Abstract
Extremism and radicalization towards violence are urgent topics in many countries. Numerous research projects are carried out, of which many focus on risk factors only. In contrast, this article contains a systematic review of the rare international research on protective factors. After screening more than 2,000 documents, we found 17 reports containing 21 analyses that specifically addressed potential protective effects and provided quantitative data. Most studies addressed religious/ethnic extremism; far-right, far-left, and mixed forms were less frequent. Thirty different protective factors showed significant effects. Many were found in single analyses, but there were various replicated factors such as self-control, adherence to law, acceptance of police legitimacy, illness, positive parenting behavior, non-violent significant others, good school achievement, non-violent peers, contact to foreigners, and a basic attachment to society. Most findings are similar to what we know from more general research on youth violence, but there are also some protective factors that seem to be more specific, particularly with regard to religious/ethnic extremism. In conclusion, it is suggested to relate the topic of extremism and violent radicalization more strongly with other fields of developmental and life course criminology. For further progress on this path, more research on protective factors and integrated theoretical concepts are needed. This will also contribute to effective prevention.
Politically and religiously motivated extremism and radicalization towards violence or terrorism are currently urgent topics in many countries. Numerous research projects and prevention programs on these topics have been launched. However, the sound empirical basis in these fields is still rather thin, which is a core problem of systematic evidence-based approaches to prevention and control (Jensen et al., 2016; LaFree & Freilich, 2018; Sageman, 2014; Silke, 2001). This situation is due to a number of issues that can only be briefly mentioned here: There is no generally used exact terminology. Extremism and radicalization refer to more or less gradual differences in attitudes and/or behavior (Bartlett & Miller, 2012; Kruglanski, Chen, Dechesne, Fisman, & Orehek, 2009). Approximately, extremism implies a verbal or active opposition to basic values in a society such as democracy, equality, liberty, rule of law, and tolerance for the faiths and beliefs of others. Some studies simply define it by relatively high scores in attitude questionnaires (Baier, Manzoni, & Bergmann, 2016), although ranking at an upper percentile may not indicate serious extremist behavior. Radicalization is a process by which a person adopts beliefs that justify the use of violence for social and political change (Doosje et al., 2016; Maskaliūnaitė, 2015). The ideologies can be far-right, far-left, religiously motivated, nationalist/separatist, or focus on special issues such as animal protection or anti-abortion (Doosje et al., 2016). There may be overlaps between these orientations. For example, depending on the social and political context, the developmental processes of different forms of extremism (e.g., Islamist and right-wing) seem to be related (Ebner, 2017).
Although various phases of radicalization have been proposed (e.g., Moghaddam, 2005; Silber & Bhatt, 2007), the respective processes that may begin with ideological interest and proceed over group affiliation to violent action are neither deterministic nor uniform (Borum, 2011; Jensen et al., 2016). Whereas the influence of group membership often plays a core role, there are also cases of so-called ‘self-radicalization’ and ‘lone wolf’ terrorism (Meloy & Genzman, 2016). Due to the diversity of phenomena and different disciplinary backgrounds of researchers, there is no widely agreed theory of extremism, radicalization, and terrorism. For example, with regard to terrorism, Borum (2011) mentions influences on the following levels: individual, group, network, organization, mass movement, socio-cultural context, and international/interstate contexts. Nasser-Eddine, Garnham, Agostino, and Caluya (2011) differentiate between rational choice, structural or societal theory, relative deprivation, social movement theory, and psychological theories. From a psychological perspective, social learning, social bonding, experience of deprivation, group dynamics, identity formation, intolerance of ambiguity, and mental health issues seem to be particularly relevant. However, the empirical bases are not very strong. Radicalization is often not directly investigated in its early stages (Rahimullah, Larmar, & Abdalla, 2013) and most studies are cross-sectional or retrospective case analyses (Scarcella, Page, & Furtado, 2016).
Although there is no comprehensive theory on extremism, radicalization, and terrorism, practice has to cope with challenges in risk assessment and management. Various risk assessment instruments have been developed, for example, VERA-2R, ERG-22+, TRAP-18, MLG, IVP, RAT, and RADAR-ItE (for overviews see King, Bender, & Lösel, 2018; Rettenberger, 2016; Scarcella et al., 2016). The risk assessment instruments mainly apply a structured clinical judgment approach that is common in other areas of violence risk prediction. Depending on the specific focus of radicalization and on institutional contexts, the measures vary (e.g., screening of vulnerable individuals in the general population vs. assessing dangerousness in already convicted offenders). Some measures differentiate between distal characteristics and proximal alarming indicators, or ‘red flags’; for example, IVP (Cole, Alison, Cole, & Alison, 2010) or TRAP-18 (Meloy, Roshdi, Glaz-Ocik, & Hoffmann, 2015). Specific measures of secret services or police are not published for plausible reasons.
In spite of various differences many items of the risk assessment instruments are similar in their content. Due to low base rates of the outcome criteria, partial unwillingness of target persons to cooperate and a lack of large-scale prospective studies, risk assessments cannot yet claim proven validity (Scarcella et al., 2016). Similar to other topics of violence, an accumulation of risk factors (Lösel & Bliesener, 2003) seems to be most indicative (Jensen et al., 2016). However, one must bear in mind that even when 100 percent of violent extremists have a specific pattern of accumulated risk factors (LaFree, Jensen, James, & Safer-Lichtenstein, 2018), the vast majority of people with such characteristics are neither extremist nor carry out violent acts. This problem of a large number of false positives led to concerns about problems of stigmatization (Royal College of Psychiatrists, 2016; Sarma, 2017).
Against this background, it is necessary to widen the perspective from risk factors to protective factors and processes. Considering protective factors can increase the validity of risk assessment and make it more sensitive for change. However, only some instruments contain at least a few items on protective factors (e.g., VERA-2R; Pressman, Duits, Rinne, & Flockton, 2016). A stronger focus on protective mechanisms is also relevant for understanding radicalization processes. For example, Doosje et al. (2016) refer to “shields of resilience” that interrupt the development from extremist orientation to membership in radical groups and to violent action (or contribute to de-radicalization). Unfortunately, there is much less research on protective factors against extremism and radicalization than on risk factors. This situation is similar to the broader literature on violence of young people where research on protective factors is also less frequent (Lösel & Farrington, 2012). Conceptual and methodological issues are more difficult because protective factors are not simply the other “side of the coin” of a risk factor but require more differentiated research methods (Lösel & Bender, 2003, 2017). For example, one may need to analyze curvilinear relations, study buffering effects in interactions, and apply hierarchical regressions (Loeber & Farrington, 2012; Lösel & Farrington, 2012).
Studies on protective factors have often been carried out within the context of resilience research. For conceptual and methodological issues in this field, see, for example, Luthar, Cicchetti, and Becker (2000), Lösel and Bender (2003), or Masten (2016). Resilience refers to a “reduced vulnerability to environmental risk experiences, the overcoming of a stress or adversity, or a relatively good outcome despite risk experiences” (Rutter, 2012, p. 336). Within this framework, the present article contains first results of a systematic review that concentrates on studies of potential protective factors against extremism and radicalization towards violence.

PRISMA flow diagram of the search procedure for the systematic review (for details and reasons of exclusion, see text).
Method
Eligibility Criteria
The eligibility criteria of our systematic review are summarized in Table 1. A pilot search revealed a small number of studies on outcomes of violent behavior. Therefore, we also included outcomes that are often precursors of violent behavior like sympathies for radical violence, the willingness to use violence or a mind-set explicitly supporting or justifying violence. Studies examining broader underlying attitudes or propensities (e.g., right-wing authoritarianism or social dominance) were not eligible. Following Doosje et al. (2016), we included research on far-right, far-left, religiously/ethnically 1 motivated, nationalist/separatist, and special issue groups. Eligible were studies on factors that should reduce the likelihood of involvement in or support of violent acts (protective factors). Studies that addressed only risk factors were excluded. In accordance with research on resilience (Lösel & Bender, 2003; Rutter, 2012) and desistance from crime (Farrall, Hough, Maruna, & Sparks, 2011; Shapland, Bottoms, & Grant, 2012), we also included studies on de-radicalization and disengagement (Barelle, 2014; Bjørgo, 2011; Horgan, 2009).
For reasons of comparability and space, this review only includes quantitative analyses of protective factors. Qualitative research will be reviewed in another part of our project. Single case studies were not eligible. The quantitative analyses had to provide sufficient statistical information (e.g., means, standard deviations, beta-weights, path coefficients, significance tests). The design could be both cross-sectional and longitudinal. Beyond correlational designs, we included intervention studies if they provided quantitative data on specified protective factors.
We concentrated on reports in English and German language, but there seemed to be no quantitative research in other languages. There was no restriction regarding time and mode of publication. ‘Unpublished’ work such as dissertations or ‘grey’ literature was also included. All countries of origin were eligible. There was also no restriction on specific scientific disciplines of authors (e.g., sociology, psychology, political science, criminology, anthropology, and cultural science).
Literature Search
We carried out two pilot tests using a very broad and a narrow research string. The results were not satisfactory and therefore we created rather complex search strings including the terms right-wing, left-wing, islamis*, salafi*, radicali*, jihad*, extremis*, and terror* combined with protect*, buffer*, resilien*, risk factor*, disengag*, deradical*, moderat*, reject*, desist*. Using these strings, we searched 15 databases (Cochrane Library, Campbell Reviews, Dissertation Abstracts, MEDLINE, PubMed, EMBASE, ERIC, German National Library, PsycINFO, Psyndex, Science Direct, Scopus, Sociological Abstracts, Sociological Collection, and World Cat). Our search procedure is visualized in a PRISMA diagram in Fig. 1.
After elimination of duplicates, there remained 354 reports. These were screened on the title and, if necessary, on the abstract level. We had to exclude 318 documents because they contained theoretical discussions, case studies, narrative reviews, country reports, police strategies, non-empirical descriptions of practice projects, and other types of reports that did not meet our inclusion criteria. The remaining 36 reports were read by two members of our team and evaluated with regard to the inclusion criteria. Only seven studies met our not very narrow inclusion criteria (see Table 1). The excluded reports contained qualitative instead of quantitative research, focussed only on risk factors, or did not provide sufficient information on statistical details for an adequate evaluation of the effects. It is a general experience in systematic reviews and meta-analyses that the search in literature databases only reveals a rather limited number of relevant studies (Schmucker & Lösel, 2011, 2015). Therefore, we also tried to find other relevant studies through a “snowball” procedure in which we used the references of eligible primary studies, references in reviews, and personal contacts to enlarge our study pool. Through this route, we inspected the titles of 2,171 documents and when some looked promising, we read the abstract. If the abstract seemed to be relevant for our topic, we inspected the article. Although a “snowball” procedure is not as representative as a systematic search of databases, it is widely used and helpful for increasing samples of primary studies. In our review, we retrieved ten more eligible documents so that we had a total of n = 17 reports that provided quantitative data on protective factors against extremism and radicalization. Some reports contained more than one different data set so that we could finally include 21 separate analyses (for reasons of simplicity in the following called “studies”).
Overview of the Eligibility Criteria*
Notes* For more details see the text.
Coding
We used a scheme to systematize the characteristics of the studies. Some main categories are shown in Table 2 (see results section). All eligible study reports were analyzed by two members of our team. Characteristics such as sample size, age range, country of origin etc. were simple facts and easy to code. In some cases, the available information was more complex. Therefore, we adhered to the terminology and categories reported by the study authors as much as possible. In some studies, we had to be more interpretative. For example, one study (Baier et al., 2016) did not explicitly use the types of extremism as described in Table 1, but equated Islamist extremism with hostility towards Germans. We followed the recommendation of Lipsey and Wilson (2001) for reviews with small study numbers and classified the content in a discussion and final consensus of two members of our researchteam.
Identified Protective Factors Against Different Types of Extremism and Radicalization
Notes. rw = right-wing ideology; lw = left-wing ideology; r/e = religious/ethnic ideology; n/s nationalist/separatist ideology.
a One sample, three extremism types in three different analyses; bOne study dividing the sample in different groups, assessing the three types of extremism selectively.
Results
Descriptive Data of Included Studies
The eligible studies differed with regard to the type of underlying extremism, design, regional context, samples, data analyses, outcomes, and measures.
About half of the 21 eligible studies (n = 11) addressed protective factors against religious/ethnic extremism. Five studies investigated protective factors against far-right extremism, two far-left, one nationalist/separatist, and two mixed orientations. Most of the studies had a cross-sectional design (n = 16), although some of these contained retrospective longitudinal information (e.g., on family or school variables). One correlational study had a prospective longitudinal design with a follow-up of one year. Four studies reported effects of interventions in a pre-post-design, of which only one included a control group.
Ten of the 17 reports originated in Europe, three in the Middle East, two in North America, and one each in Asia and Africa. The samples mainly covered young people and young adults up to the age of 30 (n = 11) which included students at school and in further education (n = 6). Six studies examinedsamples with an age from adolescence to adulthood. Two studies referred to convicted radical offenders. For details of the age ranges and sample sizes see Table 2.
With regard to the different methods of analyzing protective factors, the studies were heterogeneous. A majority addressed both risk and protective factors but separated the latter in their data analysis. The most frequently used methods were linear, hierarchic or logistic regressions (n = 9). In the remaining analyses, path models (n = 6) and group comparisons (n = 2) were applied. These designs could reveal the respective protective factors by significant negative beta weights or path coefficients for the extremist outcomes. Only one study carried out moderator analyses on interactions of the protective factors with a present risk (Pauwels & Svensson, 2017).
Regarding the outcome criteria, some studies measured only one distinct relevant outcome such as extremist attitudes (n = 3), extremist behavior (n = 6) or willingness to use violence (n = 1) as dependent variable. Others examined both behavior and attitudes (n = 1), attitudes and willingness to use violence (n = 4), or applied a composite instrument for attitudes and behavior (n = 2). While two studies drew on a special database (PIRUS), the rest relied on self-reports.
Results on Protective Factors
Most studies investigated a range of variables. For reasons of parsimony and clarity, we only present those variables that showed a significant protective effect. Our review is exploratory and not theory-testing and protective effects are difficult to detect (Lösel & Bender, 2017). Therefore, we used p < 0.10 (two-sided) as a threshold, but most significant effects were at p < 0.05 or lower. We found 52 analyses and 30 different variables that showed a protective effect. About half of the effects were on (mostly self-reported) behavior as outcome, the other half on extremist attitudes and willingness for violent action. Nearly 90 percent of the significant effects derived from cross-sectional analyses, the rest from longitudinal designs.
The protective factors were grouped according to the classification of Lösel and Farrington (2012): individual, family, school, peer group, and community/society factors. As in other classifications, there is some overlap between these categories. Our main findings are presented in Table 2. More details will be provided in our final project report.
Various protective factors were identified against more than one kind of extremism or radicalization. At the individual level, self-control reduced violence in persons with far-right, far-left, and religious/ethnic extremism, particularly in young people holding highly extremist beliefs (Pauwels & Svensson, 2017). Similarly, adherence to law was a replicated protective factor against extremist attitudes in the religious/ethnic, right-wing as well as left-wing spectrum and reduced violent left-wing extremist behavior (Baier et al., 2016). Acceptance of police legitimacy (Pauwels & De Waele, 2014) and illness/disease (Bhui, Everitt, Jones, & Correa-Velez, 2014) also had a protective effect against religious/ethic extremism in more than one analysis. It should, however, be kept in mind that these findings based on several comparisons within single projects.
At the family level, an appreciative/positive parenting behavior (Baier et al., 2016), non-violent significant others (Jasko, LaFree, & Kruglanski, 2016), and ownership of residential property (Asal, Fair, & Shellman, 2008) had a protective effect on different kinds of extremism.
There were also replicated protective effects at the school level. Both, good school achievement and bonding to school reduced far-right and far-left extremist attitudes and behavior (Baier et al., 2016; Boehnke, Hagan, & Merkens, 1998).
With regard to peer group influences, contact with non-violent peers had a protective effect against mixed types of extremism and nationalist/separatist orientations in more than one study (Cragin, Bradley, Robinson, & Steinberg, 2015; Jasko et al., 2016). Contact to foreigners protected against far-right extremism (Fuchs, 2003).
The latter result could also be assigned to the community/society level, where a basic attachment to society appeared as protective against religious/ethnic extremism (van Bergen, Ersanilli, Pels, & de Ruyter, 2016; van Bergen, Feddes, Doosje, & Pels, 2015). The scarcity of protective factors in this domain is plausible because most data were gathered from individuals and thus only indirectly reflect macro-level influences.
The majority of significant effects in Table 2 based on one study, one sample, one cultural context, and one type of extremism respectively. Insofar it was not meaningful to make comparisons between the studies. However, as most of the studies were cross-sectional, we asked whether the few potentially more valid longitudinal designs produced different findings. Only one longitudinal study (Boehnke et al., 1998) had a correlational design like the prevailing cross-sectional studies, whereas the other four contained data before and after an intervention. At first glance, the studies with a longitudinal design seemed to reveal fewer protective effects. Boehnke et al. (1998) and the intervention studies of Feddes, Mann, and Doosje (2015) and Liht and Savage (2013) each reported only one significant protective effect and Amjad and Wood (2009) and Savage (2014) found none. However, this pattern does not justify a general conclusion about the impact of design characteristics. Some intervention studies had only few outcome measures and the intervention of Amjad and Wood (2009) in Pakistan contained lectures against extremism. Due to this program it may have been unlikely to produce a change that could be interpreted as a protective effect.
Most of the single effects in Table 2 were plausible, e.g., a protective function of employment, concern about getting incarcerated, empathy with non-Muslims, becoming politically disinterested, value complexity, incarcerated or non-violent family members, higher education, and being a first generation migrant. Other results seemed to be somewhat inconsistent, such as intensive religious practice versus low importance of religion or having a wider social network versus low social capital. Some of these findings are rather specific and may be counter-intuitive at first glance. These and other results will be addressed in the following discussion.
Discussion
Many studies on extremism, radicalization and recruitment into terrorism focus on risk factors. However, most individuals with a ‘risky’ profile do not start and/or not proceed on a pathway toward violent radicalization. This issue is similar to what we know from research on drug users: Many consumed cannabis or other ‘soft’ drugs, but only a minority moved on to heroin and other ‘hard’ drugs. Accordingly, it is highly important to understand the factors and influences that protect and/or buffer against extremism and violent radicalization.
To our knowledge, our study is the first systematic review with a particular focus on protective factors on this topic. Although we retrieved a large number of reports and our eligibility criteria were not narrow, we only found a moderate number of includable studies. This may be partially due to a general scarcity of research on protective factors. More important is that politically “hot topics” typically trigger many publications and short-term projects. As sound empirical studies need time, many publications contain terminological and theoretical discussions, descriptive data, single case analyses, and practical recommendations (see also Bouhana & Wikström, 2010; LaFree & Freilich, 2018). Similarly, early reviews of the Campbell Collaboration on controlled research about “hot topics” like terrorism or human trafficking lead to (nearly) “empty reviews”. In an own meta-analysis on the ‘hot topic’ of sex offender treatment, we checked more than 3,000 documents, but could include only 29 studies that met the inclusion criteria (Schmucker & Lösel, 2015).
In spite of the limited number of studies, our findings showed a range of factors that seemed to have a protective function against extremism and violent radicalization. Various factors were remarkably similar to what has been reported in studies of protective effects against more general violence of young people (Lösel & Bender, 2017; Lösel &Farrington, 2012).
With regard to individual protective factors, self-control (low impulsivity) was one of the characteristics that are consistent to research in other fields of crime and violence (Lösel & Bender, 2017). Another protective factor was employment. This is consistent to North American research on radicalization (Jensen et al., 2016) and desistance from crime (e.g., Sampson & Laub, 2003). Although our synthesis (like other studies) subsumed good school achievement under school factors, it is also an indirect indicator of above-average intelligence. This personal resource has repeatedly shown a protective effect against crime and violence (Ttofi et al., 2016). Our finding of a protective function of anxiety about being caught and incarcerated is also in accordance with the broader literature on youth violence (Lösel & Farrington, 2012). Anxious mood goes along with higher physiological arousal and may thus buffer against risky behavior and thrill-seeking crime (Raine, 2013). With regard to the protective effect of empathy, it should be noted that this is a complex construct (Joliffe & Murray, 2012) and the result of Feddes et al. (2015) explicitly refers to empathy towards non-Muslims. This protective function can be explained by aspects of social information processing (Dodge & Pettit, 2003; Lösel, Bliesener, & Bender, 2007). Within this context, perspective taking should counteract black-and-white thinking and hostility biases in violence-prone cognitive schemata. In contrast, empathy may be a risk factor if it is only shown for the own radical Muslim group members (Rahimullah et al., 2013). The protective impact of illness in a sample of Muslim migrants in England (Bhui et al., 2014) may look very special at first sight, but in a study of “hard core” football hooligans health problems were also relevant for desistance from violence (Lösel & Bliesener, 2006).
Our findings on subjective experiences of deprivation, discrimination, and dissatisfaction with own quality of life seem to deviate from research in which these factors were risks for violence. The respective studies in our review addressed attitudes towards suicide attacks in the West Bank (Cragin et al., 2015), far-right attitudes of school students in Germany (Fuchs, 2003), and attitudes on religious extremism of young adults in Belgium (Pauwels & De Waele, 2014). We assume different mechanisms behind these findings. For example, the far-right attitudes in the German study were related to higher self-esteem and authoritarian attitudes as risk factors that may counteract feelings of discrimination and deprivation. This is in accordance with feelings of superiority in right-wing Flemish groups (De Waele & Pauwels, 2016). Feelings of dissatisfaction may also indicate internalizing behavior problems that lead to social withdrawal and thus can protect against affiliation with extremist groups.
In the British study of people with Pakistani or Bangladeshi roots, “not being born in the UK” was connected to more condemnation of violent extremism (Bhui et al., 2104). This may look counter-intuitive at first sight, but becomes plausible when one takes cultural tensions and different generations of migration into account (Mansour, 2016; see also below). Another context-specific finding relates to religious orientation. Whereas Muluk, Sumaktoyo, and Ruth (2013)found that intensive religious practice was a protective factor against committing extremist violence in Indonesia, less identification with religion was protective against violence-oriented attitudes in Muslim migrants in England (Bhui et al., 2014). These examples suggest that cultural framing conditions play a role for the meaning and impact of specific factors in different contexts.
On the family level, parenting behavior seems to have a protective effect. Although the terminology and measures differed, the significant finding on an ‘egalitarian’ parenting style is similar to the protective effects of positive parenting in other fields of youth violence (Lösel & Bender, 2017). Another family factor that showed a protective effect was ownership of residential property. This is similar to a protective function of the family’s socio-economic status in violence research (Lösel & Farrington, 2012). More complex are the findings that non-violent family members (Cragin et al., 2015) as well as family members associated with militant religious groups (Asal et al., 2008) or in custody (Cragin et al., 2015) may have a protective influence. Such results are not inconsistent when different contexts and mechanisms are taken into account. On the one hand, they may indicate vicarious reinforcement. On the other hand, they may involve a more or less conscious deterrent effect when family members faced serious negative consequences.
With regard to school factors, the findings were consistent to research in other criminological fields (Lösel & Bender, 2017): Bonding to school, good school achievement, and a higher educational level were protective factors against extremism and radicalization in several studies.
In the domain of peer relations, having contact with non-deviant peer groups or some social isolation seems to reduce the risk of extremism/radicalization. This is in accordance with the literature on protective effects against youth violence (Lösel & Farrington, 2012). The somewhat contradicting finding on a protective function of both a wider social network and low social capital may be due to different underlying mechanisms. When more social contacts include non-extremist people and when a lack of social capital also means less contact with extremist networks, the respective results are plausible.
At the community/society level, a basic attachment to or integration into society seems to protect against extremism and radicalization. This agrees with the theory on informal social control and social bonding (Hirschi, 1969). Social bonding may also enhance the above-mentioned protective function of self-control (Hirschi, 2004), although this theory needs more evidence. Another protective factor seems to be specifically relevant for religiously/ethnically oriented extremism: The study of Bhui et al. (2014) suggests that first generation immigrants are less vulnerable to extremism/radicalization than later generations. This is in accordance with findings of a particular risk for youth violence in later generations of migrants (Walburg, 2014). Whereas first generation immigrants often aim to adapt to the society, some of their offspring become deviant due to problems of integration, parenting, and lower education (Baier, Pfeiffer, Simonson, & Rabold, 2009).
As a number of protective factors seem to be similar to what we know from other fields of violence of young people, extremism and violent radicalization should become more integrated into the broader research on youth violence in developmental criminology and psychology. School bullying and violence can serve as an example for that. Although in the early phases of research, bullying was mainly seen as a school phenomenon, later studies demonstrated the developmental relations to violent behavior in other contexts and in adulthood (Bender & Lösel, 2011; Ttofi, Farrington, & Lösel, 2012).
Our review also revealed protective factors that seem to be more specific. For example, individuals who had an extremist orientation less often developed violence when they had an accepting attitude towards law, society, and police legitimacy. This is in accordance with theories on legitimacy (Tyler, 2006). A specific facet is relevant for far-right extremism because these groups emphasize law and order. Their development into radicalization may therefore involve ambivalence between conformism to state authority on the one hand and the feeling of insufficient law and order on the other hand. Such a mechanism is also suggested by Liht and Savage (2013) who found a protective effect of an acknowledgement of value complexity.
When we summarize the main findings on those protective factors that were replicated in various analyses, there are both similarities and differences between types of extremism: Self-control and a basic adherence to law were relevant for right- and left-wing as well as religious/ethnic extremism. Appreciative parenting had a protective function against left-wing and religious/ethnic extremism. Significant non-violent others and non-violent peers were protective in mixed extremist groups. Good school achievement and bonding to school were only protective for right- and left-wing extremism, but not for the religious/ethnic form. For the latter type, a basic attachment to society, illness and religious issues were specific protective factors (although not the same in different cultural contexts). However, as mentioned above, the typologies of extremism are not clear-cut and may contain some overlap.
Theoretical and Practical Conclusions
Only some of the reviewed studies investigated the same constructs and variables. As in other criminological fields, more replications are needed (Lösel, 2017). However, there were consistencies across the studies that can be used for developments in theory and practice. The above discussion has shown various theoretical explanations on the protective function of specific constructs/variables. These were mainly based on middle-range theories that should be further integrated (Pauwels & De Waele, 2014). The literature shows various promising approaches for more integrated theories to explain extremism and violent radicalization. One example is the General Aggression Model (Anderson & Bushman, 2002; DeWall, Anderson, & Bushman, 2011). Situational Action Theory (Wikström, Oberwittler, Treiber, & Hardie, 2012) can also partially explain extremism and radicalization through individual propensities, moral attitudes, self-control, and situational influences. Pauwels and De Waele (2014) presented a promising integration of constructs from control theory, procedural justice theory, general strain theory, social learning theory, and self-control theory. Our review showed that some of these constructs also have a protective pole in the process of becoming radicalized towards violent extremism. However, similar to risk research on extremism, most of the studies in our review were cross-sectional. Their regressions and path analyses provide important information, but do not really reflect developmental processes over time. In addition, many studies only rely on self-reports what may lead to an over-estimation of correlations. For a further integration of various middle-range theories, we propose to add the concept of a developmental chain reaction (Lösel & Bender, 2003). This begins with early family influences and temperament factors and leads to specific social cognitions, self-concepts/identities, peer group, and school influences (Lösel & Bender, 2003). Social information processing may play a key role in such processes when aggression-prone cognitive schemata are related to experiences of aggression in various social contexts (e.g., Lösel et al., 2007; see also Schils & Pauwels, 2016). Theories of intergroup relations and emotions (Figueiredo, Valentim, & Doosje, 2014) should be integrated as well. Overall, there is a clear need to apply more general criminological and other social science theories to the topic of violent extremism (La Free et al., 2018).
In our view it is extremely complicated to test complex integrative theories on extremism and radicalization in single studies with different data sources. However, similar to psychotherapy, already some eclectic combinations of middle range hypotheses can be useful in practice. The static and dynamic protective factors we found in our review may be helpful for approaches to prevention and intervention. The concept of developmental chain reactions implies that risk and protective factors can be addressed at various developmental phases to promote resilience. The similarities of protective factors against general youth violence and extremism suggest that early prevention programs in families, schools, and neighborhoods (Farrington, Gaffney, Lösel, & Ttofi, 2017) may be useful to prevent radicalization at an early stage. Similar to risk factors, most single protective effects are small and need to be accumulated according to dose-response-relationships (Lösel & Farrington, 2012). This requires multi-modal approaches to prevention (Lösel, Stemmler, & Bender, 2013). As theories on (causal) risk and protective factors need to be translated into practical interventions, well-controlled process and outcome evaluations are necessary. Whereas there is a body of sound evaluations of programs against intergroup prejudice (Beelmann & Heinemann, 2014), much less controlled studies address projects against extremism and radicalization (Armborst & Kober, 2017). More experimental evaluations in this field are necessary. They could not only help to make adequate decisions on “what works”, but would be the most valid tests of causal theories on extremism and radicalization.
Author’s Note
This study has been funded by the European Commission as a work package within the research consortium “Modelling the processes leading to organized crime and terrorist networks” (PROTON).
Footnotes
Bio Sketches
Friedrich Lösel, Dr. phil. (Psychology). Emeritus professor and past director at the Institute of Criminology, Cambridge University (UK), and the Institute of Psychology, University of Erlangen-Nuremberg (Germany). Research on juvenile delinquency, offender treatment, resilience, prisoners’ families, school bullying, psychopathy, and early developmental prevention. Awards, i. a.: Sellin-Glueck Award of the American Society of Criminology, Lifetime Achievement Award of the European Association of Psychology & Law, German Psychology Prize, and Stockholm Prize in Criminology.
Sonja King, M. Sc. (Psychology). Research assistant at the Institute of Psychology, University of Erlangen-Nuremberg. Research on radicalization and violent extremism with a focus on pathways and root causes, assessment, intervention, and the prisonsetting.
Doris Bender, Dr. phil. (Psychology). Senior lecturer at the Institute of Psychology, University of Erlangen-Nuremberg (Germany). Research on resilience and protective factors against antisocial development, child maltreatment and abuse, family relationships, long-term outcomes of school bullying, and extremism and radicalization.
Irina Jugl, M. Sc. (Psychology). Research assistant at the Institute of Psychology, University of Erlangen-Nuremberg. Visiting scholar at the Institute of Criminology, University of Cambridge, 09/2016 – 02/2017. Research on bullying and victimization, sports programmes as a measure of crime prevention and violent extremism.
We combined “religious” extremism with the term “ethnic” because in Islamist extremism the ethnic background is very closely related to religion. Whereas Christians need to be baptized, in Islamic cultures (like in Judaism) the religion comes along with birth and apostasy is often severely sanctioned. Accordingly, “Muslim” extremists may be more influenced by their ethnic group than by their actual “religiosity”. In general, there is not a clear red line between the different motivations and ideological backgrounds. Accordingly, Doosje et al. (2016) assign some terror groups to more than one type of extremism. For example, they assign ISIS not only to the religious type, but also include it in the nationalist-separatist category because they claim territory. The “Islamic Jihad Movement in Palestine” is motivated by religious aspirations, but also by revenge for the loss of territory after the foundation of the State of Israel in 1948. These examples suggest that “religious” extremism may be a melting pot for different kinds of motivation. Similar problems exist with regard to other types of extremism. For example, right-wing extremists may also belong to thrill-seeking violent subcultures (Lösel & Bliesener, 2006) and not be as politically motivated as it appears on first glance. There are also other similarities in the ideologies and the background of different extremist groups (Ebner, 2017; Jensen et al., 2016).
Quantitative studies included in the systematic review.
