AN EXPLORATORY TEST OF FAR-RIGHT EXTREMISTS PROTECTIVE AND PUTATIVE (RISK) FACTORS; IN BOTH VIOLENT AND NON-VIOLENT OFFENDERS USING THE EXTREMIST CRIME DATABASE By Lance Jhordan Asquith Lindsay A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Criminal Justice-Master of Science 2021 ABSTRACT AN EXPLORATORY TEST OF FAR-RIGHT EXTREMISTS PROTECTIVE AND PUTATIVE (RISK) FACTORS; IN BOTH VIOLENT AND NON-VIOLENT OFFENDERS USING THE EXTREMIST CRIME DATABASE By Lance Jhordan Asquith Lindsay Research in the field of extremism studies have steadily progressed as it emerged from the main field of terrorism studies. This progression has gone from defining terrorism and its root causes; to measuring cognitive and behavioral indicators which may predict the likelihood of radicalization toward extremism. This comes after tedious debates about typology and causes of terrorism in the field, there are still challenges to overcome as most research relies on open- source data. Through efforts to make this field more established the Extremist Crime Database (ECDB) stands out as a standard for this pursuit having some of the best open-source data and widespread usage. The database captures extremist violence from 1990 to 2014 and is a relational database that links event level data to individual offender level data. This database created by Freilich and colleagues (2014) has resulted in the ability to conduct quantitative analysis beyond descriptive statistics. The main question: are there differences in these risk and protective factors that lead to non-violent or violent outcomes of offenders in this sample? These nuances in the frequency of factors may explain why an individual followed the offending trajectory and if these differences explain offending. This thesis is dedicated to my grandma as without her I would not have this love for and interest in this pursuit of knowledge; and to my grandpa who provided a grand example of daily perseverance and duty. My motivation stems from my mentors, friends, family, and of course my two kittens Shuri and Prince Hall. iii ACKNOWLEDGMENTS I would like to thank Dr. Chermak for chairing my thesis committee and for his encouragement and support provided throughout this process without which this would not be possible. I also want to thank Dr. Carter and Dr. Scrivens for being on my committee and providing me invaluable feedback and support as well as opportunities for academic growth. I want to thank my dear friends and my colleagues for their support and motivation throughout this process. Finally, I would like to acknowledge my cats Shuri and Prince Hall who were my study buddies throughout this process and the Covid-19 pandemic. iv TABLE OF CONTENTS LIST OF TABLES ....................................................................................................................... vi INTRODUCTION......................................................................................................................... 1 Importance of this study .......................................................................................................... 2 LITERATURE REVIEW ............................................................................................................ 4 Extremist Crime Database (ECDB) ........................................................................................ 4 Radicalization To Extremism Involvement ............................................................................. 5 Group Typology in Study ........................................................................................................ 5 Pre-2000 .................................................................................................................................. 7 The Impact of the Internet ....................................................................................................... 8 Post-2000 ................................................................................................................................. 8 Putative Risk and Protective Factors ....................................................................................... 9 Extremist Studies Regarding Violent Subjects ..................................................................... 10 Extremist Studies Regarding Non-Violent Subjects ............................................................. 12 METHODS .................................................................................................................................. 15 Dependent Variable ............................................................................................................... 18 Independent Variables ........................................................................................................... 18 Demographic Variables ......................................................................................................... 18 Network Variables ................................................................................................................. 21 Financial Variables ................................................................................................................ 25 Criminal Variables ................................................................................................................. 26 RESULTS .................................................................................................................................... 29 Chi-Square Tests of Demographic Variables ........................................................................ 29 Chi-Square Tests of Network Variables ................................................................................ 31 Chi-Square Tests of Financial Variables ............................................................................... 34 Chi-Square Tests of Criminal Variables ............................................................................... 34 Multivariate Analysis ............................................................................................................ 35 DISCUSSION .............................................................................................................................. 41 BIBLIOGRAPHY ....................................................................................................................... 45 v LIST OF TABLES Table 1 Descriptive Statistics ……………………………………………………………….…28 Table 2 Group Regression Statistics …………………………………………………….……40 vi INTRODUCTION Extremist studies and domestic far right terrorism are gaining exceptional attention given the recent Capitol building breach on January 6th, 2021. The rhetoric of the former President Donald J. Trump is claimed to incite the breaching and interruption of the certification process of President Biden’s electoral college votes (Fandos & Cochrane, 2021). The activation of 20,000 National Guard members is a testament of the threat Far Right-Extremist (FRE) pose and that scholars have long warned about (Chermak et al, 2013; Klein, 2017). It is apparent that some FRE movements are past their boiling points and are ready to act on their grievances with force. For example, there were previous indications of discontent in Lansing Michigan, where armed militia members entered the Capitol on May 14th 2020 (Censky, 2020). In October 2020, the FBI announced that six anti-government suspects were involved in plotting to kidnap state official Governor Whitmer in response to COVID-19 and gun restrictions as their source of grievance (Flesher, 2020). These restrictions imposed were perceived to threaten the livelihood of FRE members, and the storming of the U.S. Capitol perhaps reflects an increase in FRE acting on their extreme ideology. These events may indicate a shift in attempting to force government officials to bend to their desire but departs from some previous FRE behavior that did not blatantly attempt government disruption (Chermak et al., 2013). These events have many individuals wondering how the growing intensity of this movement was not recognized more by the media or acknowledged by the government in the framing of domestic terrorists/extremists’ threats, and how federal law enforcement seemed unprepared for the January 6th attack (see Chermak, 2002; Schildkraut et al, 2020). Overall, scholars point out that domestic threats have been minimized compared to international threats which are over emphasized (Freilich et al., 2014; Powell 2018; Schildkraut et al, 2020). To address these issues, we need a deeper understanding of who 1 commits terrorism, and how an individual is influenced to embrace a FRE ideology. Scholars Simi & Windisch (2020) define the process of radicalizing as “Action pathways” which allows researchers and practitioners to understand that affiliation with movements could influence the nonviolent individual to become violent. In this study, these pathways or factors are considered to have influenced crime outcomes as the offenders have already been indicted and convicted (Freilich et al., 2009; Freilich et al., 2018). Importance of This Study This study attempted to contribute to the growing literature on extremism by exploring how violent offenders are different from non-violent offenders in this sample. The intent was to further refine extremist typologies using psychometric indicators while also providing new knowledge for practitioners of countering and preventing extremist violence. This study is important as there are few empirical comparisons of far right non-violent and violent individuals (Chermak et al, 2013; Clemmow, 2020). Studies that have investigated these indicators are increasing, yet a comparison of risk and protective factors modeled intentionally iteratively has not been conducted using the ECDB to date as Clemmow used a different database in the study (Clemmow, 2020). Chermak and colleagues (2010) point out that only focusing on violent crimes may lead to a lack of understanding of the escalation from non-violent to violent offending. Therefore, this study takes this important step of comparing non-violent FRE to violent FRE as surveys have shown that there is a FRE presence in most states (Freilich et al., 2009; Simi & Windisch, 2020). In addition, the survey study confirms that state and local law enforcement agencies consider FRE as a major domestic threat to the United States of America (Freilich et al., 2009). Findings in this study may lead to better understanding of violent and nonviolent offenders and one day prevention efforts and deradicalization programs might address these 2 factors. The literature review will cover material explaining why the variables in the study were selected. The methods section follows and then results for significant relationships are displayed. The discussion section will elaborate on the findings and real-world implications and general lessons learned. The final section concludes with limitations and directions for future research. 3 LITERATURE REVIEW Extremist Crime Database (ECDB) This thesis utilized the ECDB to study violent and nonviolent extremist in the United States. The ECDB includes all types of extremist groups in the United States, but the focus here is on Far-Right Extremists (FRE) as they have posed a consistent threat historically in the United States (Klein et al., 2017). These extremists are known to have varying concerns but there is consensus that these ideological concerns tend to overlap (Freilich, et al, 2014). The main groupings of these ideological concerns in this study are antigovernment and ethnic or racial in nature and help to distinguish patterns of behavior regarding extremism (Auger, 2020; LaFree, 2013; Doering et al., 2020). Given some of the issues with defining and conceptualizing terrorism and extremism, this study defined extremists using the ECDB as it is more inclusive of ideological motivated behavior as acts included have to satisfy a two-prong test: first, there is a behavioral element where a state or federal law had to be broken; and second, there is an attitudinal element where the intent of the act had to be ideologically motivated. Individuals executing a violent or non- violent act and were motivated by far right extremist ideology were the focus of this study. The ECDB defines domestic far right extremists as the following: They are fiercely nationalistic, anti-global, suspicious of federal authority, and reverent of individual liberties, especially their right to own guns and be free of taxes. They believe in conspiracy theories involving imminent threats to national sovereignty or personal liberty and beliefs that their personal or national “way of life” is under attack. Sometimes such beliefs are vague, but for some the threat originates from specific racial or religious groups. They believe that they must be prepared to defend against this attack by participating in paramilitary training or survivalism. (Freilich et al, 2014). This inclusion criteria allows for capturing violent and non-violent crimes in support of an extremist ideology. Other criteria for inclusion include that the criminal activity must have been committed in the United States including Puerto Rico. One value of using ECDB data, 4 which relies on open-source documents, is that the available information used to capture and conceptualize offending is more complete than official statistics provided on extremist crime historically (i.e. FBI yearly crime report) (Vergani et al., 2020). Most studies exclude many far- right events because they have traditionally counted as hate or bias crime; the ECDB inclusion criteria accounts for this (Hayes et al., 2018). Radicalization To Extremism Involvement A review of the literature states radicalization to extremism is a steady progression in which multiple life events can lead to ideological adoption and sometimes engagement in violence (Freilich et al., 2018; Simi & Windisch, 2020). This progression can be “instrumental” to one’s survival or ideological and related to personal or collective relative deprivation (Asal et al., 2020). If the offending is instrumental- it is meant to further such goals as self-sustenance, maintaining White supremacy or inciting a race war to counter the perceived “White Genocide” and/or “The Great Replacement” that is purportedly occurring (Simi & Windisch, 2020; Vergani et al., 2020; Gruenewald et al., 2016; Freilich et al., 2019). Gruenewald and colleagues, have pointed out that scholars and practitioners need to study characteristics at the granular level to identify at risk individuals, and slow or stop this radicalization process (2019). Group Typology In Study Anti-Government adherents heavily resent centralized government and it’s purported all expansive overreach, as it was explained above there is considerable overlap in this and racial/ethnic supremacy ideologies (Freilich et al., 2014). Yet there are differences in the main goals of the respective movements or ideologies as individuals can adhere to the goal and not be formally attached – but typically agree that less government is better (Chermak, et al. 2013; Grunewald et al., 2013). Groups or individuals share similarities to Sovereign Citizens; in which these individuals claim to be freemen and see the involvement of the federal government as 5 infringing upon their rights which they then tend to disregard with fervor (Doering et al, 2020). Concerns involve having to pay taxes, vehemently defending gun rights, protesting any increase in tax or restriction for instance on (semi or full) automatic weapons especially long gun weapons: all are a threat to ones’ livelihood (Chermak et al., 2010). At the heart of concerns for anti-government is that the government should be reduced to minimal operations in order to increase individual freedom regardless the utility or intent of the government. The increase of government programs and spending also is a common irritant of these individuals as they consider taxes government overreach (ECDB, 2013:117). Another conspiratorial belief is that official documentation such as social security numbers and drivers’ licenses, and even public health vaccinations, are methods solely for imposing government control (Freilich et al., 2018). Responses to these issues range from tax evasion, to filing court liens on officials to outright attacking actors of the state (Chermak et al., 2010; Doering et al., 2020). Militia and Patriot groups also fit into this category as they are usually avid gun right supporters and they also see the need to prepare for disasters through regular secluded training (Freilich et al., 2009; Freilich et al., 2018). Some militia members may exhibit racial or ethnic frustration, but anti-government involvement sentiment seems to be the primary concern to coalesce upon (Doering et al., 2020). An example of another grievance is anti-abortion attacks on abortion providers; due to beliefs around childbearing which bleeds into the second typology (Hayes et al., 2018). An important distinction made is that the mainstream conservative movement is not grouped here, but individuals do share the same basis of adherence to religious values (Gruenewald et al., 2013). In this perspective modernization is rejected as it promotes concepts contrary to religious beliefs, also termed “non-wave” or “counter-wave” terrorism (Auger, 2020). 6 Ethnic and/or Racial nationalist want a pure racial world of White individuals only; with all others considered a threat to their livelihood (Simi & Windisch, 2020). Ethnicity is useful in this sense because FRE concerned in this category may also hate other Caucasians such as Jewish descendants (Freilich et al., 2019). Examples of ideologies are the neo-Nazi, or anti- globalists who see the Jewish population and immigrants at fault for problems with globalism and their individual plight (Fitzpatrick et al., 2017; Holt et al., 2020). This attachment to national or global issues was found to lead to more violence in groups, and therefore may be pertinent to study in this sample (Chermak et al., 2013). This stance generally represents the “power threat hypothesis, which suggests that the size of a racial or ethnic minority group population is inversely related to discriminatory acts by Whites against minorities” (Adamcyzk et al, 2014). At the heart of this hypothesis is that national sovereignty and personal freedoms are limited when individuals migrate here and partake in this society therefore creating grievances that further fuel “irritation with immigrant groups including non-Christians” (Adamcyzk et al, 2014). Common examples of some issues were the building of the border wall or Middle East travel ban restricting the flow of legal and illegal migration of immigrants. Pre-2000 There may be a surge in far-right sentiment and criminal activities flowing from these sentiments, but far right extremism has a very long history and has cycled in waves (see Auger, 2020). For example, Louis Beam was the North American proposer of “leaderless resistance” which was promoted in the early 1980s (Gruenewald et al., 2013). This concept was taken from Che Guevara and originally embraced by far-left groups and eventually replicated by the FRE; even the use of prison tactics was embraced to avoid tactical responses or capture (Chermak et al., 2013; Smith et al., 2015). This method was meant to make it harder for law enforcement to capture FRE group members, or infiltrate such groups (Gruenewald et al., 2013). The 7 effectiveness of leaderless resistance was not realized until 1995 when a FRE lone wolf Timothy McVeigh bombed the Alfred P. Murrah building (Silva et al., 2020). This was deemed by some as a “new wave” of terrorism in which it was apparent loner or lone wolf packs were becoming a phenomenon to address (Auger, 2020; Cohen et al., 2014; Gruenewald et al., 2013). The Impact Of The Internet The Internet has been exceptionally useful for globalism and the exchange of knowledge; but this platform has been co-opted to further extremist ideologies and conspiracies (Holt et al, 2020; Scrivens et al, 2020). The use of the Internet allows individuals to learn new skills in the real world while also serving as a platform in which “cyber tool[s] and weapons of mass destruction” may likely become a major threat (Auger, 2020). Studies have found that the Internet and various platforms allow like-minded individuals to gather (i.e., echo chambers) and share material, recruit, or simply inspire people to join the cause (Holt et al., 2020). A recent example of the utility of the Internet to extremists is the use of live streaming for the Capitol insurrection mentioned above on Jan. 6th, 2021. Literature points out that FRE have long used the Internet, beginning with bulletin boards and forums moving towards the example just given (Freilich et al., 2014). Therefore, scholars and practitioners will need to keep in mind methods to analyze such cyberspaces that harbor or support extremist sentiment and the cycle or method of changing platforms (Holt et al., 2020). The use of the Internet poses a unique challenge in which we must carefully define what is free speech, hate speech, or indicators that an individual is “leaking” information about an imminent attack (Freilich et al., 2020a). Post-2000 The creation of the ECDB as a “specialized dataset” has furthered this field from anecdotal speculation to quantitative empirical analysis (LaFree, 2013) and represents what is a product of the extraordinary investment in research on terrorism following the September 11 th 8 attacks. The passing of the U.S. PATRIOT ACT allowed the relevant authorities to expand their tool kits to prevent terrorism and extremism (Freilich et al., 2020). These new tools paired with new methodology such as Intelligence-Led Policing has allowed for better techniques to infiltrate and apprehend individuals – with the Internet, anonymous tips, and informants being heavily utilized (Freilich et al., 2020b; Gruenewald et al., 2019). Along with the PATRIOT ACT the Intelligence Reform and Terrorism Prevention ACT (IRTPA) passed in 2004 which helped to provide funding for the “Information Sharing Environment[s]” well known as Fusion Centers (Gruenewald et al., 2019). Putative Risk And Protective Factors This study addressed the “need to focus more on the social features that bond individuals to communities, and broader psychological and criminological mechanisms that radicalize individuals” (Wolfowicz, 2019: 30). This study used variables that triangulated indicators of attitude, intent and behavior. This study aimed to identify putative risk and protective factors within the field of extremist studies regarding behavioral profiles and decision making. The importance is pointed out by (Wolfowicz et al., 2019:3) stating that there is a lack of understanding of who and why a person radicalizes, which is due to “the profiles of a small number of terrorists, without consideration for the distribution of such factors among nonviolent radicals, and the wider population”. This study continued this process of expanding knowledge by comparing violent and non-violent offenders’ risk and protective factors. Risk factors have been identified in the literature that is anchored in both Adverse Childhood Experiences and theory such as Developmental Life Course Theory (DLC) (Bhui et al., 2016; Windisch et al, 2020). Risk factors are criminological variables that have been found or are thought to increase the probability that an individual will commit crime. Regarding DLC the general expectation is that the more negative experiences one has the more likely they may 9 justify their use of crime. Popular examples of a risk factor are low income, low education attainment and having few strong social relationships. Protective factors are variables which reduce the likelihood an individual may resort to crime, examples include high income, legitimate employment, high education attainment and strong social bonds including family, friends, and coworkers. Scholars have pointed out the importance of exploring the Putative Risk and Protective Factors that indicates possible radicalization to extremist violence; including those of which cover traditional criminogenic factors. Gill also points out the importance of studying the process by observing lingual and spatial patterns in future research (2020:56; Becker, 2019:5). By reviewing case files which include court documents and news reports, these variables could be used to identify details of interest in order to deem the instance as protective or risky in nature. The section reviewing study measures will give the specific risk and protective factors as sorted for this study. Extremist Studies Regarding Violent Subjects A startling statistic from a START report presented data indicating that “white supremacists were responsible for 71.4% of ‘ideologically motivated homicides committed by Far Right Extremists’ between 2010-2018, with that percentage increasing to 84% for the 2015- 2018 timeframe” (Auger, 2020). Another study which reviewed the analysis “by the Institute for Economics and Peace...[found] there has been a surge in far-right terror incidents since 2010, with a 320% increase between 2014 and 2018” (Auger, 2020) This trend indicates a significant threat in the United States and is empirically confirmed by the report and is receiving more attention by the government and citizens. Grunewald and colleagues compared loner attacks to domestic extremist violence and found that most of the violent events were committed by FRE’s (2013). This same study and others found that loners were more likely to “Leak” their plans to gain support, while also gathering knowledge online in preparation of events (Cohen et al., 10 2014). Another study by Freilich and colleagues found that mass violence offenders sometimes attack individuals known to them with locations being: “30% workplace, 25% bars and malls, [and] more than 20% in schools” while other locations are random and tend to include strangers (Freilich et al., 2020c; Hayes et al., 2018). Schildkraut and colleagues found that domestic extremists were typically “white, were portrayed as ‘troubled, mentally ill loners’ who were unstable” (2020). Some studies compared online rhetoric to offline behavior and found that groups openly promoting violence were more likely to use violence (Scrivens et al., 2020). Another study pointed out that these individuals had trouble with employment and prior criminal records or interactions with the criminal justice system (Hayes et al., 2018). Organizations that had broad social connections with other extremists and were previously involved in violence were also likely to continue using violence (Vergani et al., 2020). Windisch and colleagues, found that the extremist process is generated through multiple “adverse experiences” some of which is feeling a lack of support from family or having serious household issues (2020). These adverse events may be common and therefore overlap with individuals who are non-violent. One example pointed out in the literature was that an individual only began to “resort to violence in his future relations with people because he no longer trusted their intentions” (Windisch et al., 2020). Doering and colleagues, also found that studying life events and “personal characteristics” may be better to identifying those at risk – rather than trying to identify the specific ideology first (2020). Essentially these characteristics have been found as significant in studies with similar groups; this study reviewed individuals and coded relevant variables related to life experiences which represent risk and protective factors (Adamczyk et al., 2014; Gruenewald et al., 2019). As individuals comprise both groups and loner attacks, and 11 loners have been found to display less preparatory behavior; these risk factors can help identify individuals at risk of radicalizing towards extremist behavior (Klein et al., 2017). Ultimately the overall intent is to have tailored counter-radicalization deradicalization programs that are sensitive to these significant variables’ nuances. Extremist Studies Regarding Non-Violent Subjects The ECDB tracks non-violent financial crime and found around 700 incidents committed, we can only imagine the amount of crime that has occurred since 2014 (Freilich et al., 2014). Financial crimes include bank fraud, counterfeiting, and online scams as a platform to commit non-violent crimes (Klein et al., 2017). Non-violent crime can also include “information misrepresentation, theft, vandalism, cyberattacks” and more such as providing money to support the cause or for recruiting costs and materials (Gruenewald et al., 2019). The literature on non-violent extremists is lacking and Chermak and colleagues, point out that -there were no studies that compared far right violent and nonviolent groups (2013). While studies may be in process, this author does not know of any comparing non-violent to violent extremists using the Extremist Crime Data Base. Chermak and colleagues have already discovered that “terrorist commit a large number of precursor activities to fund operations and specific attacks” (2010). In this sense there can be members who are not yet emboldened to commit violent crime, yet they participate in non-violent instrumental crime to support the cause. Anti-government related crime such as “tax refusal, gun charges, land use violations, motor vehicle violations and filing false liens” fit within this category (Chermak et al., 2010). It is crucial to see if any of these risk variables significantly differ which may help to explain the outcome of the crime; and which risk factors push or pull one from nonviolent to violent crime if any (Chermak et al., 2010). These extremists might become emboldened when confronted by government actors addressing their crimes – that may be the moment that they switch trajectories 12 to violent crime (Chermak et al., 2010). This radical view towards law enforcement officers (LE) can surprise the officer during a simple traffic stop, and is a serious threat as anti-government Sovereign citizens committed 63% of LE fatalities (Freilich et al., 2018) As the Extremist Crime Database has produced various studies on extremism with more nuanced data; scholars have called for a study as such comparing nonviolent and violent extremists’ risk and protective factors. Studying anti-government and cultural/racial extremists adds to this subset of terrorist and extremist studies by statistical testing characteristics within a sample of domestic far-right extremists. Terrorism has moved from descriptive and anecdotal studies in the pre-2000s to quantitative studies with the introduction of the ECDB and other databases post 2000. The Internet has expanded the reach of information individuals have and the possibility to connect to others who are likeminded. This expansion of social networks and capability made researchers in the field once again pivot to capture another expansive data source for analysis to understand characteristics explaining crime outcome (Wojcieszak, 2010). In closing we have learned that far-right extremism threats are serious and increasing but there continues to be a lack of comparative studies to further understand relationships that influence crime outcome between this specific group of subjects. Freilich and colleagues found four findings shared here: groups using leaderless resistance inspired by Louis Beam and others were more likely to commit violence; established hate groups linked to other hate groups also were likely to commit violence; and aggressive recruiting indicated violent proclivity; finally groups that had methods of fund creation were less likely to commit violence (Freilich et al, 2015) This study keeps in mind that all results are important, for instance if there is not a significant difference in a variable there may be an overriding factor that could explain the 13 similarity (e.g., group grievance, religious involvement). As individuals comprise groups - indicators may be in line with theory and results in extant literature. 14 METHODS The author was provided a random sample of 50 violent and 50 nonviolent FRE from the ECDB. Some core information was included for each extremist and access was provided to the search files for appending to each case. The Extremist Crime Database was compiled beginning in 2006 and data collection started from 1990; with funding from the Department of Homeland Security and the National Consortium for the Study of Terrorism and Responses to Terrorism (START). This initiative created a relational database which allowed for group level and individual level studies of offenders, including loners, terrorist groups, and included all variations of domestic extremism whether successful or failed. This database allows for violent and financial schemes to be studied along with who was attacked, if there were social ties and which group such as far-rightists, Jihadists, animal rights and environmental rights extremists were involved (ECDB codebook, 2014: 3). This database also accounts for material support and weapons stockpiling which indicates planning for an incident. The definition of Far-Right Extremism in the ECDB is clear to explicitly state that “the mainstream conservative movement and the mainstream Christian right are not included” (ECDB codebook, 2014:3). The method of creating case files is extensively covered by scholars regarding the use of the ECDB and cleaning variables (Gruenewald, Chermak, Freilich; 2013). The author systematically searched these files and conducted targeted searches for additional information to clean the risk and protective indicators of interest. The next step included following the open- source data gathering method of the ECDB. Primarily that means original court transcripts and files are considered most trustworthy and academic journals, then newspapers create a hierarchy of trusted “open source” data. The author sorted the respective individual variables and their response options into putative risk and protective factors based upon the ECDB codebook definitions of the variables 15 included in Table 1. Once all cases were cleaned the data file was imported to SPSS version 26 in order to run descriptive statistics and then create dichotomous versions of these variables representing risk or protective values. When studying risk factors, the sample of interest was viewed through the lens of the Two Pyramid Model (Wolfowicz et al., 2019). Non-violent extremist in this case have completed the opinion radicalization pathway in which they may support extremists but through select activities all of which do not include violent crimes (e.g., weapons stockpiling or material support). Whereas violent extremist will have completed the action radicalization pyramid (violent pathway) which may include non-violent instrumental crime, but they progress to violent crime making their opinion become action. This study is exploratory, and thus focused on variables from multiple criminological theories including social learning, social control, and strain theories (Freilich et al., 2018;). Protective factors are of particular interest as they may aggregate within the non-violent extremist offending sample. Comparing risk factors of non-violent offenders to violent offenders may provide further evidence of protective factor strengths in non-violent offenders. Which may direct policy to focus on creating more facilitative environments to address these risks in communities for individuals. Furthermore, differences in protective factors may help to de- radicalize or disengage violent offenders and redirect at-risk youth in the future. These factors can represent strain or turning points towards violent or non-violent extremist offenses and represent the completed “trajectory” (Freilich et al., 2018; Windisch et al., 2020). These factors were coded and added to the existing Extremist Crime Database which has some putative and protective factors coded for certain cases. Based on the existing literature the primary factors of interest in this study are taken from Wolfowicz et al., (2019). The variables 16 are categorized in Table 1 below, and they were modeled to test for differences of statistical significance. The cases were coded with a total (N=100) composed of 50 violent extremists’ offenders, and 50 non-violent extremist offenders’ and their coded occurrences of variables in the study. Descriptive statistics will be presented in the next section results, Chi-Square test of independence were conducted and will be given in the second section of results. The last section includes multivariate logistic regressions used to test for statistical significance and strength of variables on outcomes using the pseudo R2 (Nagelkerke R2) given statistical power issues with a small sample size. The final model for each iteration (which is a particular grouping of variables); includes only the most frequent and significant predicators for each section. The method for iteratively testing pairs and groups of variables was listwise proceeding from top to bottom, then offsetting the position of each individual variable N-1 times of the possible combinations. There are iterations of 2, 4, 6, and 10; in this sample there are 29 unique individual variables and many of the tests include repeating some variables to complete each test. This iterative testing method using logistic regressions was selected due to scholars in the field recognizing that “directly measuring core concepts” can be challenging, creating various models and testing the goodness of fit through blocks of alternating variables helps to overcome the lack of statistical power to make pertinent observations (Chermak & Gruenewald, 2006; Wells & Chermak, 2011). This logistic regression used is less statistically robust than other regression analyses but is used in extremist studies in order to build quantitative methods further and continue improving methods of measurement and analysis. 17 Dependent Variable The dependent variable is offender outcome measured dichotomously as violent, or nonviolent outcome tracked by the ECDB and has resulted in criminal charges and indictment. The ECDB codebook considers nonviolent crime financial and property offenses. Independent Variables There are 4 categories that the variables were grouped under: demographic, network, financial and criminal history - these variables are presented in their groups in Table 1 and were selected due to extant extremist literature and availability of variables in the database. Although 29 variables were originally selected for this study a few were removed while testing for the best fit variables. Each variable removed will be discussed as they are not in preceding tables. Demographic Variables Family was coded as (0) protective if the individual had married parents, all other familial arrangements were coded as risk (1). Risks options included: 2) Parents divorced, 3) Parents separated, 4) Mother died, 5) Father died and 6) Both parents died. Bhui and colleagues (2016) found that “adverse life events may cause people to draw on pre-existing social networks as a means of emotional support, creating opportunities to resolve dissatisfaction and garner support for extremism” (483). Meaning that without positive reinforcement at home from parents - individuals may seek other relationships outside the house that may be a negative influence and in general reduce parental influence on behavior. The sample mean for family was .18 with 18 (18%) subjects reporting risk occurrences, while 82 had protective occurrences of familial arrangements which is surprising as literature would expect otherwise. This can be due to access of data and the general coding scheme which counted lack of specific mention as an answer – resulting in this case a positive coding. 18 Children represented here as risk (1) means the individual does not have a child. This was reverse coded as (0) for having a child as it is assumed through routine activity theory that having children and a spouse keeps one busy and not willing to commit crime. The sample mean was .81 with 81 (81%) subjects reporting not having children (risk) while 19 in this sample had a child or more (considered protective). In this sample majority of subjects are considered at risk for this variable which is a surprising similarity for both populations to display. Education here is collapsed to represent risk (1) as H.S. Diploma and less, and protective (0) represents some college and more experience up to the limit of post graduate work. Risk categories were: 1) Home schooled, 2) Less than 8th grade, 3) Completed 8th grade, 4) Some H.S., 5) GED, and 6) H.S. Diploma - this is based on extent literature pointing out lower education being a risk for committing crime. Protective options were: 7) Some college or vocation, 8) Vocational graduate or associates degree, 9) College graduates, and 10) Post Graduate work. Routine Activity theory would assume that the more time someone spends attaining education the less likely they will have time or motivation to commit crime. The sample mean was .54 with 54 (54%) subjects in the risk category and 46 subjects had protective occurrences. Suicide Attempt represents that the individual engaged in a suicide mission and or was willing or expecting to die at the end of event, risk coded (1) for yes, and protective coded as (0) no. When an individual indicates they are willing to lose their life for a cause that means they are more willing to commit, view or experience violence compared to an individual who is not willing to engage in these risky actions. Misiak and colleagues support this claim stating, “Suicidal ideation and family history of mental disorders were reported by 57% and 48% of 19 extremists[respectively]” (2019:54) The sample mean was .07 with 7 (7%) subjects reported to have risk occurrences while the vast majority, 93 subjects had protective occurrences. Mental Illness here is coded as risk (1) when the individual had a mental illness and is coded as (0) protective when not deemed to have mental illness. This variable is coded as “The odds of having mental illness were over 13 times higher in the group of lone actors compared to group terrorists” (Misiak et al., 2019:53). We could then infer mental illness may have some impact in this sample. The sample mean was .14 with 14 (14%) subjects in the risk category and the vast majority 86 subjects had protective occurrences. Sex is represented here as male coded as risk (1) based on literature pointed out above, and female (0) as protective. The sample mean was .84 (84%) with 84 males as the risk category and 16 female subjects as the protective category. This sample is in line with literature and there is an overwhelming majority of males as expected. Race here is coded as risk (1) for Caucasian as based in literature they are the reference group, protective coded (0) for all other races. Protective options in this variable included: 2) Black, non-Hispanic, 3) Hispanic, 4) Asian, 5) Arab, 6) Native American, 7) Biracial, and 8) Other. Risks in this sample is 1) White, non-Hispanic and based in the literature (Simi & Windisch, 2020). The sample mean was .94 (94%) with 94 Caucasian subjects and 6 subjects of other race reported for the protective option. Foreign Born asked if the individual was born outside the United States with risk coded as (1) for US Born, and coded (0) protective for foreign born. Given the focus of domestic far- right extremists the best option for this variable is to deem U.S. born as risk as it has been found in literature there have been more domestic attacks than international (Bhui et al., 2020). In this 20 sample the mean was .96 (96%) representing the reported 96 U.S. Born subjects in the risk category, with 4 foreign born individuals representing the protective response option. Age was measured as dichotomous and is reverse coded as risk (1) representing 16-35 years old, and 36-69 years old is coded as (0) for protective based on the literature. Misiak et al., also found age to be an indicator for “sympathies of radicalization” due to individuals being open to ideas and impressionable (2019:53). In this sample the mean was .49 (49%) or 49 subjects for the risk category of 16-35 years old with 51 subjects in the protective grouping including 36 to 69 years old. Network Variables Community Status (Comm Status) represented the subjects’ status or prestige, this factor was essentially dichotomous and was reverse coded to represent risk (1) as low prestige, and (0) high prestige as protective. High prestige could be informal and or formal, for example Perliger and colleagues (2016:221) talked about dedication to a cause can lead to “higher organizational status” and that these individuals are “more likely to be married and employed and less likely to have been religious since childhood. As low prestige represents a social context in which others may not look up to the individual, popularly known as a loner type and may not “enjoy greater prestige within the group and may be more likely to be drafted into action” (Perlinger et al, 2016:221). Simply put, low prestige individuals may be perceived as disposable or expendable for the cause as the movement can still operate without the individual. In this sample the mean was .53 (53%) with 53 subjects in the risk category of low status/prestige and 47 subjects in the protective category of high status/prestige. Family Extremist [Involvement] (Fam Extrem) represented whether the individual had a family member involved in an extremist movement. Risk options included: 1) parents, 2) siblings, 3) spouse/partner, 4) children, 5) extended family members and 6) multiple members; 21 protective was coded as (1) for having family members involved in extremist movement. This variable was coded as (1) for having any family member or any of the risk options listed above present in the case file, and (0) for protective if they did not. In this sample the mean was .19 (19%) with 19 subjects reported in the risk category having family members involved in an extremist movement and the vast majority 81 subjects in the protective category had no familial connection to an extremist movement. Religious Involvement (Relig Involve) represented religious involvement in some manner, coded as risk (1) for some involvement, and (0) for no involvement. There was no need to recode this variable but there are two reasons for being able to describe religious involvement as risk. The first reason is that religious commitment and belief in religious superiority has justified killing others historically (Misiak et al., 2019). The second reason relates to anti- government arguments in which extremist religious values have found the government to be a threat to their way of life and the best example of this is the conflict of funding abortions and allowing women to have control of their body. Involvement may not be a risk itself, but in this sample the presence of the indicator pointed to an attitude and belief that may justify their crime. In this sample the mean was .10 (10%) with 10 subjects in the risk category having been religiously involved; the majority 90 subjects were in the protective category and were not religiously involved. Recruited had fourteen response categories and were recoded to include eleven of those options representing risk as (1). The remaining three options were coded as protective (0) and represented an individual acting on their own, contrary to the various methods they could be recruited. Risks options included the location of recruitment as follows: 1) Prison/Jail, 2) University/School, 3) Newsletter/Movement Propaganda, 4) Personal friend/neighbor/family 22 member, 5) Religious Institution gathering, 6) Military, 7) Internet, 9) Personal visit by member, 10) Charismatic leader, 11) Targeted system or recruited, and 13) Attended meetings. Any method of recruiting implies further social learning through those contacts and criminal skill would be expected to increase. Protective options included: 8) Self-started, 12) By chance, and 14) Individual decision. These protective variables could be considered risks in loners but fits perfectly regarding financial crimes when reviewing cases such as tax avoidance or tax schemes in which a person worked alone without support. In this sample the mean was .62 (62%) with 62 subjects in the risk category having been, while 38 were in the protective category not being recruited working alone. Military Awards (Mil Awards) was coded as binary to represent risk as yes (1) having current or prior military service and coded no (0) as no military background in the protective category. Having military experience would mean that the individual had weapons training and should be more proficient with a weapon compared to a civilian. In this sample military awards had a mean of .03 (3%) with 3 subjects in the risk category having had prior military experience and the vast majority 97 subjects were in the protective category having no military experience. Marital Status was coded as risk (1) and included: 3) Single, 4) Divorced, 5) Separated, and 6) Widowed- this is based upon the rationalization in literature that points out being partnered allows for structuring of time in a positive manner, these arrangements break from protective arrangements. Protective positive options included 1) Married Monogamous, 2) Married Polygamous, 7) Cohabitation and 8) Boyfriend/Girlfriend once again rooted in literature but results are conflicting as Perlinger and colleagues, (2016) found couples who were married and employed held higher organizational positions in extremist movements. In this sense, marital status may be protective against violent crime when compared to financial crime in which 23 partners work together in the scheme. In this sample the mean was .63 (63%) with 63 subjects reported in the risk category and 37 reported in the protective category. Personal Grievance (Personal Griev) is a variable that tracks whether the individual was motivated by a personal grievance to commit the crime charged for. Risk here is coded as (1) for yes, and protective as (0) for no personal grievance needed to commit crime. An example of a personal grievance from the data is one subject was experiencing financial problems after being laid off and he returned to his workplace and committed a violent shooting of former co- workers when trying to commit a scheme to get money from the company to pay rent. An example of a financial crime with personal grievance included one individual exaggerating their W-4 forms on taxes to avoid having taxes collected from income; this does not have a particular grievance towards an individual but in court the IRS is the complainant. Anti-government rhetoric indicates a conscious grievance when reviewing financial crimes. Another example with violent subjects included having a personal hate towards black individuals and so the individual attacked the victim based on their race and this would be considered a personal grievance rooted in ethnic superiority. In this sample the mean was .66 (66%) with 66 subjects reported in the risk category and 34 subjects were in the protective category. Political [Activity] Scale represented political activity which can represent intensity of devotion to a cause. It is coded here as (1) risk when involved in one or more activities and (0) protective when not involved in political activities. The variable was already compiled and included general activities such as attending political rallies, events, sharing political information and general political activities. In this sample the mean was .29 (29%) with 29 subjects in the risk category of being active in political activity and most subjects (71) reported not being involved in political activity. 24 Shared Info Web represented online posting behavior and once again is a way to measure dedication to a cause, coded as risk (1) for sharing, and protective (0) for no sharing activity. In this sample the mean was .22 (22%) with 22 subjects in the risk category having shared info on the web, and the vast remainder 78 subjects were in the protective category. Lone Wolf is coded as risk (1) here when acting with others, and protective (0) when acting alone. Risks options were 2) part of formal group, 3) part of informal group, and 4) acting with others no clear group boundaries. While 1) acted alone is considered protective when considering social learning and how individuals transfer skills and knowledge to commit crime – so acting alone may be less efficient when intending to commit crime. In this sample the mean was .66 (66%) with 66 subjects having acted with others and 34 subjects reported here in the protective category having acted alone. Financial Variables Community Ties (Comm Ties) represented the subjects’ living arrangement, with two options representing protective and four representing risks. The protective options in the codebook “1) lived with spouse/children and 2) lived with parents/other family”. Risk options in the codebook “3) lived alone, 4) lived with non-family, 5) in custody serving status, and 6) no stable residence”. This variable was recoded as living with a spouse and parents as protective (0) and risk (1) when living alone, living with non-family, in custody serving status and no stable residence. In this sample the mean was .53 (53%) with 53 subjects living alone or with nonfamily members in non -intimate settings and 47 subjects reported protective living arrangements. Income was coded to have low income and unemployed as risk (1) with middle and high income as protective coded (0). Risk options included 1) low income and 4) unemployed, and protective options included 2) Middle, and 3) High income. Extant literature would expect 25 unemployed and low income to commit violent acts whereas middle and high income are more likely to commit non-violent crimes (i.e., blue collar or white-collar crime). In this sample the mean was .58 (58%) with 58 subjects in the risk category of low or no income, and 42 subjects were in the protective category of medium and high income. Employ at Offense which had five original response options- when recoded two options represented risk coded as (1) for being unemployed or being employed by the extremist movement. The protective options were 2) student, 4) being retired, or 5) having legitimate employment – these options were coded as protective (0). Risks were 1) unemployed and 3) employed by extremist movement. In this sample the mean was .45 (45%) with 45 subjects reported in the risk category and 55 subjects reported in the protective category. Criminal Variables Prior arrest was coded as risk (1) for 1 or more prior arrests, and coded (0) protective for no prior arrests. This variable was a risk factor accounting for routine theory activity and social learning when operationalized to expect that prior criminal history makes a person more likely to have more contact with the criminal justice system once involved. In this sample the mean was .33 (33%) with 33 subjects reported in the risk category and the majority 67 reported in the protective category having no prior arrest. Total Arrests was coded as risk (1) for one or more total arrests, and protective coded (0) for no arrests. This variable was in line with prior arrest and expected that the more times an individual is involved in the system the more likely they will continue to be. This sample has a mean of .31 (31%) with 31 subjects who had 1 or more total arrests and majority of subjects, 69 were in the protective category and had no arrests. Ever Incarcerated was coded as (1) risk for one or more incarceration periods, and coded protective as (0) for no incarceration terms. This variable and the number of incarcerations 26 would indicate that an individual is more likely to have a network of social deviants and would become more criminally skilled. In this sample the mean was .21 (21%) with 21 subjects having had a term or multiple terms of incarceration while 79 subjects reported the protective category of having no incarceration terms. Number [of times] Incarcerated is coded as (1) for risk when incarcerated one or more times, and coded protective (0) for no terms of incarceration. In this sample the mean was .20 (20%) with 20 subjects in the risk category and 80 subjects reported the protective category option of no incarceration. 27 Table 1 Descriptive Statistics Variables N (Risk) N (Protective) Mean SD Demographic Family 18 82 .18 .386 Children 81 19 .81 .394 Education 54 46 .54 .500 Suicide Attempt 07 93 .07 .256 Mental Ill 14 86 .14 .348 Sex 84 16 .84 .386 Race 94 06 .94 .238 Foreign Born 96 04 .96 .196 Age 49 51 .49 .502 Network Community Status 53 47 .53 .501 Family Extremism 19 81 .19 .394 Religious Involvement 10 90 .10 .301 Recruited 62 38 .62 .487 Mil Awards 03 97 .03 .171 Marital Status 63 37 .63 .485 Personal Grievance 66 34 .66 .476 Political Scale 29 71 .29 .456 Shared Info Web 22 78 .22 .416 Lone Wolf 66 34 .66 .476 Financial Community Ties 53 47 .53 .501 Income 58 42 .58 .496 Employ At Offense 45 55 .45 .500 Criminal History Prior Arrests 33 67 .33 .472 Total Arrests 31 69 .31 .464 Ever Incarcerated 21 79 .21 .409 Number Incarcerated 20 80 .20 .402 28 RESULTS Findings here from descriptive statistics show general information about this study sample. Basic statistics do not allow for a deep understanding of the outcome variable. They do help to create an overview of the frequency of each risk and protective occurrence in the sample. After descriptive statistics were run Chi Square tests of significance were completed, results on each item follow. Chi-Square Tests Of Demographic Variables A chi-square test of independence was performed to examine the relation between crime outcome and family. The relation between these variables was significant at the p<.05 level, (X2 =6.775, df = 1); Violent offenders were more likely than nonviolent offenders to have family categorized here as risk. Meaning there is a difference in the family composure of this sample which may increase the likelihood of transition into violence. A chi-square test of independence was performed to examine the relation between crime outcome and children. The relation between these variables was not significant at the p<.05 level, (X2 =.585, df = 1); Violent offenders were not more likely than nonviolent offenders to have children categorized here as risk. This can be a societal trend in which costs of living is a huge deterrent to family building. A chi-square test of independence was performed to examine the relation between crime outcome and education. The relation between these variables was significant at the p<.05 level, (X2 =27.214, df = 1); Violent offenders were more likely than nonviolent offenders to have education categorized here as risk. This is in line with the literature and continues to reflect the importance of education on life choices. 29 A chi-square test of independence was performed to examine the relation between crime outcome and suicide attempt. The relation between these variables was significant at the p<.05 level, (X2 =7.527, df = 1); Violent offenders were more likely than nonviolent offenders to have suicide attempt categorized here as risk. This is in line with the literature as it stands, and this may be influenced by proximity to violence or a willingness to commit or experience violence. A chi-square test of independence was performed to examine the relation between crime outcome and mental ill. The relation between these variables was significant at the p<.05 level, (X2 =11.960, df = 1); Violent offenders were more likely than nonviolent offenders to have mental ill categorized here as risk. A chi-square test of independence was performed to examine the relation between crime outcome and sex. The relation between these variables was significant at the p<.05 level, (X2 =10.714, df = 1); Violent offenders were more likely than nonviolent offenders to have sex categorized here as risk being a male. A chi-square test of independence was performed to examine the relation between crime outcome and race. The relation between these variables was not significant at the p<.05 level, (X2 =.709, df = 1); Violent offenders were not more likely than nonviolent offenders to have race categorized here as risk. There was a majority of Caucasian offenders in this sample and this is in line with the literature as it stands. A chi-square test of independence was performed to examine the relation between crime outcome and foreign born. The relation between these variables was not significant at the p<.05 level, (X2 =1.042, df = 1); Violent offenders were not more likely than nonviolent offenders to 30 have foreign born categorized here as risk. In line with extremist literature this variable indicates that domestic born individuals are dominant in the sample. A chi-square test of independence was performed to examine the relation between crime outcome and age. The relation between these variables was significant at the p<.05 level, (X2 =54.782, df = 1); Violent offenders were more likely than nonviolent offenders to have age categorized here as risk. This variable is in line with existing literature and relates to developmental life course theory. Chi-Square Tests Of Network Variables A chi-square test of independence was performed to examine the relation between crime outcome and community status. The relation between these variables was significant at the p<.05 level, (X2 =61.060, df = 1); Violent offenders were more likely than nonviolent offenders to have community status categorized here as risk. Meaning violent extremists are more likely to experience low community prestige/status. A chi-square test of independence was performed to examine the relation between crime outcome and family extremist involvement. The relation between these variables was not significant at the p<.05 level, (X2 =.065, df = 1); Violent offenders were not more likely than nonviolent offenders to have family members categorized here as risk. Meaning there is no difference in this sample between the amount or type of community connections that existed for each subject. A chi-square test of independence was performed to examine the relation between crime outcome and religious involvement. The relation between these variables was not significant at the p<.05 level, (X2 =.444, df = 1); Violent offenders were not more likely than nonviolent offenders to have religious involvement categorized here as risk. This can be that in the United 31 States there has been a general trend of being less religiously involved in comparison to past decades. A chi-square test of independence was performed to examine the relation between crime outcome and the recruited variable. The relation between these variables was significant at the p<.05 level, (X2 =8.319, df = 1); Violent offenders were more likely than nonviolent offenders to have recruited categorized here as risk. The statistic is significant, yet the value is lower than the minimum expected count, future studies should increase the sample size and retest this variable. A chi-square test of independence was performed to examine the relation between crime outcome and military awards which represented military experience. The relation between these variables was not significant at the p<.05 level, (X2 =.344, df = 1); Violent offenders were not more likely than nonviolent offenders to have military awards categorized here as risk. Something that may be interesting is testing this again with more current data, given the collection period of subjects. A chi-square test of independence was performed to examine the relation between crime outcome and marital status. The relation between these variables was not significant at the p<.05 level, (X2 =.386, df = 1); Violent offenders were not more likely than nonviolent offenders to have marital status categorized here as risk. This is something that this author would expect to stay the same given trends in society and general divorce rates increasing. A chi-square test of independence was performed to examine the relation between crime outcome and personal grievance. The relation between these variables was significant at the p<.05 level, (X2 =25.668, df = 1); Violent offenders were less likely than nonviolent offenders to have personal grievance categorized here as risk. This is in line with literature that violent 32 killings could stem from the hateful belief that others are subhuman but may target random individuals due to opportunity. So, although the individual may not have done anything to the perpetrator – they still fall within the scope of their ideology’s particular grievance. Financial crime subjects do generally have grievances or hostile attitudes towards the government- grievance here is considered anti-government sentiment or legal arguments present in the search files. A chi-square test of independence was performed to examine the relation between crime outcome and political activity scale. The relation between these variables was not significant at the p<.05 level, (X2 =.437, df = 1); Violent offenders were not more likely than nonviolent offenders to have political scale categorized here as risk. This variable may be better disaggregated – but also may be more relevant to group studies given measurement issues with the current data. A chi-square test of independence was performed to examine the relation between crime outcome and shared info web. The relation between these variables was not significant at the p<.05 level, (X2 =.233, df = 1); Violent offenders were not more likely than nonviolent offenders to have shared info web categorized here as risk. The sharing of details about a crime given the time period of data collection makes sense, I would expect to see an increase in this variable in future studies utilizing new data. A chi-square test of independence was performed to examine the relation between crime outcome and lone wolf. The relation between these variables was not significant at the p<.05 level, (X2 =2.852, df = 1); Violent offenders were not more likely than nonviolent offenders to have lone wolf categorized here as risk. There is no difference and that means the individual non-violent subject can be at risk for becoming violent as a lone wolf or a part of a group given 33 enough adverse events. While the crime could be committed alone there is not apparent difference between loner behavior in either population. Chi-Square Tests Of Financial Variables A chi-square test of independence was performed to examine the relation between crime outcome and community ties. The relation between these variables was not significant at the p<.05 level, (X2 =.361, df = 1); Violent offenders were not more likely than nonviolent offenders to have community ties categorized here as risk. Meaning there is no difference in this sample between the amount or type of community connections that existed for each subject. A chi-square test of independence was performed to examine the relation between crime outcome and income. The relation between these variables was significant at the p<.05 level, (X2 =47.455, df = 1); Violent offenders were more likely than nonviolent offenders to have income categorized here as risk. This may continue to be the case as money is a crucial motivator for everyone. This is interesting because non-violent financial crimes may just indicate greed and selfishness. A chi-square test of independence was performed to examine the relation between crime outcome and employ at offense. The relation between these variables was significant at the p<.05 level, (X2 =9.091, df = 1); Violent offenders were more likely than nonviolent offenders to have employ at offense categorized here as risk. Violent extremists are more likely to be unemployed or work for the extremist organization in comparison to non-violent having employment or being in the protective category. Chi-Square Tests Of Criminal Variables A chi-square test of independence was performed to examine the relation between crime outcome and prior arrest. The relation between these variables was significant at the p<.05 level, (X2 =19.946, df = 1); Violent offenders were more likely than nonviolent offenders to have prior 34 arrests categorized here as risk. This is in line with current literature as we would expect those who have justice system contact to be more prepared to use violence. A chi-square test of independence was performed to examine the relation between crime outcome and total arrests. The relation between these variables was significant at the p<.05 level, (X2 =16.877, df = 1); Violent offenders were more likely than nonviolent offenders to have total arrests categorized here as risk. Meaning there is a difference between these groups when observing the respective total arrest counts. This makes sense as we expect more contact with the criminal justice system through past behavior. A chi-square test of independence was performed to examine the relation between crime outcome and number incarcerated. The relation between these variables was significant at the p<.05 level, (X2 =12.250, df = 1); Violent offenders were more likely than nonviolent offenders to have number incarcerated categorized here as risk. Multivariate Analysis There were multiple logistic models presented, but each model was completed with multiple variable combinations. The challenge faced was having many predictor variables of interest with an overall small sample. The study strategy was to attempt to identify the “best fit” variables within each general category of variables (demographic, network, financial, criminal history). Each model represented a combination of variables tested on the outcome variable for the violent and nonviolent extremists in the sample. After discussing these best fit variables, there will be discussion regarding which of the best fit variables remained significant when combined with the other variables. In model 1, only the demographic variables of interest were tested. Specifically, variables included: family, children, education, suicide attempt, mental illness, sex, race, foreign born, and 35 age. As you can see in Table 2, most of the demographic variables were not significant but this grouping had a pseudo R2 value of .805 (80.5% explanation power). Specifically, family, children, suicide attempt, mental illness, race, and foreign born were not significant. Education was significant at the p<.01 level with violent offenders being 2.30 times more likely to be less educated when compared to nonviolent offenders. Specifically, extremists who have only a high school education or less are significantly more likely to commit a violent crime. Similarly, males were significantly more likely to commit a violent offense compared to females and was significant at the p<.05 level with males being 3 times more likely to commit a violent offense. Finally, age was significant at the p<.000 level with violent offenders 4.27 times more likely to be within the 16- to 35-year-old age group when compared to nonviolent extremist. As the age of the extremist increased, so did the likelihood that the extremist would commit a violent crime. In model 2, specifically only the network variables of interest were tested. These specifically included: community status, family extremism, religious involvement, recruited, military awards, marital status, personal grievance, political scale, shared info web, and lone wolf. When you refer to Table 2 you can see once again most variables are not significant and had a pseudo R2 value of .813 (81.3% explanation power). Specifically, family extremism, religious involvement, recruited, military awards, marital status, political scale, shared info web and lone wolf. Community status was significant at the p<.000 level meaning that violent offenders in this sample were about 6 times more likely to be of lower community status and prestige compared to nonviolent offenders. Personal grievance was also significant at the p<.001 level with violent offenders being 4 times less likely to have a grievance when committing crime. This indicates that violent extremists may be committing instrumental crime or network 36 influenced crime whereas nonviolent offenders typically in this sample had a personal anti- government grievance. In model 3, specifically examined only the financial variables of interest. Variables specifically included: community ties, income, and employment when committing the offense. As you see in Table 2, community ties and employment were not significant, and this model had a pseudo R2 of .556 (55.6% explanation power). Income was significant at the p<.000 level with extremists having low or no income being 3.64 times more likely to be violent offenders when compared to nonviolent offenders. Specifically, this may indicate that these subjects may feel justified in their use of violence when trying to secure their livelihood compared to nonviolent offenders. Community ties represented living arrangements whereas employ at offense represented type of method of income, regardless having income or not influences the choices made by subjects. In model 4, specifically examined only criminal history variables of interest. Variables specifically included: prior arrest and number incarcerated. There were originally 4 variables - which were correlated in this grouping, total arrests correlated with prior arrests and ever incarcerated with number incarcerated were the less significant counterparts and were removed from the final model analysis. As you have seen in Table 2, number incarcerated was not significant and the model had a pseudo R2 of .258 (25.8% explanation power). Prior arrest was significant at the p<.01 level and violent offenders were 1.85 times more likely to have had a prior arrest when compared to nonviolent offenders. This result meant that having a prior arrest may indicate a disposition which may eventually lead to violent crime when compared to nonviolent offenders. 37 In the final step of the analysis, the author wanted to test variables from the different categories together, but the modeling was difficult for a couple of reasons. One challenge was that when focusing only on the variables that were statistically significant in Table 2 the results of combining models were not a good fit. The small sample of 100 subjects most likely is the reason that the most significant variables could not be tested together and requires a much larger sample size. Another challenge was creating groupings that were meant to be theoretically grounded, and then trying to combine them with other groupings influenced by other theories meant the effects of one variable would override and throw the whole model off course. Specifically, education, sex, and age were significant together on one round of analysis. On another test iteration income was added to the last set of variables, income and age were significant while education and sex became insignificant. In another round testing age, community status and income none of the variables were significant. Personal grievance and prior arrests were tested on the dependent variable because all other variables became insignificant – in this instance personal grievance and prior arrests were significant. The next grouping, tested included education, sex, age, and personal grievance. Education, age, and personal grievance were significant, but sex was insignificant. In the second to last combination of testing included education, age, personal grievance, and prior arrests; education stayed significant while the others became insignificant. In the second to final testing group, education, sex, community status, personal grievance, and prior arrests were combined. In that grouping community status and personal grievance were significant – then education, sex, and prior arrests were insignificant. The goal was to test various groupings and reduce the variables tested until only the most significant variables could be tested together. In the final testing group age, personal grievance, income and prior arrest were tested and resulted in a R2 value of .932; 38 income was significant at p<.05 with a B of 3.84 this last test was one of the best models. The reduction goal was achieved but creating a bigger model from those variables failed. 39 Table 2 Group Regression Statistics Variables P≤ B Nagelkerke R2 Best Fit Demographic 0.805 Family .809 0.295 Children .326 -1.172 Education .008** 2.306 Education** Suicide Attempt .999 19.532 Mental Ill .156 1.935 Sex .044* 2.822 Sex* Race .546 -1.417 Foreign Born .762 1.163 Age .000*** 4.279 Age*** Network 0.813 Community Status .000*** 5.973 Community Family Extremism .446 -0.921 Status *** Religious Involvement .513 -1.212 Recruited .224 1.338 Military Awards .607 1.281 Marital Status .151 -1.458 Personal Grievance .001** -4.088 Personal Political Scale .113 1.883 Grievance ** Shared Info Web .914 -0.119 Lone Wolf .422 -1.003 Financial 0.556 Community Ties .357 -0.523 Income .000*** 3.649 Income*** Employ At Offense .997 -0.003 Criminal History 0.258 Prior Arrests .009** 1.853 Prior Arrests** Number Incarcerated .559 0.531 *p≤.05; **p≤.01; ***p≤.001 40 DISCUSSION There were several variables that helped to differentiate between violent and nonviolent offenders. Community status represented through the network grouping was most significant in the final tests with a positive B of 5.973 at the p<.000 level. This is interesting as the data seemed to point to environment as a major difference between the two populations, this community standing inevitably impacts all other variables and lifestyle choices. This variable predicted violent extremists to be 5.97 times more likely to be of lower community standing and commit violence than nonviolent extremists. Future research may consider income using a variable with more variation, but here it was significant in the final model with p<.000 with a B of 3.64. Personal grievance was also significant at the p<.001 level and predicted that violent extremists were 4.08 times less likely to have a personal grievance with the victim compared to nonviolent extremists. This may be the case in this sample due to violent offenders attacking individuals at random, whereas nonviolent offenders committing offenses tended to have a specific anti-government grievance. The results seem to point out that violent offenders will attack at random in this sample even though their motivations are generally rooted in an ideology that groups others as targets. Literature has pointed out that there is not a need for a particular grievance to commit violence simply the opportunity to commit it may be enough if an individual already intends to commit violence. Education was significant at the p<.01 level and violent extremists were 2.30 times more likely to have a lower education level than nonviolent offenders. The final test in the model as stated was insignificant when more than 5 variables were tested together but future statistical methods and a larger sample size will retest this and other findings on this population and should also study far-left extremists and Jihadist’s extremists. Sex was significant at the p<.05 level and violent offenders were almost 3 times more likely to be male when compared to nonviolent 41 offenders. This means we would expect to continue seeing males as perpetrators of violent crime compared to females in this sample. The Criminal History grouping had Prior Arrest as the most significant predictor with a significance of p<.05 violent offenders were 1.85 times more likely to have had a prior arrest at the time of being convicted for the crime in the study (Model 4). The author learned several valuable lessons from this project. First, the major lesson learned is that more statistical power is needed and a bigger sample size to truly get better fit models. It seems as if multiple variables were measuring similar concepts, and this could have impacted the analysis by means of collinearity effects. All these variables have support in the literature but measurement issues and data challenges along with small sample size has made many significant variables appear insignificant when modeled together. What seems to be clear is that violent extremists are statistically different regarding community status, younger in age and generally do not need a grievance to commit an attack. They tend to have a prior arrest, generally are male and have lower educational attainment and low or no legitimate income. Another interesting point to be made is that these variables were significant when tested with most other variables but once tested with only significant variables the whole model collapsed. The study results were better running smaller models of varying mixed grouping combinations – the problem with this method is that it is not theoretically grounded, and the small sample size makes it clear that more than 100 subjects are required. Previous research has indicated the variables tested in this study are important, this study confirms once again the predictive powers, respectively. Future research would include more recent files on domestic extremist beyond 2014 and include popular methods such as massive data scraping efforts to identify and counter extremists (Holt, Scrivens, Chermak). Future 42 research may want to include data scientist who can help create more logical models efficiently. Future research should incorporate working with exonerees, formers of extremist movements, and clinical psychologists who work on risk and protective factors. Future research looking into risk and protective factors can help make measures more precise and result in more significant differences being identified. In this study a few variables were excluded: affiliation which had more response options of interest that were considered risks than options for protective. In the future there will have to be clean categories created when collecting data that represents protective affiliations. This is the same for religion, there were extensive subdivisions of religion, but not enough differentiation existed to justify using religion in the study; as the other options (protective) for religion were either Atheist, Agnostic, Other or radical Islam variations. In this case the study used religious involvement as a proxy for any religious involvement without a need to belabor specifying which religion. Given the dichotomous nature of this study the variables removed simply may have measured too many concepts to be distilled into two simple categories for this study. Religion and affiliation may be better used in extensive case studies but did not work well in this study. As well in this study U.S. Citizen first was included but was removed as it was a duplicate of the variable foreign born which is more relevant in this study to allow for seeing if there were foreign born domestic extremists. Although this study focused on domestic far-right extremists’ risk and protective measures can be further refined by conducting future research on far left and Jihadist violent and nonviolent offender populations. In this study the author learned just how much sample size impacts statistical precision especially with two outcome groups and 29 independent variables. In the future a sample size of 500 or more cases may be the minimum study sample to have robust results. This study has 43 pointed out that there are not many significant differences in violent and nonviolent offenders. Given the study was open source an important lesson was learned regarding the limitations of cleaning coded data and properly conceptualizing concepts in order to statistically test for significance. This study has also informed how the author will ask questions when interviewing individuals and deciding which questions should be asked. In the future gender identity may be included in the study to account for societal changes and identity what can be a risk factor. In the future researchers will have to ask formers and exonerees which protective factors they experienced and how we can create more opportunities for desisting. Researchers can ask current extremists to rank extant risk factors significance and their risks experienced. Future studies will have to account for online activities and average time spent online to capture the use of the Internet as a risk or protective factor. Future studies should equally focus on protective factors and begin studying various reference groups to build more knowledge in this area of counter- radicalization or deradicalization. Finally, in the future studies will need to be more theoretically grounded by having less variables but more pertinent variables to test with at least 500 subjects. 44 BIBLIOGRAPHY 45 BIBLIOGRAPHY Adamczyk, A., Gruenewald, J., Chermak, S. M. & Freilich, J. D. (2014). The relationship between hate groups and far-right ideological violence. Journal of Contemporary Criminal Justice, 30(3), 310-322. Asal, V., Chermak, S. M., Fitzgerald, S. & Freilich, J. D. (2020). Organizational-level characteristics in right-wing extremist groups in the United States over time. Criminal Justice Review, 45(2), 250-266. Auger (2020). Right-Wing Terror: A Fifth global wave. Perspectives on Terrorism, 14(3), 87-97. Bhui, K, Silva, M.J., Topciu, R.A., & Jones, E. (2016). Pathways to sympathies for violent protest and terrorism. British Journal of Psychiatry, 209(6), 483-490. Censky, A. (2020). Heavily armed protestors gather again at Michigan Capitol to decry Stay-At- Home order. NPR, Retrieved from https://www.npr.org/2020/05/14/855918852/heavily- armed-protestors-gather-again-at-michigans-capitol-denouncing-home-order , Accessed 19 Jan 2021. Chermak, S.M., & Gruenewald, J. (2006). The Media’s Coverage of Domestic Terrorism. Justice Quarterly, 23(4), 429-460. Chermak, S. M., Freilich, J. D. & Simone Jr., J. (2010). Surveying American State Police agencies about Lone Wolves, Far-Right criminality, and Far-Right Jihadist criminal collaboration. Studies in Conflict & Terrorism, 33(11), 1019-1041. Chermak, S. M., Freilich, J. & Suttmoeller, M. (2013). The organizational dynamics of Far-Right hate groups in the United States: Comparing Violent and Nonviolent Organizations, Studies in Conflict & Terrorism, 36(3), 193-218. Clemmow, C. (2020). Risk Factors and indicators for engagement in violent extremism. [Unpublished Doctoral Dissertation]. University College London. Cohen, K., Johansson, F., Kaati, L. & Mork, J. C. (2014). Detecting linguistic markers for radical violence in social media. Terrorism and Political Violence, 26(1), 246-256. Doering, S., Davies, G & Corrado, R. (2020). Reconceptualizing ideology and extremism: Toward an empirically-based typology. Studies in Conflict & Terrorism. Retrieved from: https://doi.org/10.1080/1057610X.2020.1793452 # 46 Fandos, N. & Cochrane, E. (2021). After Pro-Trump mob swarms capitol, Congress confirms Biden’s Win. The New York Times, The New York Times, Retrieved from www.nytimes.com/2021/01/06/us/politics/congress-gop-subvert-election.html , Accessed 19 Jan. 2021. Federal Bureau of Investigation [FBI]. (2005). Terrorism: 2002-2005. United States Department of Justice Report. Retrieved from http://www.fbi.gov/stats- services/publications/terrorism-2002-2005 Also retrieved from https://www.fbi.gov/investigate/terrorism n.d. Accessed 19 Jan 2021. Fitzpatrick, K. M., Gruenewald, J., Smith, B. L. & Roberts. (2017). A community-level comparison of terrorism movements in the United States. Studies in Conflict & Terrorism, 40(5), 399-418. Flesher, J. (2020). 6 men indicted in alleged plot to kidnap Michigan Governor. AP NEWS, Retrieved from https://apnews.com/article/gretchen-whitmer-michigan-indictments- coronavirus-pandemic-traverse-city- 10f7e02c57004da9843f89650edd4510#:~:text=(AP)%20%E2%80%94%20A%20federal %20grand,angry%20over%20her%coronavirus%20policies , Accessed 19 Jan. 2021. Freilich, J. D., Chermak, S. M. & Simone Jr., J. (2009). Surveying American State police agencies about terrorism threats, terrorism sources, and terrorism definitions. Terrorism and Political Violence, 21:450-475. Freilich, J. D., Chermak, S. M., Belli, R., Gruenewald, G. & Parkin, W. (2014). Introducing the United States Extremis Crime Database (ECDB). Terrorism and Political Violence, 26(2), 372-384. Freilich JD, Chermak SM, Gruenewald J (2015) The future of terrorism research: a review essay. International Journal of Comparative Applied Criminal Justice 39(4):353–369 Freilich, J. D., Chermak, S. M., Gruenewald, J., Parkin, W. S. & Klein, B. R. (2018). Patterns of fatal extreme-right crime in the United States. Perspectives on Terrorism, 12(6), 38-51. Freilich, J. D., Parkin, S., Gruenewald, J. & Chermak, S. M. (2019). Comparing extremist perpetrators of suicide and non-suicide attacks in the United States. Terrorism and Political Violence, 31(5), 943-965. Freilich, J. D., Bejan, V., Parkin, W. S., Chermak, S. M. & Gruenewald, J. (2020a). An intervention analysis of fatal far-right extremist violence within a vector-autoregressive framework. Dynamics of Asymmetric Conflict, 13(2), 143-171. 47 Freilich, J.D., Chermak, S. M. & Klein, B. R. (2020b). Investigating the applicability of situational crime prevention to the public mass violence context. Criminology & Public Policy, Special Issue Article: Countering Mass Violent in the United States, 19:271-293. Freilich, J.D., Chermak, S. M. & Klein, B. R. (2020c). Investigating the applicability of situational crime prevention to the public mass violence context. Criminology & Public Policy, Special Issue Article: Countering Mass Violent in the United States, 19:271-293. Gruenewald, Chermak, Freilich (2013). Distinguishing loner attacks from other domestic extremist violence: Comparison of far-right homicide incident and offender characteristics. Criminology & Public Policy, 12(1), 65-92. Gruenewald, J., Dooley, M. G., Suttmoeller, M. J., Chermak, S. M. & Freilich, J. D. (2016). A Mixed-Method analysis of fatal attacks on police by far-right extremists. Police Quarterly, 19(2), 216-245. Gruenewald, J., Drawve, G. & Smith, B. (2019). The situated contexts f American terrorism: A conjunctive analysis of case configurations. Criminal Justice and Behavior, 46(6), 884- 901. Gruenewald, J., Klein, B. R., Drawve, G., Smith, B. L. & Ratcliff, K. (2019). Suspicious preoperational activities and law enforcement interdiction of terrorist plots. Policing: An International Journal, 42(1), 89-107. Hayes, B. E., Mills, C. E., Freilich, J. D. & Chermak, S. M. (2018). Are honor killings unique? A comparison of honor killings, domestic violence homicides, and hate homicides by Far- Right extremists. Homicide Studies, 22(1), 70-93. Holt, T. J., Freilich, J. D. & Chermak, S. M. (2020). Examining the online expression of ideology among Far-Right Extremist forum users. Terrorism and Political Violence. https://doi.org/10.1080/09546553.2019.1701446 Klein, B. R., Gruenewald, J. & Smith, B. L. (2017). Opportunity, group structure, temporal patterns, and successful outcomes of Far-Right terrorism incidents in the United States. Crime & Delinquency, 63(10), 1224-1249. LaFree, G. (2013). Lone-offender terrorists. Criminology & Public Policy, 12(1), 59-62. Misiak B, Samochowiec J, Bhui K, Schouler-Ocak M, Demunter H, Kuey L et al (2019) A systematic review on the relationship between mental health, radicalization and mass violence. European Psychiatry 56, 51–59 48 Perliger A, Koehler-Derrick G, Pedahzur A (2016) The gap between participation and violence: why we need to disaggregate terrorist ‘profiles’. Int Stud Quart 60(2):220–229 Powell, K. A. (2018). Framing Islam/creating fear: an analysis of U.S. media coverage of terrorism from 2011-2016. Religions, 9(9), 257–271. https://doi.org/10.3390/rel9090257 Schildkraut, J., Gruenewald, J., Gilliam, M & Novak, M. (2020). Framing mass shootings as a social problem: A comparison of ideologically and non-ideologically motivated attacks. Aggression and Violent behavior, https://doi.org/10.1016/j.avb.2020.101533 Scrivens, R., Wojciechowski, T. W., Freilich, J. D., Chermak, S. M. & Frank, R. (2020). Comparing the online posting behaviors of violent and Non-violent Right-Wing Extremist forums. In press. Silva, J. R., Duran, C., Freilich, J. D & Chermak, S. M. (2020). Addressing the myths of terrorism in America. International Criminal Justice Review, 30(3), 302-324. Simi, P. & Windisch, S. (2020). Why radicalization fails: Barriers to mass casualty terrorism. Terrorism and Political Violence, 32(4), 831-850. Smith, B. L., Gruenewald, J., Roberts, P. & Damphousse, K. R. (2015). The emergence of Lone Wolf terrorism: Patterns of behavior and implications for intervention. Terrorism and Counterterrorism Today; Sociology of Crime, Law and Deviance, 20, 91-110. Vergani, M., Navarro, C., Freilich, J. D. & Chermak, S. M. (2020). Comparing different sources of data to examine trends of hate crime in absence of official registers. American Journal of Criminal Justice, Retrieved from: https://doi.org/10.1007/s12103-020-09567-9 Wells, W. & Chermak, S. M. (2011). Individual-Level Risk Factors for Gun Victimization in a Sample of Probationers. Journal of Interpersonal Violence, 26(11), 2143-2164. Windisch, S., Simi, P., Blee, K. & DeMichele, M. (2020). Measuring the extent and nature of adverse childhood experiences (ACE) among former White supremacists. Terrorism and Political Violence, https://doi.org/10.1080/09546553.2020.1767604Wojcieszak M (2010) ‘Don’t talk to me’: effects of ideologically homogeneous online groups and politically dissimilar offline ties on extremism. New Media Soc 12(4):637–655 49