army veteran and avowed white supremacist. Harpham, the attempted MLK-Day backpack bomber who had threatened Spokane earlier that year, was finally in custody (CBS News, 2011; Morlin, 2011a).

Nearly seven months later, and roughly 2000 miles to the northwest, Mary Ann Morgan, an anti-government ideologue, slowed her truck as she approached the U.S.-Canada border crossing near Port Alcan, Alaska. Upon informing border guards of her possession of a semiautomatic handgun, Morgan’s journey quickly came to an end as Canadian law enforcement agents discovered a collection of instructions on ricin production and notes on bomb-making in her truck’s cab (Hopper, 2011). Although Morgan claimed she was driving a friend to Minnesota, the true destination of the 53-year-old secretary of the Alaska Peacemakers Militia was unknown (Burke, 2011; Morlin, 2011b).

© 2019 Society for Terrorism Research

CONTACT Thomas R. Guarrieri tguar@umd.edu National Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland, PO Box Number 266, 5245 Greenbelt Rd., College Park, MD 20740, USA

Supplemental data for this article can be accessed https://doi.org/10.1080/19434472.2019.1698633 This article has been republished with minor changes. These changes do not impact the academic content of the article.

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 2021, VOL. 13, NO. 1, 23–42 https://doi.org/10.1080/19434472.2019.1698633

The stories of Kevin William Harpham and Mary Ann Morgan are similar in that both individuals subscribed to extremist ideologies and evidence suggests that both were plan- ning an attack. However, there is an important difference between these two would-be terrorists. While Harpham’s weapon of choice was an improvised explosive device in the form of a backpack bomb, Morgan sought to use the biological toxin ricin. Are there any identifiable demographic characteristics that could indicate Morgan’s pursuit of ricin compared to Harpham’s choice of a conventional IED? Extending this question beyond these two individuals, what are the demographic and ideological predictors of extremists’ pursuit of unconventional weapon modalities, such as weapons that use chemical or biological agents?

While recent efforts have been made to distinguish between violent and non-violent extremist activity (LaFree, Jensen, James, & Safer-Lichtenstein, 2018) as well as weapon choices among chemical, biological, radiological, and nuclear (CBRN) adversaries (Binder & Ackerman, 2015), there has been little examination of the indicators differentiating extremists who pursue unconventional weapons from extremists who pursue conven- tional weapons. In this study, the first of its kind to our knowledge, we examine the demo- graphic and ideological factors that might predict whether certain violent extremists will pursue chemical/biological (CB) agents for use in a plot.

While we believe this research makes important contributions to the literature, it is important to note that we do not believe this research will serve as a conclusive assess- ment of predictors of individuals and unconventional weapons use. We recognize that there are many additional factors involved in predicting weapon selection that we do not control for in this exploratory study.1 The merits of this study rest on the novel theor- etical and quantitative contributions to the literature in which we derive testable hypoth- eses from grounded theoretical underpinnings and utilize statistical techniques not previously employed by other studies of non-state adversaries and unconventional weapons. We aim for this study to act as a starting point for future explorations employing maximum likelihood estimation techniques but not an ending point that claims our findings to be indisputable.

We separate our manuscript into four sections. The next section examines the current state of the literature on extremism and radicalization, as well as the literature on uncon- ventional CBRN adversaries. The third section lays out our theories and hypotheses. The section thereafter describes our research design and provides information on our data and variables. The final section displays our results.

Literature

There has been much interesting research on extremists in terms of the radicalization process. The processes of extremist radicalization, and even the exact definitions of the terms radicalization and extremist themselves, are debated (Borum, 2011a; Laqueur, 2000; Maskaliūnaité, 2015). However, scholars generally agree that radicalization involves a pathway along which an individual internalizes a set of extremist beliefs (Borum, 2011b; Jensen, Seate, & James, 2018). The precise contours of an individual’s pathway and their resultant extremist beliefs are determined by an array of micro-, meso-, and macro-level factors that guide them toward extremism (Jensen et al., 2018; Lee, 2011; Maskaliūnaité,

24 T. R. GUARRIERI AND C. J. MEISEL

2015). At the micro-level, individuals’ characteristics, beliefs, and experiences serve as drivers of extremism.

For the meso-level factors, it is no surprise that an individual’s immediate surroundings affect their propensity to radicalize. For example, Pels and de Ruyter (2012) find that the presence of radical peers increases the likelihood a child in the Netherlands will hold extre- mist beliefs themselves. The results of LaFree et al. (2018) highlight the existence of this effect among ideological extremists in the United States, showing that having radical peers increases an extremist’s propensity for violence by as much as 140% compared to those without radical peers. In fact, they assert that the presence of ‘radical peers is the strongest predictor of violence’ among ideological extremists in the U.S. from 1948 to 2013 (LaFree et al., 2018, p. 20).

For the macro-level conditions, extremist groups’ reactions to government can affect the propensity for violence. For example, the far left in the U.S. has taken a predominately non-violent turn since the 1970s (Miller, 2017) following decades of government crack- downs (Cunningham, 2003). Conversely, a forceful federal response to a far-right extremist at Ruby Ridge in 1992 has inspired anti-government attacks in the United States for over two decades (Berger, 2012), spanning from the 1995 Oklahoma City bombing, to 2017’s attempted Oklahoma City copycat, Jerry Varnell (Fernandez, 2017). Across all ideologies, scholars Shellman, Levey, and Young (2013, p. 331) present the notion of ‘phase shifts,’ where groups modulate their degree of violence based on a series of factors, including government repression.

Together, micro-, meso-, and macro-level factors help predict an individual’s vulner- ability to radicalization and their propensity for violence. However, the factors that predict an individual’s chosen modality of violence remains a mystery. One effort to capture extremist attack types is from Gill et al. (2017). They observe that the Internet helps familiarize extremists with tools and tactics with which they personally have no experience, and this could lead individuals to attack modalities of greater technical sophis- tication. Although Gill et al.’s (2017, p. 113) findings highlight a relationship between Inter- net usage and ‘[t]echnically more difficult attacks,’ their notion of technical difficulty includes improvised explosive devices, which are not CBRN weapons. While it is certainly true that extremists’ ‘[m]otivations can be fickle at the individual level’ (Hegghammer, 2013, p. 12), this gap is characteristic of the broader literature, which lacks an explanation of the factors that predict use of unconventional weapons. Our work aims to close this gap by tying the lessons learned from studies on radicalization and violent extremism to the existing body of work on unconventional weapons use by non-state adversaries.

In the literature on unconventional weapons there have been advances in the study of CBRN terrorism over the past decades. Hurwitz (1982) examines the possibility of CB use among terrorists, whereas later work by Simon (1989) investigates why terrorist groups might, or might not, use biological weapons. In the seminal book Toxic Terror, edited by Jonathan Tucker (2000), contributors review historical cases of terrorist use of chemical and biological weapons. These works contributed to the basis of future studies, such as those employing CBRN terrorism incident datasets.

In an effort to identify the individual-level characteristics of adversaries who pursue unconventional weapons, many researchers have used data to conduct useful studies to advance the literature. Ackerman and Pinson (2014) conduct a study comparing the characteristics of CBRN pursuit by lone actors and autonomous cells (LA/ACs) to those

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 25

characteristics displayed by formal terrorist organizations. In another study, Ellis (2014) explores the capabilities of lone actors in obtaining CBRN materials that can be weapo- nized. He also examines strategies and counter-measures to prevent individuals from acquiring materials needed to build weapons of mass destruction. Other studies specifi- cally focus on chemical and biological adversaries (Ackerman et al., 2014), while chemical, biological, and RN weapon pursuit and use by terrorist organizations has been a topic of investigation using quantitative methods (Asal, Ackerman, & Rethemeyer, 2012; Campbell & Murdie, 2018).

In a recent article, Lindekilde, O’Connor, and Schuurman (2019) explore the link between radicalization and patterns of attack, finding that different patterns of radicaliza- tion are linked with different modes of attack planning. Along with recent datasets on CBRN actors, such as the Profiles of Incidents involving CBRN and Non-State Actors (POICN) database (Binder & Ackerman, 2019), such studies continue to advance the scien- tific understanding of extremists and unconventional weapon use. Our effort to determine the significance of individual-level factors predicting whether extremists will pursue unconventional weapon modalities builds upon these studies filling a current void in both the literature on extremist radicalization and unconventional weapons use by non- state actors.

Theory and hypotheses

Despite the gap in the literature examining the predictors of an extremist’s pursuit of unconventional weapons, there are numerous theories that inform testable hypotheses regarding CB weapon selection. We draw all of our theoretical expectations for predictors of CB pursuit from scholarly research and empirical trends. While our study addresses a gap in the literature, there are, nonetheless, numerous informative studies that we use to derive our hypotheses.

First, given that age exhibits a strong correlation with both criminal activity (LaFree et al., 2018) and terrorist activity (Klausen, Morrill, & Libretti, 2016) it could be a predictor of extremist pursuit of unconventional weapons. Following a ‘terrorist-age crime curve,’ as characterized by Klausen et al. (2016, p. 30), an individual’s propensity to engage in terrorist activity increases as they age into their 20s before dramatically reducing after the age of 35. While these authors identify this pattern among Islamist extremists, similar patterns are evident among the far left in radical environmental and animal rights groups (Carson, LaFree, & Dugan, 2012) and among single-issue nationalists in India’s state of Bengal (Lee, 2011).2

In reference to extremist weapon selection, one theory is that older extremists are more likely to pursue CB weapons relative to younger extremists because they have more lived experiences, knowledge, and insight that they can reference in determining an attack modality. There is high investment required to acquire and weaponize CB material com- pared to conventional weapons (Maurer, 2009). Older extremists, given the more time and experience they could have had within their ideological circles, could be more likely to pursue a CB attack modality because they have experience to guide more intricate or sophisticated weapons. Moreover, older extremists would have had more years of exposure to different modalities of attack, and this could increase their likelihood of

26 T. R. GUARRIERI AND C. J. MEISEL

opting to pursue an unconventional weapon. Based on these theoretical underpinnings, we hypothesize that age will be a predictor of CB pursuit among extremists.

H1: If an extremist is older, then there is a higher likelihood of them pursuing CB relative to younger extremists.

Gender is another variable that is a predictor of certain social and political behaviors. However, as Sjoberg (2011) cautions, it is important to clarify that any such differential effects between genders are likely due to culturally and organizationally imposed gender roles, which is a phenomenon Schils and Pauwels (2014, p. 45) describe as ‘[g]endered socialization.’ For example, women are often relegated to what are viewed by an extremist organization as non-combat roles (Talbot, 2000) while young extremists are often men (Jarman, 2004; Lee, 2011). Yet, at least where white supremacist groups are concerned, Blee (2005) notes there has been an increasing number of women since the turn of the 21st century who have sought empowerment through extremist violence. She notes that one of the characteristics of female extremists in such white supremacist organizations is that they generally maintain indirect involvement. The indirect involve- ment of females in racial terrorism comports with studies that show that women are more inclined toward indirect aggression and violence in general (Crick & Grotpeter, 1995; Fesbach, 1969). As one female scholar, Tracy Vaillancourt, states explicitly, ‘when we aggress against somebody, we do it indirectly’ (Davis, 2013) and this is known in the academic literature as ‘indirect aggression’ (Vaillancourt, 2013, p. 1). This is not to say that females do not have active roles in terrorist groups, only that, according to some studies, there might be differences, on average, in the types of violence employed.

In reference to CB weapon modalities, female extremists’ inclination toward indirect aggression could make them likely to pursue CB compared to their male counterparts because CB agents can be used to attack targets without direct confrontation. In other words, unlike the use of a handgun, the use of chemical or biological agents to conduct an attack can be accomplished indirectly; that is, the attacker need not come into direct contact with the target. Given this, attack modalities using CB agents could be more attrac- tive for females. Gender, therefore, could be an indicator of extremist pursuit of CB.

H2: If an extremist is female, then there is a higher likelihood of them pursuing CB relative to males.

Education may also be a predictor of extremist pursuit of CB. As Hoffman (2006, p. 249) states, ‘solid training… and technological competence are the essential prerequisites for a successful operation.’ Thus, extremists with a higher level of education could be more likely to pursue CB because advanced training in chemistry, biology, and/or physics often occurs in a university setting. While there are many infamous cases of less educated individuals pursuing CB agents, we hypothesize that this is not the case in the aggregate. While similar, although more basic, knowledge may be acquired at the high school level, the types of weapons that are likely to be produced (e.g. chlorine gas) require mass quantities that are difficult to conceal (Meulenbelt & Nieuwenhuizen, 2015).

An example of education likely playing a role in unconventional weapon selection is the case of Ahmed Abassi. According to the Chemical and Biological Non-state Adversaries Database (CABNSAD) (National Consortium for the Study of Terrorism and Responses

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 27

to Terrorism, 2017), Abassi was a graduate student working toward a Master’s degree in chemical engineering and therefore likely possessed advanced knowledge in chemicals and had access to expertise in the natural sciences. Motivated by Islamist ideology, there is evidence that he expressed a desire to plot a terrorist attack using an uncon- ventional weapon with the goal of causing mass casualties. Abassi is alleged to have suggested the use of biological agents disseminated through the air or water. Before the plot could proceed further, he was charged with ‘knowingly making false statements to immigration authorities to facilitate an act of terrorism’ and deported from the United States (National Consortium for the Study of Terrorism and Responses to Terrorism, 2017).

The case of Ahmed Abassi provides a good example of how education can play a role in determining the weapon modality for extremists. Extremists who are more highly edu- cated are not only exposed to greater opportunities to network with individuals who possess knowledge and expertise in CB agents but also are likely to acquire the critical thinking skills during their education needed for more elaborate plots. Interestingly, despite Abassi’s background in chemical engineering, evidence suggests that he wanted to use a biological agent. One reason for this could be that Abassi believed it would be possible to kill more people with a biological agent than a chemical agent, thus relying on his knowledge of chemicals to make such an assessment. Notwithstanding, we hypothesize that there is a relationship between education and CB pursuit.

H3: If an extremist is more educated, then there is a higher likelihood of them pursing CB rela- tive to less educated extremists.

One of the most oft repeated theories in terrorism is that employment status plays a role in whether individuals will engage in violence. Researchers find that under-employment – the combination of an advanced education and poor employment prospects – helps drive an individual toward committing terrorist acts (Klausen et al., 2016; Lee, 2011). Unem- ployed chemists, biologists, and physicists may pose a particularly potent threat with respect to terrorist acts using unconventional means (Gunaratna, 2002; Meulenbelt & Nieu- wenhuizen, 2015; Zwolski, 2011). Therefore, education and employment are both factors that should be considered in research on extremists. Since CB takes ‘large investment of capital and human resources’ (Maurer, 2009, p. 61), if extremists are jobless then this could be a predictor of a higher likelihood of them pursuing CB relative to employed extre- mists because of the additional time they have available for preparing such attacks. This theory is bolstered by research that finds that unstructured time is associated with violent extremism (Becker, 2019). Extremists who are employed, particularly in jobs that require at least 40 hours of work per week, do not have as much time to spend on researching CB materials and methods of attack.

Moreover, full-time terrorists – individuals whose sole vocation is terrorist activity – have the time to invest in acquiring or developing a weapon that requires additional effort compared to conventional munitions. These are individuals who are not employed in the licit workforce and who can dedicate themselves fully to the terrorist cause in lieu of legal employment. Jobless extremists can also be terrorists who are leaders of organiz- ations and therefore have the resources needed to pursue CB. Thus, we hypothesize that there is a relationship between employment and extremists’ pursuit of unconven- tional weapons. We consider full-time terrorists as jobless because full-time terrorists,

28 T. R. GUARRIERI AND C. J. MEISEL

similar to drug traffickers whose sole income is from illicit means, are not in the legal work- force. Likewise, students are outside of the active workforce even if they do have a job, and we believe that students also have increased time to dedicate to unconventional weapons. Thus, we treat jobless extremists and student extremists as more likely to pursue CB rela- tive to employed extremists.

H4: If an extremist is jobless or a student, then there is a higher likelihood of them pursuing CB relative to employed extremists.

Ideology also influences the actions of extremist groups and their members and therefore may also predict whether extremists will pursue unconventional weapons. There is evi- dence that violent extremists’ tactical decisions (and thus also weapon choices) are par- tially driven by their ideology (Hoffman, 2006). For acts of terrorism, where intergroup competition is concerned, Nemeth broadly asserts that left-wing organizations respond with less terrorism, while religious and nationalist terrorist groups respond with more ter- rorism (Nemeth, 2014). Carson et al. (2012) find that radical environmental and animal rights groups tend to prefer non-violent activism (e.g. property damage) and, in general, avoid physical harm to others but this is not the case for numerous other ideological groups. Far-left terrorists typically pursue symbolic violence (Hoffman, 2006), while some white supremacists maintain a preference for interpersonal violence (Simi & Windisch, 2018).

According to a study of adversaries employing CB material, Ackerman and Binder (2015) find personal/idiosyncratic motives to be the leading non-criminal motivation of perpetra- tors employing CB. Of 163 high-certainty perpetrators – cases in which the criteria for inclusion in the dataset requires a higher level of certainty – 24% of perpetrators held per- sonal/idiosyncratic motives, which was the highest non-criminal motive for employing CB.3 While this percentage does not hold in some samples of the dataset that include cases in which there is uncertainty about the perpetrator’s inclusion, Ackerman and Binder (2015, p. 13) state that the data analysis ‘does make clear that, of those with non-criminal motives for employing CB, the two leading motivations across the samples were personal/idiosyncratic motives and religious ideologies,’ with personal/idiosyncratic motives being the highest, or tied for the highest, when chemical cases and biological cases are examined separately.

According to scholars of extremist radicalization, we can consider actors with personal/ idiosyncratic motives as single-issue extremists since ‘single-issue extremists are individ- uals that are motivated primarily by a single issue, rather than a broad ideology,’ and this includes ethnonationalists and extremists with idiosyncratic ideologies, such as Ted Kaczynski (the so-called ‘Unabomber’) (National Consortium for the Study of Terrorism and Responses to Terrorism, 2018, p. 13). Given the findings of descriptive studies regard- ing the ideologies of CB adversaries, single-issue extremists could be more likely to pursue CB compared to other ideological extremists. Therefore, we expect that single-issue extre- mists will have a higher likelihood of pursing CB compared to extremists with alternative ideologies. Concordantly, we hypothesize that Islamist, far-right, and far-left extremists will have a lower likelihood of pursuing CB compared to single-issue extremists.

H5: If an extremist is Islamist, Far right, or Far left, then there is a lower likelihood of them pursing CB relative to Single-issue extremists.

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 29

Research design

Data

To evaluate these hypotheses we utilize data from two datasets. The dataset from which we sample non-CB radicalized extremists is the Profiles of Individual Radicalization in the United States (PIRUS) dataset (National Consortium for the Study of Terrorism and Responses to Terrorism, 2018). The dataset from which we draw our sample of individuals who pursued CB is from the Chemical and Biological Non-state Adversaries Database (CABNSAD) (Ackerman & Binder, 2015; National Consortium for the Study of Terrorism and Responses to Terrorism, 2017). Our sample from both of these datasets includes lone actors and individuals who are members or leaders of terrorist groups or cells.4

PIRUS version 3.0 includes a random sample of non-state actors who radicalized into an extremist ideology within the United States and were exposed for committing, or attempt- ing to commit, a plot during the years 1948 through 2016. PIRUS data were recorded from open sources. These sources include ‘court documents, online news articles, newspaper archives, open-source non-government reports (e.g. the Southern Poverty Law Center), unclassified government reports (e.g. annual FBI terrorist reports), and existing terror- ism-related data sets (e.g. the Global Terrorism Database)’ (LaFree et al., 2018, p. 12). Indi- viduals who are coded in both CABNSAD and PIRUS were removed from the PIRUS dataset as to not double count perpetrators.5

CABNSAD includes all known use, acquisition, or attempted use or attempted acqui- sition of chemical and biological weapons by non-state actors worldwide as found in open sources. The dataset includes ideologically and non-ideologically motivated actors, but excludes the following: state actors, unwitting actors (i.e. individuals who were unaware they possessed CB agents), smugglers, hoaxers, acts of interpersonal violence using unsophisticated means (e.g. rat poison rather than higher-end agents like ricin), and acts including chemical agents used for their explosive or corrosive qualities (National Consortium for the Study of Terrorism and Responses to Terrorism, 2017). For the purposes of this study, all non-ideologically motivated actors were omitted from our sample since we are focused on actors with extremist ideologies. Moreover, since our theories focus on the characteristics of extremists who decide to pursue CB, we limit our sample to per- petrators coded as the decision-makers.6

While it may be considered unorthodox to combine two separate samples in one analy- sis, this method has gained traction within the field of epidemiology (e.g. Blettner, Sauer- brei, Schlehofer, Scheuchenpflug, & Freidenreich, 1999; Bower et al., 2003; Collaborative Group on Hormonal Factors in Breast Cancer, 1996) and, to some extent, household survey analysis (e.g. Hanna, Cordery, Steel, Davis, & Harrold, 2017; Wendt, 2007). We con- sider the combination of two samples valid here given the similarity of the data collection efforts and the intended use of the datasets in terrorism studies research. Indeed, both PIRUS and CABNSAD relied on English-language, open-source materials which were scanned as comprehensively as possible by researchers for a complete set of cases of inter- est.7 Although CABNSAD’s focus is global and PIRUS’s is United States-centric, we do not have theoretical reasons why the individual-level indicators of CB pursuit would be different between the samples.8

While dataset differences admittedly pose a challenge to our analysis, our sentiment is similar to that of Bower et al. (2003, p. 5):

30 T. R. GUARRIERI AND C. J. MEISEL

Such methodological compromises mean that the present analysis can never approach the precision of primary data collection. However, it is a matter of debate (and economics) whether this additional precision is worth the extra cost and time delay… or whether secondary analysis provides a reasonably accurate estimate to inform policy and practice.

Aside from its reasonable accuracy, a quantitative analysis of these data is worth pursuing given the limitations of previous qualitative studies. Qualitative designs, while adept at tracing processes like radicalization, often select on the dependent variable (King, Keohane, & Verba, 1994). Multiple regression can fill this gap and act as a supplement to inferences made by prior qualitative studies (Collier, 1995). Additionally, our study offers the ability to separate out confounding factors and mediating effects (Miller, 2005). Given the importance of addressing these crucial issues, we move forward with this exploratory research, acknowledging that what follows amounts to a starting point – hopefully the first of many as the threat of CB weapons persists, and perhaps even grows (Yeo, 2015).

Dependent variable

For our analysis, we use a sample of 180 individuals drawn from PIRUS and CABNSAD. The dependent variable is binary, coded as 0 if there is no evidence of the extremist pursuing CB and coded as 1 if there is evidence that the extremist pursued CB. The distribution of the variable shows that there are far fewer cases in which an extremist pursued CB as opposed to a conventional weapon. Of the 180 extremists, 34 are coded as having pursued CB agents and 146 extremists are not coded as CB actors. Thus, 18.89% of the sample are CB actors.

Independent variables

To evaluate our hypotheses, we use multiple independent variables. Our first hypothesis examines the relationship between an extremist’s age and pursuit of CB. To evaluate this hypothesis, we include a variable for age that we treat as continuous. In the sample, the age of extremists ranges from 15 years old to 71 years old, with a mean age of 32. Despite the range of the age variable, the distribution is skewed, as more than 50% of the sample is 29 years of age or younger. For our second hypothesis, we include a variable for gender. This is a binary variable coded as 0 if the extremist is female and coded as 1 if the extremist is male.9 There are 7 females and 173 males in the sample. Given the highly skewed distribution, the data initially suggest that the average violent ideological extremist is a young man.

For our third hypothesis, we include an ordinal variable for education. This variable con- sists of three levels and corresponds to the highest level of education an extremist attained. Extremists whose education is high school or less are coded as 1, extremists who attended an undergraduate college are coded as 2, and extremists who attended a postgraduate school are coded as 3.10 Over half of the individuals in our sample had at least begun pursu- ing a college education (92 are coded as having attended undergraduate college and 25 are coded as having attended a postgraduate school). Of the remaining extremists, 63 are coded as high school or less.

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 31

Our fourth hypothesis focuses on the association between employment status and pursuit of CB agents. We code our employment variable as a categorical measure with three categories: employed, jobless, and student.11 In the sample, 90 extremists were employed at the time of their violent incidents, which is 50% of the cases. Of the remaining individuals, 61 were jobless (33.89% of the sample) and 29 were students (16.11% of the sample). Since this is a categorical variable, we create dummy variables (called Employed, Jobless, and Student) for each of these categories, as to exclude one of the categories from our analysis to mitigate issues of perfect collinearity.

Our fifth hypothesis looks at the role of ideologies on violent extremists’ pursuit of CB agents. For this variable, we manually coded the ideologies of the individuals included in CABNSAD according to the four general categories found in PIRUS: single issue, Islamist, far left, and far right. In non-obvious cases, we performed additional research into the individ- ual actor, or their group, and we discussed which ideology most closely matched the per- petrator’s affiliated group, beliefs, and/or actions. Individuals with ideologies that could not be categorized were excluded from our sample. For this variable, 75 actors have an Islamist ideology, which is the highest category in the sample (41.67%), while the second highest category consists of far-right actors, of which there are 46 cases (25.56%). There are 39 extremists who fall into the category of single issue (21.67%), while the fewest number of cases consist of far-left ideological actors, of which there are 20 (11.11%). Since our ideology variable is categorical, we again create dummy vari- ables (called Single Issue, Islamist, Far Left, and Far Right) for each of these categories.

For theoretical reasons, we include a control variable called Group Member that is coded as 1 if the individual is in a group and coded as 0 otherwise. We include this control because individuals who are leaders or decision-makers in a group could have different characteristics than lone actors in terms of indicators predicting CB pursuit. For example, while higher education might be an indicator of CB pursuit among lone actors, this might not be the case for individuals in groups who choose to pursue CB because they can rely on other group members’ knowledge, expertise, resources, and support. Therefore, it is important to control for whether an individual is a decision- maker in a group.

For the sample drawn from PIRUS, we code an individual whose role is recorded as a ‘leader’ in a group as 1. However we exclude individuals whose group role is recorded as ‘follower’ or ‘loose associate’ from our sample since we cannot assume that these individuals were decision-makers involved in the choice of weapon to use in a violent plot. An individual whose role is recorded as ‘Not Applicable’ because the individual was not part of a group is coded as 0. For the sample drawn from CABNSAD, we code Group Member as 1 if the individual conducted their activities as a member of an organization or small cell and 0 if they are a lone actor. Out of the 180 cases, 96 individuals (53.33%) are not members of a group and 84 (46.67%) are coded as members of a group.

Model specification

We consider three different modeling approaches to best estimate our model. We employ maximum likelihood estimation using a logit link function as our main model. We also employ two alternative logit models that are specifically designed to address issues that

32 T. R. GUARRIERI AND C. J. MEISEL

arise from small sample sizes or skewed distributions of the dependent variable where there are fewer cases in which Y = 1. The two modeling types we use are the rare event logit model developed by King and Zeng (2001a) and penalized maximum likelihood esti- mation (PMLE) developed by Firth (1993).

The rare event logit model (RE logit) uses a bias-corrective technique that subtracts an approximation of the bias resulting from a small sample size from the coefficient estimates (Leitgöb, 2013). Bias becomes particularly problematic for maximum likelihood estimation when the proportion of events where Y = 1 is smaller than 5% (King & Zeng, 2001a). While our distribution of the dependent variable is not this severe, it nonetheless serves as a useful sensitivity analysis to employ alternative types of models given our sample size and skewed distributions. To address this bias, RE logit utilizes a weighted least squares regression to estimate the bias and calculates bias-corrected beta estimates. This bias term affects the constant term, which is also corrected using this method. As King and Zeng (2001b, p. 702) explain, ‘when the results make a difference, they are better than logit, and when they do not make a difference they give the same answer as logit.’

Another approach is penalized maximum likelihood estimation (PMLE). This method introduces a penalization term that is sensitive to smaller sample sizes and a lower pro- portion of cases where Y = 1 (Leitgöb, 2013). The PMLE method ‘allows convergence to finite estimates with very sparse data’ and also overcomes problems of separation (Coveney, 2008). This is a bias-preventative technique that ‘penalizes the likelihood [func- tion] by the Jeffreys invariant prior’ (Firth, 1993, p. 36). Effectively, PMLE modifies the esti- mation method in a manner that introduces a bias to counter the asymptotic bias that affects model estimation with small sample sizes.

Results

In Table 1, we present the results for each of the three estimation methods we employ. Taken together, the logit, PMLE, and RE logit models all consistently display the relatively small but statistically significant effect of Age on the propensity for violent ideological extremists to pursue CB weapons. Put simply, older violent extremists are more likely to pursue CB weapons relative to their younger counterparts. This provides support for our first hypothesis. Corroborating our theoretical expectations, age is one of the character- istics of extremists that affects weapon selection.

While our second hypothesis theorizes that gender has an effect on whether an extre- mist is a CB actor, we find no evidence to support this theory. The lack of statistical signifi- cance, however, could be a result of data limitations. Out of the 180 extremists in our sample, 7 are female and 173 are male. Given this extremely skewed distribution, it is unsurprising that this variable is statistically insignificant.

For our third hypothesis, we expect that more educated extremists will be more likely to pursue CB compared to less educated extremists. Similar to our hypothesis regarding gender, we do not find any evidence that education is significant to extremist pursuit of CB agents or weapons. One of the reasons that we might not find evidence in support of this hypothesis is the wide variation in education level among some of the extremists who pursue CB. While there are instances of extremists like Abassi, indicating that education may play a role in whether an extremist will be a CB actor, there are also many cases in which individuals with no higher than a high school education attempt

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 33

to acquire CB agents for a plot. Such examples of these include right-wing extremists with less than a high school education attempting to weaponize ricin.

Concerning our fourth hypothesis regarding employment status, both jobless and student extremists are more likely to pursue CB weapons relative to those who are employed. We theorize that jobless extremists and student extremists have more time to dedicate to the planning of an attack using unconventional methods, such as those employing CB agents. On the other hand, extremists who maintain legal employment and can only dedicate themselves to their extremist activities part time are more likely to pursue conventional attack modalities. We find evidence for this theory, as our results indicate that jobless and student extremists are more likely to be CB actors.

Our fifth hypothesis focuses on ideology, which we argue influences the actions of extremist groups and therefore should also influence the weapon selection of violent actors. In our analysis, we set single-issue extremists as the reference category since we hypothesize that they are more likely to pursue CB compared to other ideological extre- mists, such as those with Islamist, far-right, or far-left ideologies. Our findings provide support for our hypothesis. Our Islamist variable is statistically significant with a negative coefficient across all three modeling specifications at p < 0.10 threshold. Based on this result, we find some evidence that Islamist extremists are less likely to be CB actors com- pared to single-issue extremists. Likewise, the Far Left variable is also statistically significant and in the negative direction at p < 0.10 threshold. For our variable Far Right, we also find evidence for our hypothesis since this is in the negative direction and statistically signifi- cant at p < 0.05 threshold in the logit model and p < 0.10 threshold in the PMLE and RE logit models. While not as strong as the other findings we report, we nonetheless find ten- tative evidence to support our hypothesis that Islamist, far-left, and far-right extremists are less likely to be CB actors compared to single-issue extremists.

Table 1. Beta coefficients for logit, PMLE, and RE logit models. Variables Logit PMLE RE Logit

Age 0.054*** 0.048** 0.047** (0.020) (0.023) (0.018)

Gender −0.195 −0.268 −0.263 (0.726) (1.096) (0.688)

Education 0.091 0.083 0.082 (0.358) (0.366) (0.339)

Jobless 2.695*** 2.417*** 2.396*** (0.637) (0.568) (0.603)

Student 2.700*** 2.431*** 2.423*** (0.980) (0.878) (0.928)

Islamist −1.149* −1.051* −1.048* (0.592) (0.629) (0.561)

Far Left −1.806* −1.563* −1.538* (0.951) (0.851) (0.901)

Far Right −1.542** −1.387* −1.371* (0.741) (0.713) (0.702)

Group Member 1.772*** 1.616*** 1.607*** (0.561) (0.498) (0.532)

Constant −5.040*** −4.452*** −4.412*** (1.438) (1.656) (1.362)

Wald χ2 40.42 33.28 35.58 N 180 180 180

Note: p < 0.01***, p < 0.05**, p < 0.10*. Standard errors in parentheses. DV: CB Actor.

34 T. R. GUARRIERI AND C. J. MEISEL

For our control variable accounting for group membership, we also find a statistically significant relationship. In all three models, the variable is significant and in the positive direction. This indicates that members of an organization are more likely to pursue CB than individuals not in a group. While this is included as a control variable for theoretical reasons, this finding is unsurprising. Groups are likely to be more resource rich compared to a lone actor and can pool resources, and this could influence the pursuit of CB. Addition- ally, terrorist organizations provide a network that could increase the attractiveness among decision-makers and leaders of more complex plots, such as those that employ CB agents.

To examine the substantive effects of our models, we report odds ratios. The odds ratios for the logit model results are presented in Table 2. Calculated by exponentiating the logit coefficients (i.e. Euler’s number, e, raised to the power of the coefficient),12 odds ratios express the difference in the probability of the event (here, a violent extremist opting to pursue CB agents) relative to the probability of the non-event (here, a violent extremist not opting to pursue CB agents). For example, an odds ratio of 1 denotes an equal likelihood (i.e. 100% as likely), whereas an odds ratio of 2 indicates an event is twice as likely (i.e. 200%), and an odds ratio of 0.75 indicates an event is three quarters as likely (i.e. 75%). Odds ratios can also be read as more or less likely where, for example, an odds ratio of 0.75 would indicate that the event under consideration is 25% less likely than the non-event. For ease of interpretation when explaining substantive results, we report odds ratios around 1 in terms of percentages, while we report large odds ratios (odds ratios that equal a 1000% or greater increase/decrease) in terms of ‘x times more/less likely.’

Looking at the odds ratios in the logit model, we can determine the substantive effects of each independent variable on extremist CB pursuit. Beginning with age, there is a 5.5%

Table 2. Odds ratios for logit model. Variables Logit

Age 1.055*** (1.016–1.096)

Gender 0.823 (0.198–3.415)

Education 1.095 (0.543–2.210)

Jobless 14.803*** (4.248–51.584)

Student 14.878*** (2.179–101.561)

Islamist 0.317* (0.099–1.012)

Far Left 0.164* (0.025–1.061)

Far Right 0.214** (0.050–0.913)

Group Member 5.883*** 1.958–17.674

Constant 0.006*** (0.000–0.108)

Wald χ2 40.42 N 180

Note: p < 0.01***, p < 0.05**, p < 0.10*. 95% confidence intervals in parentheses. DV: CB Actor.

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 35

increased likelihood among violent extremists to pursue CB weapons for each additional year of age. The variables Gender and Education are not statistically significant and so we do not calculate substantive effects.13 It is important to note that increased variance due to multicollinearity is likely to have moderately reduced the statistical significance of each variable with uncentered variance inflation factors for Age, Gender, and Education of 11.48, 15.12, and 9.36, respectively.14

Both Jobless and Student have strong, positive effects on the likelihood of CB weapons pursuit relative to employed violent extremists. Jobless and student extremists were each roughly 15 times as likely to pursue CB agents relative to employed violent extremists. Employment status again demonstrates a large, significant (p < 0.01) effect on the propen- sity for CB agent pursuit. For ideology type, violent Islamist extremists were roughly 68% less likely than single-issue extremists (the reference category) to pursue CB agents. Far- left extremists, too, exhibited a roughly 84% decreased likelihood of CB agent pursuit rela- tive to single-issue extremists. Each category, however, is statistically significant at p < 0.10 threshold. Meanwhile, the propensity for far-right extremists to pursue CB agents is stat- istically significant with a 79% decrease in likelihood compared to single-issue extremists. Last, for the variable Group Member, our results indicate that a member of a group is roughly 6 times more likely to pursue CB than non-group members.

To aid in the assessment of the relative effects of each variable within and across each model, the odds ratios for the logit, PMLE, and RE logit models are depicted in Figure 1. We use a logarithmic scale along the horizontal axis for two reasons. First, mathematically, the unit of output of a logit function is log odds. Second, for interpretation, an odds ratio of 0.1 indicates a one-tenth relative likelihood (i.e. base odds divided by ten) and an odds ratio of 10 indicates a ten-fold relative likelihood (i.e. base oddsmultiplied by ten). The logarithmic scale transforms these values so that their distance from 1 on the horizontal axis is visually identical (see Hosseinpoor & Abouzahr, 2010). Point estimates are represented by the respective symbols indicated in the legend and confidence intervals are illustrated by the horizontal line passing through each symbol. The dashed vertical line at 1 on the horizontal axis indicates even odds – or no effect relative to the reference category – and confidence intervals crossing this dashed vertical line denote a variable’s lack of statistical significance at the 95% level of confidence.

Due to the properties of logs, readers interested in considering a different reference category need only to follow the subsequent steps to calculate each category’s new odds ratio.15 First, take the natural log of each category’s odds ratio to obtain its log odds coefficient. Then, subtract the coefficient of the new reference category (e.g. Far Right) from the coefficient of the variable of interest (e.g. Islamist). Finally, exponentiate the difference to obtain the new odds ratio.

Conclusion

Although there have been some scientific studies on adversaries who pursue unconven- tional weapons, this remains an inchoate literature. This exploratory study presents, to our knowledge, a first attempt to identify statistically significant individual-level predictors of extremists’ pursuit of unconventional weapon modalities, such as CB agents. Since there has been little examination of unconventional weapons choices among violent extremists, we believe this study makes an important contribution to the literature on CB adversaries.

36 T. R. GUARRIERI AND C. J. MEISEL

Using three different estimation techniques, we find that older extremists are more likely to be CB actors than younger extremists, extremists who are jobless or students are more likely to be CB actors than employed extremists, and Islamist, far-right, and far-left extre- mists are less likely to be CB actors than single-issue extremists. We do not find any evi- dence that gender or education have an effect on whether an extremist will pursue CB agents.

While we believe that this study is an important step in gaining a better understanding of violent extremists, there are future expansions of this research agenda that we believe would benefit scholarship on countering violent extremism. One avenue of future research includes utilizing more computationally advanced statistical techniques to evaluate hypotheses that we are not able to examine in this study due to data limitations. There are some interesting hypotheses we would like to evaluate that include interaction terms, but given the limitations of our sample, we would not be able to trust the results of maximum likelihood estimation. One method of handling missing data is to utilize imputation techniques. This would allow for testing of novel hypotheses that require a larger dataset with greater variation among the distributions of the variables.

Notwithstanding future avenues of research, our study provides results that scholars can build upon. Given the psychological effect that CB attacks have on a target population, it is all the more important that policy-makers have a better understanding of the charac- teristics that separate CB extremists from those planning conventional attacks. Over the

Figure 1. Odds ratios and confidence intervals for logit, PMLE, and RE logit models. Note: Odds are relative to the reference category: female, employed, single-issue extremists who do not belong to a group. Odds ratios are depicted on a logarithmic scale. Figure reports 95% confidence intervals. DV: CB Actor.

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 37

past decade, terrorism studies have been enhanced by more rigorous, scientific research, and as technology continues to progress, it is increasingly important for such research to focus on unconventional weapons.

Notes

1. Many of the limitations regarding additional control variables are a result of the paucity of data for CBRN actors. There are numerous additional variables related to internet usage, gun own- ership, personal relationships, and specific policy beliefs (to name a few) that could also be predictors of pursuit of unconventional weapons. Unfortunately, we were not able to find available data for these variables.

2. In the study for India, findings are not to the point of statistical significance when controlling for other factors.

3. According to the study, 26% of perpetrators had criminal motives. 4. For a non-leader member of a group, we only include cases in the sample drawn from

CABNSAD where they are coded as a decision-makers to pursue CB since this is our theoretical population of interest. Please see Footnote 6 for additional information about this distinction.

5. We verified this by comparing the two datasets looking for matched names, as well as key word searches through the PIRUS data for CB-related terms and phrases. We coded four indi- viduals with unknown, but possible, CB or RN weapon pursuits as unknown.

6. The CABNSAD codebook defines decision-makers in the following manner: ‘Perpetrator was involved in the decision to pursue/use CB. In the case of an organization, this may include organizational leaders.’ For organizational leaders, the dataset’s creators add:

As a general principle, organization leadership figures are included in CABNSAD where there is some indication that they played a part in decision-making related to the devel- opment or use of CB weapons. In the case of weakly structured organizations, or organ- izations in which the pursuit of a CB capability was undertaken on the independent initiative of a lower-level figure it may be inappropriate to include apex leadership (National Consortium for the Study of Terrorism and Responses to Terrorism, 2017, p. 9).

7. Note that once PIRUS researchers had assembled a complete set of cases, they then sampled from that set randomly.

8. However, given different cultural environments, there may be structural conditions within different states that affect extremist weapon choice. Given an already small sample of cases in the data, we do not investigate structural factors, but we anticipate that our theoretical expectations regarding individual-level demographic characteristics exclusively are generaliz- able to different countries.

9. This is not to say that gender is binary, but PIRUS and CABNSAD do not include data on extre- mists who identify as neither male nor female.

10. Given that the CABNSAD coding schema for education is less granular than PIRUS, we col- lapsed the PIRUS education categories to the four levels in CABNSAD, leaving the aforemen- tioned four levels.

11. We also collapse the PIRUS employment categories into the CABNSAD categories with the slight alteration of reclassifying unemployed individuals as ‘jobless,’ which leaves open the possibility that an individual was illicitly, or informally, employed. The datasets also included a category for ‘retired’ but there were no individuals in our sample that were coded as being retired.

12. e≈ 2.71828. The coefficients are exponentiated because the standard logistic regression output is expressed in log odds, or the natural log of the odds ratios.

13. Given the possibility that the relationship between Education and CB pursuit is non-monotonic and non-linear, as a robustness check we estimated an alternative model with indicator vari- ables describing the impact of each education level relative to ‘High School or less.’ Consistent with our results, the education dummy variables were not statistically significant.

38 T. R. GUARRIERI AND C. J. MEISEL

14. As a general guideline, variance inflation factors greater than 10 are considered to indicate high levels of multicollinearity, which can meaningfully impact statistical significance.

15. Note that this does not mean a variable’s effect changes. However, the relative odds do change because of the change in the baseline category.

Acknowledgments

The authors would like to thank Gary Ackerman, Markus Binder, Rebecca Bryan, Dennis Foster, and Patrick James for their invaluable contributions to this research, as well as the two anonymous reviewers and University at Buffalo’s Department of Political Science. The authors are solely respon- sible for any errors.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This research did not receive external funding.

Notes on contributors

Thomas R. Guarrieri is an Assistant Research Scientist at the National Consortium for the Study of Terrorism and Responses to Terrorism (START) and the Director of Undergraduate Studies for the University of Maryland’s Terrorism Studies Department. He holds a Ph.D. in Political Science from the University of Missouri.

Collin J. Meisel is a researcher with the Frederick S. Pardee Center for International Futures’ Diplo- metrics project, which gathers and analyzes data on issues related to international security. He holds an M.P.P from Georgetown University.

References

Ackerman, G. A., Asal, V., Johns, M., Binder, M. K., Murdie, A., Bale, J. M., & Rethemeyer, R. K. (2014). Anatomizing chemical and biological non-state adversaries identifying the adversary, final report. Retrieved from https://calhoun.nps.edu/handle/10945/46068

Ackerman, G. A., & Binder, M. K. (2015, April). Deriving new understandings of CB terrorism potenti- alities from individual perpetrator characteristics using the chemical and biological non-state adversaries database (CABNSAD). Presented at the 2015 International Studies Association Annual Convention, New Orleans, LA.

Ackerman, G. A., & Pinson, L. (2014). An army of one: Assessing CBRN pursuit and use by lone wolves and autonomous cells. Terrorism and Political Violence, 26(1), 226–245.

Asal, V. H., Ackerman, G. A., & Rethemeyer, R. K. (2012). Connections can be toxic: Terrorist organiz- ational factors and pursuit of CBRN weapons. Studies in Conflict & Terrorism, 35(3), 229–254.

Becker, M. H. (2019). When extremists become violent: Examining the association between social control, social learning, and engagement in violent extremism. Studies in Conflict & Terrorism. Retrieved from https://www.tandfonline.com/doi/full/10.1080/1057610X.2019.1626093

Berger, J. M. (2012, April 18). Patriot games. Foreign Policy. Retrieved from https://foreignpolicy.com/ 2012/04/18/patriot games/

Binder, M. K., & Ackerman, G. A. (2019). Pick your POICN: Introducing the profiles of incidents invol- ving CBRN and non-state actors (POICN) database. Studies in Conflict & Terrorism. Retrieved from https://www.tandfonline.com/doi/full/10.1080/1057610X.2019.1577541

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 39

Blee, K. M. (2005). Women and organized racial terrorism in the United States. Studies in Conflict & Terrorism, 28(5), 421–433.

Blettner, M., Sauerbrei, W., Schlehofer, B., Scheuchenpflug, T., & Freidenreich, C. (1999). Traditional reviews, meta-analyses and pooled analyses in epidemiology. International Journal of Epidemiology, 28(1), 1–9.

Borum, R. (2011a). Radicalization into violent extremism I: A review of social science theories. Journal of Strategic Security, 4(4), 7–36.

Borum, R. (2011b). Radicalization into violent extremism II: A review of conceptual models and empirical research. Journal of Strategic Security, 4(4), 37–62.

Bower, P., Byford, S., Barber, J., Beecham, J., Simpson, S., Friedli, K.,… Harvey, I. (2003). Meta-analysis of data on cost from trials of counselling in primary care: Using individual patient data to over- come sample size limitations in economic analysis. British Medical Journal, 326(2003), 1–6.

Burke, J. (2011, November 8). Who is Alaska militia member Mary Ann Morgan? Anchorage Daily News. Retrieved from https://www.adn.com/alaska-news/article/who-alaska-militia-member- mary-ann-morgan/2011/11/09/

Campbell, B., & Murdie, A. (2018). Keep the informants talking: The pursuit and use of CBRN weapons by terrorist organizations. Studies in Conflict & Terrorism. Retrieved from https://www.tandfonline. com/doi/full/10.1080/1057610X.2018.1531517

Carson, J. V., LaFree, G., & Dugan, L. (2012). Terrorist and non-terrorist criminal attacks by radical environmental and animal rights groups in the United States, 1970–2017. Terrorism and Political Violence, 24(2), 295–319.

CBS News. (2011, March 9). FBI arrests suspect in Wash. MLK day bomb case. Retrieved from https:// www.cbsnews.com/news/fbi-arrests-suspect-in-wash-mlk-day-bomb-case/

Collaborative Group on Hormonal Factors in Breast Cancer. (1996). Breast cancer and hormonal con- traceptives: Collaborative reanalysis of individual data on 53,297 women with breast cancers and 100,239 women without breast cancer from 54 epidemiological studies. The Lancet, 347(9017), 1713–1727.

Collier, D. (2005, May). A statistical rationale for qualitative research. Presented at the NSF workshop on interdisciplinary standards for systemic qualitative research, Arlington, VA.

Coveney, J. (2008, July 11). Firthlogit, Version 1.0. STATA Help. Retrieved from http://fmwww.bc.edu/ RePEc/bocode/f/firthlogit.html

Crick, N. R., & Grotpeter, J. K. (1995). Relational aggression, gender, and social-psychological adjust- ment. Child Development, 66(3), 710–722.

Cunningham, D. (2003). Understanding state responses to left- versus right-wing threats: The FBI’s repression of the new left and the Ku Klux Klan. Social Science History, 27(3), 326–370.

Davis, S. (2013, October 13). Women use “indirect aggression” to succeed, study says. CBS News. Retrieved from https://www.cbsnews.com/news/women-use-indirect-aggression-to-succeed- study-says/

Ellis, P. D. (2014). Lone Wolf terrorism and weapons of mass destruction: An examination of capabili- ties and countermeasures. Terrorism and Political Violence, 26(1), 211–225.

Fernandez, M. (2017, August 14). Bombing plot in Oklahoma City is thwarted with arrest, F.B.I. says. New York Times. Retrieved from https://www.nytimes.com/2017/08/14/us/oklahoma-van-bomb- domestic-terrorism.html

Fesbach, N. D. (1969). Sex differences in children’s modes of aggressive responses toward outsiders. Merrill Palmer Quarterly of Behavior and Development, 15(3), 249–258.

Firth, D. (1993). Bias reduction of maximum likelihood estimates. Biometrika, 80, 27–38. Gill, P., Corner, E., Conway, M., Thornton, A., Bloom, M., & Horgan, J. (2017). Terrorist use of the inter-

net by the numbers: Quantifying behaviors, patterns, and processes. Criminology & Public Policy, 16 (1), 99–117.

Gunaratna, R. (2002). Inside Al Qaeda: Global network of terror. New York: Columbia University Press. Hanna, J., Cordery, D. V., Steel, D. G., Davis, W., & Harrold, T. C. (2017). The impact of the mode of

survey administration on estimates of daily smoking for mobile phone only users. BMC Medical Research Methodology, 17(65), 1–9.

40 T. R. GUARRIERI AND C. J. MEISEL

Hegghammer, T. (2013). Should I stay or should I go? Explaining variation in Western Jihadists’ choice between domestic and foreign fighting. American Political Science Review, 107(1), 1–15.

Hoffman, B. (2006). Inside terrorism. New York, NY: Columbia University Press. Hopper, T. (2011, November 7). Alaska militia member tried to enter Yukon with pistol and ‘hoard’ of

bomb making instructions. National Post. Retrieved from http://nationalpost.com/news/canada/ alaskan-militia-member-tried-to-enter-yukon-with-bomb-instructions

Hosseinpoor, A. R., & Abouzahr, C. (2010). Graphical presentation of relative measures of association. The Lancet, 375(9722), 1254.

Hurwitz, E. (1982). Terrorists and chemical/biological weapons. Naval War College Review, 35(3), 36–40.

Jarman, N. (2004). From war to peace? Changing patterns of violence in Northern Ireland, 1990–2003. Terrorism and Political Violence, 16(3), 420–438.

Jensen, M. A., Seate, A. A., & James, P. A. (2018). Radicalization to violence: A pathway approach to studying extremism. Terrorism and Political Violence. Retrieved from https://www.tandfonline. com/doi/full/10.1080/09546553.2018.1442330

King, G., Keohane, R. O., & Verba, S. (1994). Designing social inquiry: Scientific inference in qualitative research. Princeton, NJ: Princeton University Press.

King, G., & Zeng, L. (2001a). Logistic regression in rare events data. Political Analysis, 9(2), 137–163. King, G., & Zeng, L. (2001b). Explaining rare events in international relations. International

Organization, 55(3), 693–715. Klausen, J., Morrill, T., & Libretti, R. (2016). The terrorist-age crime curve: An analysis of American

Islamist terrorist offenders and age-specific propensity for participation in violent and nonviolent incidents. Social Science Quarterly, 97(1), 19–32.

LaFree, G., Jensen, M. A., James, P. A., & Safer-Lichtenstein, A. (2018). Correlates of violent political extremism in the United States. Criminology; An interdisciplinary Journal, 56(2), 233–268.

Laqueur, W. (2000). The new terrorism: Fanaticism and the arms of mass destruction. New York: Oxford University Press.

Lee, A. (2011). Who becomes a terrorist? Poverty, education, and the origins of political violence. World Politics, 63(2), 203–245.

Leitgöb, H. (2013). The problem with modeling rare events in ML-based logistic regression: Assessing potential remedies via MC simulations. Presentation at the 2013 European Survey Research Association Meeting, Ljubljana, Slovenia.

Lindekilde, L., O’Connor, F., & Schuurman, B. (2019). Radicalization patterns and modes of attack plan- ning and preparation among lone-actor terrorists: An exploratory analysis. Behavioral Sciences of Terrorism and Political Aggression, 11(2), 113–133.

Maskaliūnaité, A. (2015). Exploring the theories of radicalization. International Studies. Interdisciplinary Political and Cultural Journal, 17(1), 9–26.

Maurer, S. M. (2009). WMD terrorism: Science and policy choices. Cambridge, MA: MIT Press. Meulenbelt, S. E., & Nieuwenhuizen, M. S. (2015). Non-state actors’ pursuit of CBRN weapons: From

motivation to potential humanitarian consequences. International Review of the Red Cross, 97 (899), 831–858.

Miller, E. (2017). Ideological motivations of terrorism in the United States, 1970–2016. START. Retrieved from https://www.start.umd.edu/pubs/START_IdeologicalMotivationsOfTerrorismInUS_ Nov2017.pdf

Miller, J. E. (2005). The Chicago guide to writing about multivariate analysis. Chicago, IL: University of Chicago Press.

Morlin, B. (2011a, March 10). The Spokane bomb attempt: Who is Kevin William Harpham? Southern Poverty Law Center. Retrieved from https://www.splcenter.org/hatewatch/2011/03/10/spokane- bomb-attempt-who-kevin-william-harpham

Morlin, B. (2011b, November 7). Feds hope to deny bail to Alaska militia suspect. Southern Poverty Law Center. Retrieved from https://www.splcenter.org/hatewatch/2011/11/07/feds-hope-deny- bail-alaska-militiasuspect

BEHAVIORAL SCIENCES OF TERRORISM AND POLITICAL AGGRESSION 41

National Consortium for the Study of Terrorism and Responses to Terrorism (START). (2017). Chemical and Biological Non-state Adversaries Database, Version 2.71 [Data file]. Retrieved from Markus Binder (mkbinder@umd.edu)

National Consortium for the Study of Terrorism and Responses to Terrorism (START). (2018). Profiles of individual radicalization in the United States [Data file]. Retrieved from http://www.start.umd. edu/pirus

Nemeth, S. (2014). The effect of competition on terrorist group operations. Journal of Conflict Resolution, 58(2), 336–362.

Pels, T., & de Ruyter, D. J. (2012). The influence of education and socialization on radicalization: An exploration of theoretical presumptions and empirical research. Child & Youth Care Forum, 41 (3), 311–325.

Schils, N., & Pauwels, L. (2014). Explaining violent extremism for subgroups by gender and immigrant background, using SAT as a framework. Journal of Strategic Security, 7(3), 27–47.

Shellman, S. M., Levey, B. P., & Young, J. K. (2013). Shifting sands: Explaining and predicting phase shifts by dissident organizations. Journal of Peace Research, 50(3), 319–336.

Simi, P., & Windisch, S. (2018). Why radicalization fails; barriers to mass casualty violence. Terrorism and Political Violence. Retrieved from https://www.tandfonline.com/doi/full/10.1080/09546553. 2017.1409212

Simon, J. D. (1989). Terrorists and the potential use of biological weapons: A discussion of possibilities. Santa Monica, CA: RAND Corporation.

Sjoberg, L. (2011). Conclusion. In L. Sjoberg & C. E. Gentry (Eds.), Women, gender, and terrorism. (pp. 227–240) Athens, GA: University of Georgia Press.

Talbot, R. (2000). Myths in the representation of women terrorists. Éire-Ireland, 35(3-4), 165–186. Tucker, J. B. (2000). Introduction. In J. B. Tucker (Ed.), Toxic terror: Assessing terrorist use of chemical and

biological weapons (pp. 1–14). Cambridge, MA: BCSIA. Vaillancourt, T. (2013). Do human females use indirect aggression as an intrasex competition strat-

egy? Philosophical Transactions of the Royal Society B: Biological Sciences, 368(1631), 1–7. Wendt, M. (2007). Considerations before pooling data from two different cycles of the general social

survey. Statistics Canada: Social and Aboriginal Statistics Division. Retrieved from http://www23. statcan.gc.ca/imdbbmdi/document/8011_D1_T9_V1-eng.pdf

Yeo, W. (2015, September 2). The growing threat of CBRN weapons. RSIS Commentary, 188. Retrieved from https://www.rsis.edu.sg/wp-content/uploads/2015/09/CO15188.pdf

Zwolski, K. (2011). Unrecognized and unwelcome? The role of the EU in preventing the proliferation of CBRN weapons, materials and knowledge. Perspectives on European Politics and Society, 12(4), 477–492.

42 T. R. GUARRIERI AND C. J. MEISEL

Copyright of Behavioral Sciences of Terrorism & Political Aggression is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.