1 Capstone right-wing extremism
Crisis and Security Management Faculty of Governance and Global Affairs
Understanding extremism from a criminological perspective
A case study on social structure and social learning theory applied to right-extremist behaviour
Master’s Thesis Leiden University
Pepijn Hack
Thesis supervisor: Dr. Y.D. Veilleux-Lepage Second reader: S. Wittendorp MA
Word count: 22075 29-06-2020
2 Abstract
In recent years multiple scholars have argued for the applicability of Akers’ social structure and social learning theory in explaining ideologically motivated crimes. To this date little research has been done and the results remain inconclusive. The aim of this thesis is to fills this gap in the literature using a qualitative case study analysis. Using the Global Terrorism Database, five cases of far-right extremist attacks between 2011 and 2017 have been selected. Publicly available data has been used to gather information on the subjects. For each case study a life course overview has been created, the subjects’ entry into extremism has been described and the different theoretical variables are outlined. This thesis concludes that social learning variables are applicable to most cases and social learning mechanisms are able to explain the subjects’ behaviour. However, the structural variables presented in Akers’ theory were not present in most cases and the mechanisms laid out in the theory were not applicable. After discussing the limitations of this thesis possible avenues for future research and practical recommendations for countering extremist behaviour are provided.
3 Table of contents
Abstract ... 2
Introduction ... 4
Literature review ... 9
Explanation of Social Learning Theory (SLT) ... 11
Literature on SLT ... 13
Explanation of Social Structure and Social Learning Theory (SSSL) ... 16
Literature on SSSL ... 18
Methodology ... 21
Data... 22
Research quality and limitations ... 24
Results ... 26 Anders Breivik... 26 Elliot Rodger ... 30 Alexandre Bissonnette ... 34 Thomas Mair ... 37 Dylann Roof ... 41 Conclusion ... 46 Discussion ... 49 Bibliography ... 54
4 Introduction
On the 9th of October 2019 Stephan Balliet shot two people and attempted to kill many
more in a synagogue in Halle, Germany (NOS, 2019). During his attack he was livestreaming his actions to the online livestreaming platform twitch.tv. The broadcast was seen by nearly 2200 people before the video was taken down (AD, 2019a). A public prosecutor spokesperson told German media that there were sufficient leads to think of this as a right-wing extremist attack. The first words Balliet spoke in his video were “Hi, I’m Anon and I don’t belief the Holocaust happened” (Lindhout, 2019, p. 1). Anon, short for anonymous, is a popular handle on the internet forums 4chan and its even more radical but much lesser used brother 8chan. On both websites the video of the attack was shared after it took place. Balliet wanted his attack to be seen by likeminded people (Lindhout, 2019). This also becomes clear from the manifesto Balliet posted shortly before going live. He states he wanted to “prove the viability of improvised weapons”, “increase the moral of other suppressed Whites by spreading the combat footage”, “kill as many anti-Whites as possible, Jews preferred”, and as a “bonus: Don’t die” (Balliet, 2019, p. 13).
The way in which Balliet let the world be part of his act of terror, a first-person livestream and uploading his manifesto, reminds us of a previous attack in Christchurch, New Zealand. In March 2019, the Australian Brenton Tarrant killed 51 Muslims in two mosques (AD, 2019b). Tarrant also uploaded a manifesto before his attack, and he livestreamed his actions on 4chan and Facebook. In his manifesto he said he was inspired by yet another right-wing extremist, Anders Breivik. After uploading a 1500-page “European Declaration of Independence”, Breivik killed 77 people during his terrorist attack in Oslo and Utøya. In his document he explained his motivations and pro-white worldviews (AD, 2019b). But these were not the only attacks. Philip Manshaus in Oslo, Patrick Crusion in El Paso, John Earnest in Poway, and Robert Bowes in Pittsburgh. All very recent (attempted) terrorist attacks with an extremist motivation from actors who are not part of an organised terrorist group. From a societal point of view, it is necessary to understand the way in which people such as Balliet and others like him enter into extremism and radicalize to the point that they carry out their attack. It is also important to understand what can be done against it.
Social structure and social learning theory provides a framework to understand why people commit crimes (Akers, 2009). The social structure part of the theory explains how variations in the social organisation and culture of communities, combined with the locations of individuals and groups in the social system explain variations in crime rates. The social learning part of the theory explains crime as “learned behaviour through social interaction with
5 others” (Akers & Sellers, 2004, p. 89). The chance that a person will commit a crime is increased when they differentially associate, directly or indirectly, with others who commit similar crimes. They adopt definitions favourable to criminal behaviour through differential association and imitation and anticipate greater reward than punishment for their behaviour (Akers & Silverman, 2014). The social structural variables provide the context in which social learning takes place. Akers’ theory states that these structural variables do not predict crime itself, but when these variables are present in the community or on a personal level of the subject, a person is more likely to show social learning variables, which in turn predict crime (Akers & Silverman, 2014).
Multiple scholars have argued that social structure and social learning theory should also be applicable to acts of terrorism or extremist behaviour (Akers, 2009; Akers & Sellers, 2011; Akers & Silverman, 2014; Akins & Winfree, 2016; Breen, 2019; Freilich & LaFree, 2015; Ryan, Vanderlick, & Matthews, 2007). Although there has been some research regarding social structure and or social learning theory, all studies so far have shown one or more shortcomings. Almost all studies have focused on testing only (a part of) social learning theory, ignoring the social structure elements which are crucial in Aker’s theory. Similarly, studies have only used data from radicalised people or terrorists in the US, not taking into account different cultures and countries. This thesis tries to overcome these shortcomings by using a framework that looks at both social structure and social learning elements, in multiple case studies, to fully try to comprehend to what extent SSSL theory can helps us understand extremist behaviour.
This thesis not only helps getting a better understanding of extremist behaviour but also provides clear policy implications on how to fight it. If we have a better understanding of the mechanisms that influence extremist behaviour and extremists’ thought processes, anti-terrorism agencies, social workers, police agencies and policy makers can make better informed decisions on how to fight against it. The specific research question this thesis answers is “To what extent can Aker’ social structure and social learning theory be used to explain
far-right extremist behaviour”.1 This thesis argues that social structure and social learning theory
1 This thesis tries to explain extremist behaviour, seen as a form of ‘non-conforming’ behaviour such as described
by Akers in his theory. It does not try to explain a ‘radicalisation process’. Despite the fact that an argument could be made to see the change in someone’s beliefs as being the same as his or hers radicalisation process, the notion of radicalisation is a contested subject with a long history of many different definitions, understandings and nuances (Coolsaet, 2019). Furthermore, it is contested because of the multitude of positive non-harmful connotations that ‘radicalism’ has (von Behr, Reding, Edwards, & Gribbon, 2013). For the purpose of this thesis it is not of added value to use the term ‘radicalisation’. Akers theory tries to explain why people commit both
6 is only partially able to explain extremist behaviour. The social learning variables within the theory are largely applicable, but the social structure variables are largely not applicable to the five case studies.
To answer the research question, a qualitative case study analysis has been performed. Using the Global Terrorism Database (LaFree & Dugan, 2007), five cases of far-right extremist attacks between 2011 and 2017 have been selected. Publicly available data such as newspaper articles, court and investigation documents, psychiatric evaluations, blogs, scholarly work, books, and the perpetrator’s own publications such as manifestos and social media accounts have been used to gather information on the subjects. These cases have then been analysed using a social structure and social learning framework to test the applicability of the multitude of variables included in this theory.
In the following sections the concepts used in the research question are defined. In the literature review chapter, an overview will be given of how and why using a criminological theory to explain extremist behaviour makes sense, extremist behaviour is after all a form of crime. After that, social structure and social learning theory will be explained and the current body of literature regarding this theory will be illustrated. This section shows that the literature regarding social learning and social structure theory is still inconclusive and further testing needs to be done. In the methodology section a detailed overview will be given of the used methods, how the data has been selected and analysed, and the research quality and limitations of this design will be discussed. In the discussion, an answer to the research question will be presented, which will be discussed in relation to the current body of literature. It is concluded that social learning variables can partially explain extremist behaviour whereas social structure variables are only very limited in application. The results show similarities to the current body of literature, but also show differences that need further exploration. Lastly, the limitations of the research will be discussed with regards to possible avenues for future research, and based on the results of this thesis, practical recommendations will be provided for countering extremist behaviour.
Before going into discussing the current body of literature regarding social structure and social learning theory, the concepts used in the research question need to be defined. In both Dutch as well as international scholarly works the concepts of terrorism, (violent) extremism and political violence are often used interchangeably (Coolsaet, 2019;
Nasser-conforming and nonNasser-conforming behaviour (Akers, 1998). It does not make distinctions between different types of nonconforming behaviour.
7 Eddine, Garnham, Agostino, & Caluya, 2011). A proper distinction between the concepts of terrorism and extremist behaviour is therefore difficult to make. However, for the purpose of this thesis the same definition of extremism as Carlsson et al. (2019) and LaFree, Jensen, James, and Safer-Lichtenstein (2018) is used. Extremist behaviour is seen as (the threat of) violence used to further particular beliefs, including those of a political, religious, social or ideological nature by a person defined as extreme on a given scale (such as political, religious, social or ideological).
Secondly, this thesis uses Castelli Gattinara and Pirro’s view of the far right as a subcultural group. These authors describe three forms of far-right groups; political parties, social movements and a “conglomeration of groups within the subcultural environment” (Castelli Gattinara & Pirro, 2019, p. 449). They see the far right as an umbrella concept in which far-right actors are defined as groups or individuals located on the “right end of the ideological left-right continuum” (Castelli Gattinara & Pirro, 2019, p. 450). The subjects described in this thesis all fall under this ‘umbrella’ definition of far-right actors as a subcultural group of individuals. They do not belong to political parties that aim to win elections, nor do they belong to social movements or networks that try to mobilise the public opinion. They are actors that hold ideological views that overlap with the far right, and some are even trying to mobilise others to start taking action, but none of them are part of a group.2
Thirdly, this thesis refrains from defining terrorism itself but instead relies on the inclusion criteria used in the Global Terrorism Database to select its cases (LaFree & Dugan, 2007). In the codebook of this database terrorism is defined as “the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious, or social goal through fear, coercion, or intimidation” (University of Maryland, 2019). To include an attack into the database, all three of these factors must be present: (1) “the incident must be intentional”, (2) “the incident must entail some level of violence or immediate threat of violence”, and (3) the perpetrators of the incidents must be sub-national actors” (University of Maryland, 2019, pp. 10-11). Furthermore, two of the following three criteria must be present:
2 Some authors would describe these actors as ‘lone wolfs’, ‘lone actors’ or any other words used to describe
terrorist who do not belong to an organised network (Bullington, 2019; Ellis, et al., 2016; Pantucci, 2011). These concepts are however not uncontested in scholarly literature (Mchauley & Segal, 2009; Schuurman, et al., 2019; Spaaij & Hamm, 2015). Critics state that there are very few examples of true “lone wolfs” who radicalized, planned, prepared, and acted out their terrorist attack on their own. It is also argued that “lone wolf” terrorism is nothing but a social construct and the answer if they exist depends on the definitions you use. For the purpose of this thesis it is not necessary to delve into the question whether or not the subjects described are lone actors or not, because it is not an imperative part of Akers’ theory. The name we give our subjects does not change the actions they have taken in the past.
8 (1) “the act must be aimed at attaining a political, economic, religious, or social goal”, (2) “There must be evidence of an intention to coerce, intimidate, or convey some other message to a larger audience (or audiences) than the immediate victims”, and (3) the action must be outside the context of legitimate warfare activities” (University of Maryland, 2019, p. 11). An additional filtering mechanism is present for cases which are doubted to be proper terrorism attacks because they could also be seen as other types of crime. In all five cases described in this thesis this was not the case.
9 Literature review
When studying terrorism and extremist behaviour in general, scholars have used insights from different disciplines such as political science (Abrahms, 2012; Neumann & Kleinmann, 2013), psychology (Horgan, 2008; Kruglanski, et al., 2014), communication studies (Corman, Trethewey, & Goodall, 2008; Al-Lami, 2009), and sociology (Boyns & Ballard, 2004; Oberschall, 2004; Walsch, 2016). Recently, LaFree and Freilich (2016) published the Handbook of the Criminology of Terrorism. Before this book was published there were some loose efforts of explaining terrorism and extremism by using criminological theories. However, this book marked the beginning of a more organised and coordinated effort. In this book the authors bundled insights from multiple prominent criminological theories such as strain theory (Agnew, 2016), situational and situational action theory (Hsu & Newman, 2016; Wikström & Bouhana, 2016), social learning theory (Akins & Winfree, 2016), and victimization theories (Parkin, 2016). Most of these chapters consisted of theoretical applications of a certain theory to the concept of terrorism. Akins and Winfree (2016) for example make a strong plea for the applicability of social learning theory in explaining how people become terrorists, but do not empirically test their assumptions.
From this point of departure criminological scholars have since started to look into the empirical applicability of different theories using different kinds of methods. Larsen and Jensen (2019) have looked at Jihadist rap from a subcultural perspective where others have used content analysis of Jihadists materials and a single case study approach to test rational choice theory (Perry & Hasisi, 2014). Simi, Sporer, and Bubolz (2016) used a life-course criminological approach by interviewing 44 former violent white supremacist group members. Freilich, Chermak, and Caspi (2009) looked at this same phenomenon by using a case study approach. Some scholars have also used databases such as the US Extremist Crime Database (Freilich, Adamczyk, Chermak, Boyd, & Parkin, 2015), the PIRUS database (LaFree et al., 2018) or the Global Terrorism Database (Pyrooz, LaFree, Decker, & James, 2017).
This development of integrating criminology into the field of extremism and terrorism studies is interesting because terrorism is after all at its core a form of criminological behaviour (Freilich & Lafree, 2015; Liem & Bakker, 2019). Although there are some differences in the goals that terrorists and regular criminals pursue, criminals often strive towards economic gains and do not wish to influence socio-political ideas, they are both forms of deviant behaviour which diverges from commonly accepted norms and values. Furthermore, conventional criminals and terrorist often show similarities in sociodemographic characteristics such as age and gender (LaFree & Ackerman, 2009). These similarities are part of the reason social
10 structure and social learning theory shows promise. It is primarily a theory to explain nonconforming behaviour which should therefore also be applicable to forms of extremism (Akers & Silverman, 2014).
In the last decade, a number of scholars have used theoretical arguments to advocate its use in explain extremism (Akers, 2009; Akers & Sellers, 2011; Akers & Silverman, 2014; Akins & Winfree, 2016; Breen, 2019; Freilich & LaFree, 2015; Ryan et al., 2007). However, it has been only since very recently that scholars have started to empirically test (parts of) social learning theory or social structure and social learning theory in relation to terrorism, radicalisation or extremism (Becker, 2019; Carson & James, 2019; Holt, Freilich, Chermak, & LaFree, 2018; LaFree et al., 2018; Mills, Freilich, Chermak, Holt, & LaFree, 2019; Pauwels & Schils, 2016).
Becker (2019), LaFree et al. (2018), and Pauwels and Schils (2016) are three examples of articles which have done tests using quantitative data. While this gives great insight into the correlations between certain variables and allows for “large-N” tests, it does come with certain trade-offs. LaFree et al. (2018), for example, only look at radical peers and family as a proxy for differential association. They do not test the other key elements of social learning theory, imitation, definitions, and differential reinforcement. More broadly, all three studies have a methodological issue which is using proxies for concepts from social learning theory. When looking at the operationalisation of differential association all three studies test this by looking at the question if these individuals had any radical peers or family members. When this was the case, they coded this in a dichotomous way with someone having either one or more radical peers or family members or having no radical peers or family members at all. This is understandable from a methodological point of view since this makes it easier to quantitatively test certain hypotheses, but this does not capture the true range of options and mechanisms explained by the theory. Just having a radical peer or family member does not say much about the nature of the relationship. Akers (2009) explicitly states that the nature, the intensity, and the duration of association with others influences the way in which definitions are shaped and differential reinforcement takes place. Similarly, when looking at dichotomous variables in a cross-sectional design it is difficult to capture the passage of time during which a person becomes an extremist. Since this process is often understood as a gradual process that does not happen instantaneous (van den Bos, Loseman, & Doosje, 2009), it adds to our understanding to not only look at it from a quantitative point of view but also from a qualitative point of view. This allows for a better description and deeper understanding of real-life events such as individuals’ life-courses and behaviour (Decorte, 2016). Especially when looking at the
11 mechanisms described by SSSL these nuances get lost by using quantitative research methods. When looking at extremism, case studies allow us to “capture and explain change in viewpoints which can be conceptualized as a developing process rather than a stable state of thinking and action” (Pennington, 2015, p. 904).
Having said this, there are some case study approaches to testing SLT or SSSL (Carson & James, 2019; Holt et al., 2018; Mills et al., 2019). However, these all have some form of methodological or theoretical shortcomings as well. Holt et al. (2018) and Mills et al. (2019) only look at social learning theory and only use the role of peers and family as their main focus. While both studies help in our understanding of social learning theory they therefore do not test the assumptions made by SSSL or test for the other elements of SLT. Similarly, while Carson & James (2019) do take SSST into account, they take a limited approach in selecting their case studies by only looking at one animal rights and one environmental activists case from the US. Except for Pauwels and Schils (2016) who look at Belgium adolescents, all mentioned studies on either SLT or SSSL look only at cases from within the US obtained from the same database. This thesis tries to combine all these factors to both look at SLT and SSSL, look at cases from inside and outside the US, use qualitative data to add onto the knowledge gained through quantitative tests and look at a higher amount of case studies. Overall, this will broaden our current understanding of social structure and social learning theory as a means of explaining why people become extremists, carry out their attack and subsequently help in building better and more effective policies to counter this process.
Explanation of Social Learning Theory (SLT)
Akers’ social learning theory incorporates four key elements: (1) differential association, (2) definitions, (3) imitation, and (4) differential reinforcement (Lilly, Cullen, & Ball, 2015). Crime, as well as normal conforming behaviour, is seen as “learned behaviour through social interaction with others” (Akers & Sellers, 2004, p. 89). This core idea stresses the importance of differential association which influences the definitions a person might have regarding certain behaviour (Lilly et al., 2015).
Through differential association groups influence individuals’ definitions and behaviour (Akers & Silverman, 2014). If a group holds definitions supportive of certain actions, the individual is more likely to perform these actions. Akers makes a distinction between two dimensions of differential association: interactional and normative. Interactional differential association involves either direct (e.g. friends spoken to at school) or indirect identification with more distant groups (role models on the internet, a political party, or
12 extremist groups). The internet can play a role in blurring this distinction between direct and indirect association (Lilly et al., 2015). The normative dimension is the “different patterns of norms and values to which an individual is exposed through this association” (Akers & Sellers, 2004, p. 85).
The second key concept, definitions, is a major factor in causing criminal behaviour. For the majority of their actions, people act based on their cognitions or how they define situations and what actions they believe are appropriate (Lily et al., 2015). These internal definitions can be general, broadly counting for many actions, or only applicable to certain specific behaviour. This can result in defining crime as a whole as wrong but punching someone in the face after they disrespected you as the right thing to do (Akers & Sellers, 2004, p. 86). Another example which is often seen in Moroccan culture is the notion of honour killings. Most Moroccans define killing someone as morally wrong, but sometimes is it necessary to protect the honour of the family in which case it is permissible or even obligatory. The more positive definitions someone holds of a certain action in relation to negative definitions, the greater the chances are that the act will be committed. Definitions can also be neutralizing, defining the act as morally wrong but justified under certain circumstances (Akers & Sellers, 2004, p. 86). The amount of exposure to certain acts highly influence the type of definitions someone holds. The third concept, imitations, is Akers’ way of introducing a way in which people start committing certain crimes (Lily et al., 2015, p. 60). Shortly put, people engage in behaviour after they observe similar behaviour in others. Individuals can deduce possible outcomes based on the behaviour of other people and the consequences of that behaviour for those other people. Imitation especially plays a role in when individuals admire the other person (Akers & Sellers, 2004). Whether the behaviour shown by other will be imitated therefore not only depends on the behaviour itself or the consequences of the behaviour, but also by the relationship between the individual and the other. Similar to how differential association can take place with more distant groups such as role models on the internet, imitation also is not limited to the physical and direct dimension (Akers & Silverman, 2014).
The last concept, differential social reinforcement, is introduced as the determining factor in whether behaviour is repeated or not (Lily et al., 2015, p. 60). Akers defined this concept as “the balance of anticipated or actual rewards and punishments that follow or are consequences of behaviour” (Akers & Sellers, 2004, p. 87). The stronger certain consequences are, the more likely someone is to refrain from or repeat certain behaviour. These consequences can be non-social, such as the physical feeling after taking drugs, but they are mostly social,
13 involving other people (Akers & Sellers, 2004, p. 88). They can come from the belief of fulfilling ideological, political, religious, or other goals (Akers & Silverman, 2014).
Literature on SLT
For the last five decades social learning theory has been one of the major theories used in the school of criminology (Akers & Silverman, 2014). It has not only stood the test of time on its own but has also been used as part of different integrated theories of crime, including rational choice (Pratt, Cullen, Blevins, Daigle, & Madensen, 2006), social disorganisation (Lowenkamp, Cullen, & Pratt, 2003), life-course theories (Wiesner, Capaldi, & Patterson, 2003), and neuropsychological theories of crime (Beaver, Wright, & DeLisi, 2008). It has become a standard to include social learning variables as control variables when testing for other criminological theories (Pratt et al., 2010).
In their meta-analysis consisting of 133 empirical studies between 1974 and 2003, Pratt et al. (2010) have found an overall modest to strong support for social learning variables. Their first conclusion is that the empirical support for social learning theory can compete with other criminological theories which have been subjected to meta-analysis such as self-control, rational choice, social bonds and deterrence theory (Pratt, et al., 2010). Despite varying effect sizes depending on how key concepts are measured and overall research design, “the variables specified by social learning theory tend to be strong, yet not invariant, predictors of criminal and deviant behaviour” (Pratt, et al., 2010, p. 788). The second conclusion is that some of the different elements of social learning theory have been tested substantially more often than others. Where differential association and definitions have been tested extensively and are often included as control variables for other studies, the differential reinforcement and imitation elements have received less attention in empirical studies (Pratt, et al., 2010). Thirdly, Pratt et al. (2010) conclude that these same variables, differential reinforcement, and imitation, have received less empirical support compared to the other two variables. Effect sizes were generally weak or modest at best, whereas differential association and definitions are consistently strong predictors of criminal behaviour. They argue that this could be because of a lack of proper operationalisation of these constructs or because they have simply been tested fewer times. Overall, Pratt et al. (2010) conclude that despite these shortcomings, the results of their meta-analysis “clearly show that it deserves its status as one of the core perspectives in criminology” (Pratt, et al., 2010, p. 790).
As stated above, there are some recent examples of studies testing some or all social learning variables in relation to the concept of extremism (Becker, 2019; Carson & James,
14 2019; Holt et al., 2018; LaFree et al., 2018; Mills et al., 2019; Pauwels & Schils, 2016). The first test of social learning theory in relation to extremism was done by Pauwels and Schils (2016). They are the only authors who do not make use of the PIRUS3 database to test their assumptions. Instead they make use of survey data among Belgian high school students between 16 and 24 years old. Using this data, they perform a binary logistic regression to test if exposure to online violent extremism is related to self-reported political violence. They also test for several criminological theories to explain possible differences. The authors find that there is a relation between exposure to violent extremism online and light forms of self-reported political violence aimed at property. They interpret this as evidence for social learning variables. Similarly, differential association with peers holding racist attitudes which takes place offline is strongly related with acts of political violence among the participants. Pauwels and Schils also find (partial) evidence for rival theories such as strain, social bonds, self-control, but conclude that social learning variables are central to a model explaining political violence among high school students.
LaFree et al. (2018) also make use of quantitative methods. They use a multivariate logistic regression analysis to test variables obtained from several criminological theories in relation with violent and non-violent extremist. Just like the following five studies LaFree et al. (2018) obtain their data from the PIRUS database containing 1757 US extremists. Regarding social learning variables they limit themselves to the proxy variables radical peers and radical family members. They find support for both variables with radical peers as being the strongest predictor of violence among all tested independent variables. The variable radical family members was significant at the bivariate level, but lost significance when controlling for other variables in the multivariate model.
Becker (2019), making use of the PIRUS database, carries out a bivariate chi-square association test between social learning and social control variables and engagement in violent extremism. He makes use of seven dichotomous proxy variables that together are used to test for the concepts of differential association, definitions, imitation, and differential reinforcement. He does not make a distinction between the different elements of social learning but uses the seven proxy variables to look at the overall social learning concept. The author
3 The Profiles of Individual Radicalization in the United States database includes individual level information on
the attributes, radicalization process and backgrounds of 2,200 (violent) extremists who adhere to a far right, far left, single issue or Islamist ideology in the US covering 1948-2018 (START, 2018). The database is hosted by the National Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland (MD), United States.
15 finds mixed support for the hypothesis that social learning variables can explain the difference between violent and non-violent extremists. The variables gang membership and group membership are positively correlated with violent extremism among radicalized individuals. The author finds weak overall support for social control variables.
Holt et al. (2018) also make use of the PIRUS database to select their cases. But instead of using quantitative analysis they make use of a case study approach consisting of two far right and two jihadist perpetrators, with one engaged in non-violent activity and one engaged in violent activities within each ideological group. They test if social control and social learning variables can explain the difference in radicalization between both ideological groups. The authors make use of publicly available documents such as court documents, newspaper articles, website, and watch group articles to create biographies of the four cases. In contrast to Becker (2019), Holt and colleagues find more support for social control variables than for social learning variables. They find some evidence for definitions of peers in support of radical ideologies but did not test for imitation or differential reinforcement. Another explanation for this difference could come from the fact that they only looked at peer association as a proxy for differential association whereas Becker (2019) made use of multiple variables. The authors also stress the important role the internet played in three out of four cases and that it would be useful for future research to look into this.
Two other studies that use case study approaches are Carson and James (2019) and Mills et al. (2019). Carson and James (2019) test social learning and social structure variables in one animal rights and one environmental activists case from the US. They select their two cases from the PIRUS database as two “average” cases about whom a lot of information is available. The authors gather data from publicly available documents such as documentaries, newspaper articles, and court documents to document a life-course overview of their two subjects. Carson and James (2019) conclude that in both cases social learning variables played a big role in the radicalization process. However, in one case there was more direct association with peers and in the other case there was more indirect association with reference groups online.
Mills et al. (2019) take the recommendations from Holt et al. (2018) and look at how social learning and social control variables can explain the difference between off- and online pathways to terrorist violence and hate crime. They select four different cases from the PIRUS database consisting of two jihadists and two far-right perpetrators with one of each group engaging in hate crime and one engaging in anti-government terrorist activities. Overall, they conclude that in all four cases there were elements of both social control and social learning
16 variables. Regarding the off- and online pathways to extremist violence they find that the internet played a larger role for the two far-right cases than for the two jihadist cases.
Overall, these six studies show differing results. Becker (2019), Carson and James (2019) and Mills et al. (2019) show strong support for social learning variables in their studies. On the other hand, LaFree et al. (2018) and Pauwels and Schils (2016) show moderate support whereas Holt et al. (2018) find weak support and find stronger support for the rival variables of social control. These results show similarities to the meta-analysis of Pratt et al. (2010) who found moderate to strong support for most social learning variables but noticed similar differences in support between studies and between variables that were tested. This goes to show that although some testing has been done regarding SLT, results remain inconclusive and depend on the methodological soundness of the research.
Explanation of Social Structure and Social Learning Theory (SSSL)
Akers (1998) realised that a micro explanation of crime alone was not enough to explain all forms of criminal behaviour. He therefore built upon his social learning theory by extending it to the meso and macro domain introducing social structure. Structural theories of crime look at the larger context of society and its characteristics which influence crime rates in certain areas. Micro-level theories on the other hand, explain the processes by which individuals’ criminal behaviour is influenced (Akers, 2009). Despite acknowledging the standardised methodological approach of separating structural from behavioural elements in order to develop and test theories, Akers (1998) stressed the importance of an integrated theory in which social structural situations influence individual behaviour.
The variables differential association, definitions, imitation, and differential reinforcement described in social learning theory, mediate social structure’s effect on individual behaviour and therefore, by extension crime rates (Akers, 2009). The social learning variables have a direct impact on an individual’s behaviour whereas social structure provides the environment that impacts behaviour through social the learning process (Akers & Sellers, 2004). The social structure should be seen as the context in which social learning takes place. Crime rates are influenced by the extent to which structural variables provide opportunity for social learning to take place.
In the SSSL model (see figure 1), four dimensions can be distinguished: Differential Social Organization, Differential Location, Theoretically Defined Structural Causes, and Differential Social Location. The first dimension, Differential Social Organisation, consists of structural variables on a community or society level that are known correlates of crime (Akers,
17 2009). These include the community size, population density, age, sex or racial composition, and other geographical variables (Akers & Sellers, 2011). These correlates of crime can be seen apart from theoretical explanations of crime. There are several known and unknown characteristics of the culture of society, communities, neighbourhoods, or groups that have distinctively higher crime rates compared to others. The differences between these social structures make up the first dimension of Akers’ conceptualization.
The second dimension, Differential Location in the Social Structure, are the individual sociodemographic and socioeconomic variables that are known predictors of crime in societies, communities, and regions (Akers, 2009). There are known or unknown variables such as class, race/ethnicity, gender, marital status, age, religion, occupation, and other factors of social differentiation. Akers argues that these individual characteristics locate where people stand within the social structure. These variables can be seen as the root causes of crime, but more often the relationship between them and differentiating levels of crime is seen as the thing to be explained. We know for example that young men are more likely to commit crimes compared to women in the same age range (van Dijk, Huisman, & Nieuwbeerta, 2014). The question for criminological theorists is to find a logical explanation for this phenomenon. One might for example think of a theory which reflects that gender-role expectations are different for men and women of that age. Akers sees the combined characteristics of individuals as “direct indicators of the differential location of groups or categories of individuals in the social structure” (Akers, 2009, p. 333).
The third dimension, Social Disorganization and Conflict, are what Akers calls theoretically defined structural causes (Akers, 2009). He recognises that other criminological theorists have come up with structural causes of crime. These are for example social disorganization theory, anomie and strain theory and conflict theory, but as for the other variables in the other dimensions, this list is by no means exhaustive. These structural causes are not simple predictors of crime, such as the variables listed under dimension one and two, but these are the explanations for why these variables cause differences in crime rates between parts of society (Akers, 2009). The idea that these variables could influence crime rates originates from Sutherland’s notion that crime is caused by social organization for or against crime (Sutherland, 1947). Differential social organisation can create opportunities and environments that influence micro-level social learning variables. A social system is socially organized against crime if there is a strong cohesion between its members, if there is a consensus on the norms and values that are upheld by members of the community, and if there
18 is little disruptive conflict (Akers, 2009). Conversely, if the opposite is true a community is socially organised for crime.
The fourth and last dimension of social structure is Differential Social Location in Primary, Secondary, and Reference Groups. These are the groups that are referred to in the differential association concept in the social learning concepts. They form the agents of informal and semiformal social control and socialisation (Akers, 2009). Individuals relate to these groups, and they provide the social environments, opportunities and situations that discourage or promote his or her conforming or criminal behaviour. These groups provide the social context in which differential association, the forming of definitions, imitation and differential social reinforcement takes place. These groups are not limited to delinquent peers, but can also include family, reference groups online, political parties or non-delinquent peers.
Figure 1. The social structure social learning model. Adapted from Akers, Ronald. L. (2009). Social Learning and Social Structure. New York: Routledge.
Literature on SSSL
To this day there are still not many studies that use the full social learning social structure theory as its conceptual framework. Similarly, most studies look at US student survey data to test their assumptions. Capece and Lanza-Kaduce (2013), Lee, Akers, and Borg (2004), Schaefer, Vito, Marcum, Higgins, and Ricketts (2015), Whaley, Smith, and Hayes-Smith (2011) all make use of survey data gathered among US high school students. They find
19 moderate to strong support for SSSL theory in explaining respectively, binge drinking, substance abuse, heroin use and drug and alcohol use. With the exception of Capece and Lanza-Kaduce (2013), they find results which are consistent with Akers’ hypothesis that social learning variables mediate social structure variables.
Bellair, Roscigno, and McNulty (2003) and Cooper and Klein (2018) look at broader forms of (youth) deviance. Cooper and Klein (2018) make use of US college student data to explain youth deviance. They find mixed support for SSSL. Bellair et al. (2013) additionally also make use of adult survey data to see if labour market opportunities correlate with adolescent delinquency seeing labour market opportunities as a proxy for social structure variables. They find that service sector employment opportunity directly decreases the chances of violent delinquency, thus finding support for SSSL.
Not all authors have limited themselves to use US data, however. Dim and Elabor-Idemudia (2018) use the Nigerian Demographic and Health Survey of 2013 to see if SSSL theory can explain the severity of physical intimate violence against women in Nigeria. Their results are consistent with social structure variables and show that the sociocultural context influences the social realities in which the average Nigerian women lives. Kim, Akers, & Yun (2013) use survey data among South Korean adolescents to examine if SSSL variables can explain their alcohol use. The results show that the social learning variables mediate the effects of the social structure components of alcohol use. These two studies show that the theory holds explanatory value not only in the US but also in non-Western societies when trying to explain adolescent alcohol usage and violence against women.
Morris and Higgins (2010) and Hawdon, Bernatzky, and Costello (2019) have taken a more modern approach to see if SSSL can explain more recent forms of online crimes. Morris and Higgins (2010) look at survey data to see if SSSL variables can explain digital piracy among US students. The results show moderate support for the explanatory value of SSSL variables in explaining digital piracy. The authors stress the importance of these findings in relation to the rise of cyber (related) crimes in the world. Hawdon et al. (2019) make use of a quantitative analysis to look at the relation between the risk of exposure to violence-advocating materials and online behaviour of US college students. They find that there is a positive correlation between using social media platforms and the virtual spaces individuals frequent and exposure to extremist materials. They explain this correlation by using social learning and social structure theory arguing that people who become exposed through online media sympathize with it collectively and learn and change their position regarding violence through interactions.
20 The one study that has looked into the relation between social structure and social learning variables and extremism is from Carson and James (2019) which has also been discussed in the literature on SLT. Carson and James (2019) make use of a case study approach to test both SSSL variables on one animal rights and one environmental activists case from the US. They find little support for social structure elements in the first case but much support for social learning variables. In the second case they find many social structure elements as well as social learning elements. Overall, they conclude that SSSL offers promise in explaining all forms of crime, including terrorism but call for further testing on a wider range of case studies including perpetrators that adhere to more violent ideologies.
Overall, the previous section showed that studies regarding SLT and extremism show differing results. Three studies showed strong support, two studies found moderate support and one study found no support for social learning variables. This current section on the literature regarding SSSL also showed mixed results. Overall, studies looking at regular types of crime have shown moderate to strong support for SSSL variables. However, the one study which has been done looking at extremist behaviour found support for SSSL variables in only one of their two case studies. This shows that not only regarding social learning theory, but especially regarding social structure and social learning theory, further testing needs to be done.
21 Methodology
This thesis aims to fill the clear gaps in the current literature regarding SSSL and extremism laid out in the previous chapter. Social structure social learning theory remains an understudied yet promising theory which could yield useful policy related insights. While there has been some research on social structure and or social learning theory, all studies so far have shown one or more shortcomings. Almost all studies have focused on testing only (a part of) social learning theory, ignoring the social structure elements which are crucial in Akers’ theory. Similarly, studies have only used data from extremist people in the US, not taking into account different cultures and countries. This thesis tries to overcome these shortcomings by using a framework that looks at both social structure and social learning elements, in multiple case studies, to fully try to comprehend to what extent SSSL theory can helps us understand why people become extremists and carry out their attacks.
To test the statements made by Akers (2009) that social structure and social learning theory can be used to explain extremist behaviour, this thesis makes use of a qualitative case study analysis. The aim is to test if the mechanisms laid out by Akers (2009) can help us explain peoples’ entry into extremism and their extremism actions. While quantitative empirical studies are perhaps better in terms of generalization of the results, there are certain benefits for using a qualitative case study approach. This approach does not have the limitation of having to use (dichotomous) proxies for intricate and difficult to measure concepts which SSSL theory is certainly rich of. This helps for example, in getting a better understanding of the nature of relationship between the case study object and his family and peers. Similarly, the process of becoming an extremist, which takes place during a certain time period, is difficult to measure in cross-sectional data analysis and is better suited for a case study approach. Using a qualitative method allows for a more comprehensive descriptive understanding of the process (Pennington, 2015).
Social structure and social learning theory consists of eight concepts which make up the main mechanism used to explain extremist behaviour. There is a distinction to be made between the social structural variables and the social learning variables. The social learning variables are differential association, definitions, imitation, and differential reinforcement (Lilly et al., 2015). Applied to extremist behaviour, the chance that a person will show extremist behaviour is increased when they differentially associate, directly or indirectly, with others who show extremist behaviour. They adopt definitions favourable to extremist behaviour through differential association and imitation and anticipate greater reward than punishment for their behaviour (Akers & Silverman, 2014).
22 Social structure variables consist of differential social organization, differential location, theoretically defined structural causes, and differential social location. Applied to extremist behaviour, Akers’ theory states that these structural variables, do not predict extremist behaviour itself, but when these variables are present in the community or on a personal level of the subject, a person is more likely to show social learning variables (Akers & Silverman, 2014). For example, when someone is aged between 17 and 21, lives in a bad neighbourhood in which problems of social disorganisation, conflict and strain play a role, his parents divorced at an early age and he has mental health problems, the theory does not predict he is more likely to show extremist behaviour, but instead the social learning variables are more likely to be present. In turn, these social learning variables directly influence the probability of him showing extremist behaviour.
Data
To fully test all aspects of SSSL theory, this thesis makes use of a purposeful case selection method, selecting five cases from the last decade. For the aim of this thesis it is necessary that cases are selected on the basis of the amount of data that is available on them. In order to test Akers’ social structure and social learning theory, the richer and more complete the information is on the cases the better our understanding of mechanisms and theoretical elements will be.
Secondly, this thesis focusses on extremists that are part of Castelli Gattinara and Pirro’s third group they describe in their framework. They see the far right as an umbrella concept in which far-right actors are defined as groups or individuals located on the “right end of the ideological left-right continuum” (Castelli Gattinara & Pirro, 2019, p. 450). The subjects described in the results section of this thesis all fall under this ‘umbrella’ definition of far-right actors as a subcultural group of individuals. This is done because Akers and Silverman (2014) specifically state that social learning theory should also be capable of explaining violent ideologically motivated crimes. Gottfredson and Hirschi (1990) on the other hand specifically exclude ideologically motivated offenders in their general theory of crime. They state that their self-control theory only applies to non-ideological, routine offenders, because ideologically motivated offenders do not suffer from the same deficiency in self-control levels. Perpetrators of non-ideological and ideological crimes do not overlap according to them. It would therefore be interesting to see if, other than self-control theory, social learning theory can indeed explain far-right extremism such as predicted by Akers and Silverman (2014).
23 Thirdly, the cases have been selected to be somewhat recent ranging from 2011 to 2017. This is done to make sure that enough time has passed to allow for a large amount of information to be available. Similarly, it makes sure that enough time has passed for legal cases to be concluded.
The list of cases itself has been compiled by using the Global Terrorism Database (LaFree & Dugan, 2007). This database contains over 190,000 terrorist attacks compiled by The University of Maryland’s National Consortium. From these 190,000 cases a selection of five cases has been made which will be used to answer the research question. To select these five cases only cases between 2011 and 2017 have been selected. Secondly, only cases from the regions “North-America” and “Western Europe” have been selected. This is done for linguistic reasons. Thirdly, the following perpetrator group names were selected to include a wide as possible array of far-right extremists: “Right-wing extremists”, “White supremacists/nationalists”, “Incel extremists”, “Anti-Muslim extremists, “Anti-Semitic extremists”, “Anti-Immigrant extremists”, “Neo Fascist extremists”, and “Neo Nazi extremists” (Castelli Gattinara & Pirro, 2019).
From here, cases with more than three casualties or high impact targets were selected. This was done because attacks with more casualties usually cause a larger amount of media attention (Kearns, Betus, & Lemieux, 2019). Lastly, all cases were removed in which the perpetrator died during the attack, both because of his own accord or because of police fire. This is done in order to allow the possibly to include court records into the analysis. There is one exception to this, which is the case of Elliot Rodger. Not only did he upload a 141-page manifesto in which he talks about his entire life-course, but he also uploaded multiple videos explaining what he was about to do and why. Furthermore, the comprehensive investigation report on the shooting is available online and provides another insight into his life and entry into extremism. These three things make that, for the purpose of this thesis, enough information can be found on Rodger without needing additional court documentation. From the remaining eleven cases, five have been selected to cover as many countries as possible. The final cases are Anders Breivik (Norway), Elliot Rodger (United States), Dylann Roof (United States), Thomas Alexander Mair (United Kingdom), and Alexandre Bissonnette (Canada).
In order to create life-story overviews and provide information regarding SSSL elements this thesis relies on publicly available documents. Newspaper articles, court and investigation documents, psychiatric evaluations, blogs, scholarly work, books and the perpetrator’s own publications such as manifestos and social media accounts were used to gather information on the subjects. This method is consistent with previous scholars who have
24 used publicly available documents to do case study analyses on radicalisation (Carson & James, 2019; Holt et al., 2018; Mills et al., 2019). To find these documents the same procedures will be followed as used in the compiling of the US Extremist Crime Database (Freilich, Cherman, Belli, Gruenewald, & Parkin, 2014). This consisted of using official government publications and searching for suspects and victims in over 30 web search engines. For this thesis, this list has been limited to relevant search engines and has been expanded with DuckDuckGO, The Internet Archive, Lexis Nexis, Factiva, and pdf-based search engines.
To create a full understanding of why these subjects became extremists and to apply all SSSL elements a case study template has been created (see appendix 1). This template is based on Holt et al. (2018)’s work on social learning and social control. It has been modified to include all SSSL elements as depicted in figure 1. Additionally, conceptualisations from other scholars studying SSSL theory have been used to add onto the template (Becker, 2019; Carson & James, 2019). All documents used to find information on the subjects have been coded using this framework. The outcome of this coding has been used to describe the subjects’ life-course overview, his entry into extremism, and describe to what extent the different variables of SSSL theory are applicable to the case.
Research quality and limitations
The chosen research method does come with certain limitations. First off all, issues regarding case study research are well documented. For critiques of case-study research see (King, Keohane, & Verba, 1994; Geddes, 2003); for support see (Brady & Collier, 2004; George & Bennett, 2005), with the most important limitation being the external validity. Conclusions that are drawn based on qualitative case study analysis do give great insight into the cases themselves but cannot be used to externalise conclusions to the bigger population. The external validity of this thesis and most qualitative case studies in general is therefore quite limited (Kleemans, Korf, & Staring, 2008).
A second limitation of this thesis is that it primarily relies on publicly available data. For example, it does not make use of interviews with the subjects themselves, which is commonly used in other life-course based study designs. This has three implications. First, not all information regarding all aspects of their life-course will be available as well as information on how they entered into extremism. This thesis is reliant on information gathered by others who could have had a different purpose than the purpose of this study. Compared to asking subjects the questions themselves this thesis relies on others to do it for us. Secondly, the information has already been processed by others and crucial information could therefore have
25 been interpreted differently or discarded. Thirdly, this study primarily relies on publications in the media for one of the five cases. This means that a certain media bias should be taken into account when interpreting the results. Attacks that have caused more casualties for example cause more publications which causes there to be more information available on certain subjects (Kearns et al., 2019).
Another limitation of the chosen method is that this thesis only looks at “successful” attacks. Mainly for practical reasons, there is often very little information about foiled attacks. This can either be because the media did not report on it or the intelligence agencies or police did not publicly announce them (Carlsson, et al., 2019). There could be a potential bias in the subjects we study when only looking at successful attacks. This is similar to how criminology as a field relies primarily on crimes which got reported. There could be a structural difference between offenders who get caught for example, and offenders who do not get caught. One of the ways in which criminologists deal with this potential bias is holding confidential questionnaires among the general public to see how many people commit crimes without getting caught. However, this solution is not as suitable when trying to identify potential extremist behaviour. Trying to find cases in which an attack was not committed is not only difficult, it is also time consuming and beyond the scope of this thesis.
All in all, these limitations imply a certain bias in the findings of this thesis and the conclusions should be read in the light of these limitations. However, this does not mean the results of this thesis are meaningless. The aim of this thesis is not to prove any causal relations or generalize the findings to a broader population, but to generalise theoretically. The results should provide a clear picture of the extent to which Akers’ social learning theory can explain a specific form of nonconforming behaviour. The units of analysis are not chosen at random but are an example of theoretical sampling (Boeije et al., 2009). Theories need to constantly be tested by using new respondents to verify, adjust or falsify the theory (Kleemans et al., 2008). Furthermore, all subjects in this thesis have been selected by using the Global Terrorism Database in order to prevent any media bias in the case selection. Similarly, in most cases official court documents, police investigation records and psychiatric evaluations have been used to increase the accuracy of the data gathering process.
26 Results
In the following results section all five case studies will be discussed. The analysis has been performed using the case study template (see appendix 1). First of all, in all the five cases a life course overview will be given from birth till the attack taking place. Secondly a short summary will be given of every subjects’ entry into extremism. Furthermore, in all five cases, all eight social learning and social structure variables will be discussed to the extend to which information could be found on them. In addition, every subject’s mental health status prior and during the attack will shortly be described.
Anders Breivik
Life course overview: Breivik was born on 13-02-1979. His father was a recently divorced diplomate and his mother a nurse and a mother of a four-year-old daughter (Seierstad, 2015). His mother was depressed and left his father six months after he was born. When Breivik was two, he was put in foster care for two weekends per month (Seierstad, 2015). His mother said she could not take care of him and his sister and wanted to smash their heads against each other, according to child protection services (Seierstad, 2015). This lasted only for six months because his mother did not think the foster home was good enough for him and asked for a foster home that could take full care of Breivik. When he was four, his mother changed her mind and after the Child Protection initially planned to claim a forced out-of-home placement, a combination of circumstances led to this being abandoned (Olsen, 2016).
During primary school, things were going relatively well with Breivik. He made some friends, joined the boy scouts, played soccer with classmates, and did his best at school. He did occasionally bully other children and was no longer allowed to play with animals because he hurt them (Seierstad, 2015). However, he also showed his good side by helping other children who were bullied.
During high school, Breivik ended up in the taggers (graffiti) circuit (Seierstad, 2015). He became quite popular but remained a bit of an outsider who was never truly respected by the others. At some point the other taggers got fed up with him and kicked him out of the group. From then on, he started to get bullied at school and lost his friends. Because he got arrested twice by the police for vandalism, his father also broke off all ties with him when Breivik was fifteen (Seierstad, 2015). After this he changed schools and made friends with a group of rich kids. Just before the end of the last year however, Breivik dropped out of school. He decided to start working full time for an advertising company (Seierstad, 2015).
27 At the age of 18, he joined the Fremskrittsparti, a Norwegian conservative-liberal political party (Seierstad, 2015). A few years later, he and four others set up the board of the local youth party. In this youth party he became active, but he had a hard time climbing to the top. Once again, he tried to fit in, but others saw him as too meddlesome and subordinate to become a real politician (Seierstad, 2015). When he is 24, he stopped being active within the party and went his own way. After this, Breivik started selling fake diplomas with which he earned a large amount of money (Seierstad, 2015). He sold them on the internet for hundreds of crowns each. Not only did he brag about this to his friends from the last high school he went to, he also called his father who he has not spoken to in nine years to tell him how great he was doing. At the end of the call his father promised to call back soon but never did (Seierstad, 2015).
When Breivik was 27, he stopped with his business in false diploma’s and started to focus on stock trading. This went quite poorly however, which caused him to lose a lot of money and forced him to live with his mother again (Seierstad, 2015). At his mother’s he primarily played the online strategy game World of Warcraft. He took little breaks, barely ate, got mad at his mother for interrupting him and barely slept. The next two years of his life were spent behind his screen in his moms house (Seierstad, 2015).
When Breivik saw his friends again, he was a changed man. No longer did he spend his time gaming. Instead he was looking at websites such as Stormfront, Gates of Vienna and Jihad Watch (Seierstad, 2015). He developed a strong hatred against the Islam and was convinced Muslims wanted to overthrow the white way of life in Norway (Seierstad, 2015). He also started working on his book ‘2083 – A European Declaration of Independence’. In this manifesto he wrote his primary motive for his terrorist attacks (Berwick, 2011). To prepare for his attack he bought a farm so he could work on creating his bomb without being disturbed. On the 22nd of July 2011, his preparations were complete, and he started his attack.
Entry into extremism: After Breivik had spent 2 years playing World of Warcraft he started browsing right-extremist websites (Townsend & Traynor, 2011). He took his previous points of interest he still had from his time as a member of the Fremskrittsparti and posted these on these websites. He started off by reading what others had published and posted messages saying he was working on his own book (Seierstad, 2015). He tried to come into contact with one of his heroes called “Fjordman” who was very active in the community. When he did not get a response he tried again, but this time with a more critical tone. He said that Fjordman’s ideas were not radical enough. He said that stopping immigration with democratic means would not be enough and that undemocratic means were necessary (Seierstad, 2015, pp. 267-268).
28 Not only did countries need to stop immigration, they also needed to deport immigrants who had already invaded Norway. Despite this more radical tone, he did not get the attention from his heroes he wanted. Therefore, he convinced himself he needed to take action himself (Seierstad, 2015). He then proceeded to work on gathering materials for his attack and bought a remote farm to have a place to work without being disturbed. He ended up using lots of things he read on the extremist websites for his own book (Seierstad, 2015).
Differential association: During his high school years Breivik made friends with whom he would stay in contact with for a couple of years. He met them offline and they provided conforming norms and values to Breivik. They were well educated. Most of them ended up as lawyers, businessmen, financial bankers or had other well-paid jobs (Seierstad, 2015). However, the duration, intensity and frequency of these interactions was low compared to the other groups Breivik came into contact with. While he was gaming, he spent two years talking to other people online for up to 18 hours a day (Seierstad, 2015). Nonetheless, they primarily discussed the game they were playing and did not have discussions related to conforming or nonconforming behaviour. The last group Breivik had interactions with was the group he met online on extremist websites. Again, he spent a lot of time on these websites which caused the frequency, intensity, and duration of these interactions to be high. He also had some discussions with his mother regarding his worldviews (Seierstad, 2015). However, she did not provide him with conforming definitions, but instead accepted his worldview as her own and even spread it further to her own friends.
Definitions: On these websites Breivik came into contact with positive definitions of his crime. The norms and values these websites promoted were nonconforming. They openly incited violence and were against immigrants and the ruling parties which they blamed for allowing the immigrants to take over Norway (Seierstad, 2015). One of the messages on these forums was that a car-bomb, placed on the exact spot Breivik would park his van during the attack, would cause the biggest explosion and take down the entire building (Seierstad, 2015). After being exposed to the extremist content and peers on these websites Breivik’s behaviour changed. He quit playing videogames and spent all his time reading forums, writing his book, and preparing for his attack (Seierstad, 2015).
Imitation: Breivik did not directly see others commit similar crimes. He did however, read much on the Unabomber, copying parts of his manifesto into his own, changing only the words leftism to cultural Marxism.
Differential reinforcement: Breivik knew he was going to prison if he continued with his plan. This was a risk he was willing to take. He was afraid he was going to be tortured in
