Collective moral disengagement and its associations with bullying perpetration and victimization in students

Abstract The aim of the current study was to examine whether collective moral disengagement in the classroom was associated with bullying perpetration and victimisation. One-thousand-and-fifty-four students from 70 classrooms in 29 schools in the middle and southern parts of Sweden completed a questionnaire in their classroom. In line with the hypotheses, the bivariate correlation analyses at the classroom level showed that students who belonged to classrooms with lower collective moral disengagement were less likely to be victimised by bullying or engaged in bullying perpetration. Moreover, when controlling for gender and age at the individual level and including collective moral disengagement in the same model at the classroom level, multilevel analyses revealed that students who belonged to classrooms with a higher level of collective moral disengagement were more likely to be engaged in bullying perpetration or to be targets of bullying victimisation.


Introduction
School bullying can be defined as an interaction pattern in which 'an individual or a group of individuals repeatedly attacks, humiliates, and/or excludes a relatively powerless person' (Salmivalli, 2010, p. 112). In addition, face-to-face bullying in schools can include physical, verbal and relational forms (Varjas et al., 2009). Bullying is a social phenomenon rather than an individual problem that takes place in peer groups and social contexts (Hymel et al., 2015;Pepler et al., 2010;Salmivalli, 2010). Evidence that bullying is a social phenomenon is supported by the fact that peer bystanders are present in most bullying incidents according to observational studies (Craig & Pepler, 1998;Craig et al., 2000) and play various participant roles (Salmivalli, 1999). In this context, social factors can contribute to understanding school bullying as researchers have found that the prevalence of bullying varies as a function of school climate (Cornell & Huang, 2016;Cornell et al., 2015;Gerlinger & Wo, 2016;Gregory et al., 2010), classroom climate (Thornberg et al., 2018), and peer norms at the classroom level (Salmivalli & Voeten, 2004). In a meta-analysis examining research on factors affecting bullying, peer influence had the largest overall effect size when compared to community factors, school climate, family/home environment, and peer status (Cook et al., 2010).
In the current study, we adopted a social cognitive framework (Bandura, 1997(Bandura, , 2000(Bandura, , 2016, which emphasises an interplay or a reciprocal influence between behaviour, personal factors, and external environment (the so-called triadic codetermination process; Bandura, 1997Bandura, , 2000Bandura, , 2016. This framework further states that people strive to exercise control or agency over events that affect their lives. In this context, human agency operates within interdependent relationships between personal, behavioural, and environmental/contextual influences. Bandura (1997Bandura ( , 2000 argues that concepts of human agency have been essentially confined to personal agency in terms of cognitive, affective, motivational, and choice processes, whereas the social-cognitive theory extends it to include collective agency. 'People do not live their lives in individual autonomy. Indeed, many of the outcomes they seek are achievable only through interdependent efforts' (Bandura, 2000, p. 75). Collective agency means that people are interdependent, pool their competences and resources, and work together to solve problems and gain shared goals (Bandura, 1997;Fernandez-Ballesteros et al., 2002). Thus, to better understand school bullying from a social-cognitive perspective, we need to know more about the social context such as the peer group and the classroom context, and examine how social-cognitive factors at the group level are associated with bullying perpetration and victimisation. In the present study, we have delimited our investigation to classroom collective moral disengagement.

Classroom collective moral disengagement
According to social-cognitive theory (Bandura, 2002(Bandura, , 2016, the exercise of moral agency includes two possible forms. The inhibitive form refers to refraining from inhumane actions, such as bullying perpetration. The proactive form refers to humane actions, such as helping someone in distress. As all domains of human agency, moral agency is the product of a complex interplay between personal, behavioural, and environmental influences (Bandura, 2002(Bandura, , 2016. Bullying perpetration is a failure of the inhibitive forms of moral agency and can, at least in part, be understood in terms of moral disengagement, which refers to a set of self-serving cognitive distortions by which self-regulated mechanisms can be deactivated and moral self-sanctions can be disengaged. Such distortions facilitate inhumane behaviour without any feelings of remorse or guilt. Examples of moral disengagement mechanisms are moral justification (i.e. justifying the means by providing a greater moral ends), diffusion of responsibility (i.e. diluting personal responsibility because other individuals are present and engaged in the same negative behaviour as well), distorting the consequences (i.e. thinking that there are no real harmful effects of the negative behaviour), and blaming the victim (i.e. convincing oneself that the victims deserve their suffering; Bandura, 2002Bandura, , 2016. Research has shown that among children and youth, moral disengagement is positively associated with delinquency (Bandura et al., 1996;Bandura et al., 2001;Hyde et al., 2010), aggression (Bandura et al., 2001;Barchia & Bussey, 2011;Paciello et al., 2008;Pornari & Wood, 2010), bullying perpetration (Caravita et al., 2012;Gini et al., 2011;Kokkinos & Kipritsi, 2018;Pepler et al., 2008;Pozzoli et al., 2016;Thornberg et al., 2015), and cyberbullying perpetration Perren, & Gutzwiller-Helfenfinger, 2012). Meta-analyses have confirmed that children and adolescents who displayed higher levels of moral disengagement were more inclined to bully others Killer et al., 2019).
Although moral disengagement has largely been examined at the individual level, it has been proposed to work at the group-level as well (White et al., 2009). Moral disengagement can be widespread and shared across group members and thus considered to be a part of the peer, classroom and school culture or climate, depending on the level of the analysis. Within the social-cognitive framework (Bandura, 2016), collective moral disengagement has been constructed to refer to moral disengagement beliefs that are shared within a significant social group . It is not simply the aggregation of the moral disengagement of its individual members, but a group-level phenomenon of perceived shared beliefs produced by the group dynamics (Bandura, 2016). Caravita et al. (2014) demonstrated the presence of the social influence of moral disengagement in peer groups by showing that students in early adolescence became more similar to their friends in their proneness to morally disengage.
With reference to the social-cognitive theory of moral agency (Bandura, 2016), it is plausible to assume that inhumane or immoral behaviour such as bullying should be more prevalent in groups, such as classroom groups, which display high levels of collective moral disengagement. In accordance with that assumption, research has in fact shown that collective moral disengagement at the classroom level is associated with aggression , bullying perpetration (Bj€ arehed et al., 2019;Kollerov a et al., 2018;Thornberg et al., 2019), and bullying victimisation (Kollerov a et al., 2018). Classroom collective moral disengagement can therefore be interpreted as a group characteristic at the classroom level with the potential of influencing the prevalence of peer aggression and bullying among classmates. Still, very few studies have examined whether collective moral disengagement is associated with bullying and victimisation. Therefore, more studies are needed to further test and replicate the associations between these variables. Furthermore, it is still unknown whether collective moral disengagement at the classroom level interacts with gender at the individual level to help explain the variances of bullying perpetration and victimisation among schoolchildren. Previous research has found that boys were more engaged in bullying perpetration than girls (for metaanalyses, see Cook et al., 2010;Mitsopoulou & Giovazolias, 2015), although gender failed to significantly predict bullying victimisation (Cook et al., 2010). Investigating whether classroom collective moral disengagement might moderate or influence the gender effect on bullying perpetration and (the lack of gender effect on) victimisation would shed some light on gender differences in school bullying. As far as we know, the current study is the first to examine these possible cross-level interaction effects.

Current study
The aim of the current study was to examine whether collective moral disengagement at the classroom level was associated with bullying perpetration and victimisation. In accordance with the social cognitive theory on moral agency (Bandura, 2016) and previous studies (Bj€ arehed et al., 2019;Kollerov a et al., 2018;Thornberg et al., 2019), we hypothesised that classroom collective moral disengagement would be associated with greater bullying perpetration and victimisation. In the current study, bullying perpetration and victimisation are treated as continuous variables.
Gender and age were included as covariates. In line with previous meta-analyses (Cook et al., 2010;Mitsopoulou & Giovazolias, 2015), we hypothesised that being a boy would be associated with a higher probability of bullying perpetration, whereas gender would be unrelated to bullying victimisation.
Considering age, in their meta-analysis, Cook et al. (2010) found that the association between age and bullying perpetration was positive but weak. Their analysis included studies with participants in K-12 settings (5-18 years old). In contrast, another metaanalysis found a weak but significant negative association between age and bullying perpetration. However, this analysis only included studies with a longitudinal or prospective design and with participants between 11 and 18 years (Kljakovic & Hunt, 2016). Based on their systematic review, Alvarez-Garc ıa et al. (2015) concluded that the likelihood of bullying increased with age until about age 14 years, when it decreased, but in some studies, the decrease started from age 11-12 years. Due to the mixed findings and the small effect sizes reported in the meta-analyses, age was included as a covariate in the present study but without a clear hypothesis in relation to bullying perpetration. In addition, meta-analyses have shown a non-significant association between age and bullying victimisation (Cook et al., 2010;Kljakovic & Hunt, 2016). We therefore hypothesised that age would be unrelated with bullying victimisation.
Finally, with reference to social cognitive theory (Bandura, 1997(Bandura, , 2000(Bandura, , 2016 emphasising that behaviours are produced by interdependent associations between individual and contextual factors, we tested two possible cross-level interaction effects: (a) classroom collective moral disengagement x gender, (b) classroom collective moral disengagement x age. However, because of the lack of previous empirical research, the literature did not offer us any clear hypotheses to test. Possible cross-level interaction effects were examined in an exploratory fashion.

Participants
In the current study, participants included 1054 students from 70 classrooms in 29 schools located the middle and southern parts of Sweden. There were 487 girls and 567 boys from grades 4-6 with an age range of 10-14 years old (M age ¼ 11.63, SD ¼ 0.83). We did not measure socioeconomic status at the individual level, but as a result of purposeful sampling of schools, the sample of public schools included a wide range of socio-geographic areas as well as socioeconomic backgrounds of students. The vast majority of the participants were of Swedish ethnicity, and only a small minority (6%) had a foreign background, that is, either they were born in another country or both their parents were born in another country. The Swedish school system consists of a kindergarten year (the year a child turns six) and then nine years of compulsory schooling, including elementary school (grades 1-6) and secondary school (grades 7-9). Generally, elementary school students are in the same class all day, have a home classroom in which the vast majority of their classes take place, and have the same teacher in most school subjects. Students have the same teacher for grades 1-3, and then a new teacher works with them for grades 4-6. In secondary school, students meet a variety of subject teachers, and they change classrooms for each subject.
The original sample consisted of 1416 students (666 girls and 750 boys). In the study, 256 students were excluded because their parents did not grant active consent; 58 students were absent at the data collection session; 16 students did not give their assent to participate; five students were excluded due to participation difficulties (i.e. language, reading or cognitive difficulties); and 27 students were dropped because they did not fill out any information on at least one of the scales included in the study, leading to the final sample of 1054 students. This investigation received ethical approval from the Regional Ethical Review Board at Link€ oping, and we obtained active, written parental consent and student assent for all participating students.

Procedure
The participants completed a questionnaire in their classroom. Trained graduate students in psychology (the sixth, seventh and eighth authors of this article) were present in the classrooms during the survey administration. They explained the study procedures, reassured students that their participation was voluntary and confidential, and assisted the participants who needed help. The participants responded anonymously to the questionnaire. To ensure privacy, the participants were instructed to move away from each other and separate their desks.

Socio-demographic scale
Participants completed a socio-demographic scale that included questions about their age (i.e. 'How old are you?' followed by, 'I'm … … years and … … months old'), gender (0 ¼ girl, 1 ¼ boy), and Swedish vs. foreign background (i.e. 'Were you born in Sweden? Was your mother born in Sweden? Was your father born in Sweden?'). Participants who reported that they were born in another country and/or that both their parents were born in another country were categorised with a foreign background.
Collective moral disengagement A Swedish translation of Gini and colleagues' (2015) 17-item scale was used to measure collective moral disengagement. The scale was translated through a back-translation procedure. The 17 items covered the eight moral disengagement mechanisms (e.g. 'it is alright to fight to protect friends', 'if kids fight and misbehave in school it is their teacher's fault', 'insults among children do not hurt anyone'). For each item, the participants were asked to rate, 'in your classroom, how many kids think that … [item]' on a 5-point scale: 'None', 'About a quarter (25%)', 'About a half (50%)', About three quarter (75%)', and 'Everyone'. The 17 items were averaged into one scale  which was supported with the current study's sample as follows: v 2 Robust (119)¼425.276, p<.001, RMSEA Robust ¼0.036 (90%C.I.: 0.032; 0.039), CFI Robust ¼0.964, SRMR Robust ¼0.05, Cronbach's a ¼ .83. Collective moral disengagement for the classroom group was then computed as the classroom average.

Bullying perpetration and victimisation
A Swedish translation of Student Survey of Bullying Behaviour-Revised 2A (Varjas et al., 2006) was used to measure bullying perpetration and victimisation. The scale was translated through a back-translation procedure. In the victimisation scale, the participants were asked, 'How often in the past couple of months have older, bigger, more popular, or more powerful kids picked on you by … ' followed by 13 items (e.g. 'hitting or kicking you', 'saying mean things to you', and 'spreading rumors about you'). They rated each item on a four-point scale (0 ¼ 'not at all', 1 ¼ 'just once or twice', 2 ¼ '2-3 times a month', and 3 ¼ 'once a week or more'. This translation resulted in the following data providing psychometric support to the Swedish translation of this scale (i.e. v 2 Robust (65)¼273.563, p<.001, RMSEA Robust ¼ 0.026 (90%C.I.: 0.023; 0.029), CFI Robust ¼ 0.987, SRMR Robust ¼0.092, Cronbach's a ¼ .93). In the bullying scale, the participants were asked, 'How often in the past couple of months have YOU picked on younger, smaller, less popular, or less powerful kids by … ' followed by 12 items (e.g. 'hitting or kicking them', 'saying means thing to them', and 'spreading rumors about them'). The participants rated each item on the same four-point scale as in the victimisation scale with the following data to support the psychometric characteristics of this scale: v 2 Robust (54)¼95.977, p<.001, RMSEA Robust ¼0.015 (90%C.I.: 0.01; 0.02), CFI Robust ¼0.985, SRMR Robust ¼0.056, Cronbach's a ¼ .83.

Statistical models
Separate multilevel regression models were analysed for the dependent variables victimisation and bullying. First, a model with only the control variables gender and age was estimated for each dependent variable, allowing the intercept to vary between classrooms. Model 1 is shown below: where DV ij is the victimisation and bullying score, respectively, for the ith student in the jth classroom, a is the intercept in classroom j, b to b 2 are regression slopes for individual effects, E ij is a student residual, a is the mean intercept across classes, and u j is a classroom residual. It is assumed that u j $ N 0, r 2 u À Á , e ij $Nð0, r 2 e Þ and covðu j , e ij Þ ¼ 0, where r 2 u is the variance between classrooms, and r 2 e is the variance within classrooms. In the second model, the class variable collective moral disengagement (CMD) was added. Model 2 is shown below: Where b 1 to b 2 are regression slopes for individual effects and c 1 is a regression slope for the class effect. The assumptions for model 2 are the same as for model 1.
To explore the possibility that the relationships between gender and the dependent variable, and age and the dependent variable, were different in different classrooms, we added equations for the slopes of gender and age. Model 3 is shown below: Where b 1j and b 2j are the regression slopes for gender and age respectively in classroom j, b 1 and b 2 are mean regression coefficients across classrooms, c 1 to c 3 are regression slopes for classroom effects, and u 1j to u 2j are classroom residuals. It is assumed that the classroom residuals have a multivariate normal distribution with mean vector 0 and covariance matrix W: If we substitute the bottom two equations for b 1j and b 2j into the top equation, we can see that the effect of CMD on the relationships between the dependent variable and the individual variables (gender and age) will be estimated by two interaction terms. Table 1 presents descriptive statistics for victimisation, bullying, age, and collective moral disengagement. As shown, most participants reported low values on the Victimisation and Bullying scales (skew Bul ¼3.68, skew Vic ¼3.14). Pairwise correlations for class variables (class means for victimisation and bullying) are presented in Table 2. As shown, both victimisation and bullying were positively correlated with collective moral disengagement on a class level. In other words, the prevalence of bullying and victimisation were less in classrooms with lower levels of collective moral disengagement.  Table 3 displays estimates and standard errors from analyses in R for models 1, 2 and 3, with dependent variables victimisation and bullying, respectively. All variables, except gender, are grand mean centred. Effect sizes in Table 3 are standardised (for all quantitative variables) or partially standardised (for gender) regression coefficients (see Lorah, 2018). They can thus be interpreted as the expected change in the number of standard deviations in the dependent variable, followed by a one standard deviation change (or difference in gender) in the independent variable.

Victimisation
As shown by the intraclass correlation (ICC), 4.6% of the variation in victimisation scores was between classrooms. In addition, the variance between classrooms was significant (p<.001, from a likelihood ratio test). There were no significant effects of gender or age in either model. The class variable was added in model 2, and there was a positive effect of collective moral disengagement on victimisation. Classrooms with higher levels of collective moral disengagement scored higher, on average, on victimisation. This effect was still significant when interaction terms were added (in model 3). In addition, the variance of the gender slope was significant (p¼.014, from a likelihood ratio test), indicating that the relationship between gender and victimisation was different in different classroom.  .051 Note. ÃÃ p<.01, ÃÃÃ p<.001. The variance of the gender-(model 3: Bullying) and age-slopes (model 3: victimisation, bullying) were estimated close to zero, and were omitted for estimation purposes.

Bullying
As presented in Table 3, 5.1% of the total variation in bullying scores was between classrooms. In addition, the variance between classrooms was significant according to a likelihood ratio test (p<.001). Boys scored higher on bullying, and this effect was still significant as more variables were added (in models 2 and 3), and classrooms with higher levels of collective moral disengagement tended to score higher on bullying (in models 2 and 3).

Discussion
Bullying is an inhumane  or immoral behaviour (Romera et al., 2019) that takes place and is embedded in a social context (Hymel et al., 2015;Pepler et al., 2010;Salmivalli, 2010). Further research on how group characteristics might counteract or undermine moral standards and facilitate bullying among schoolchildren would therefore expand our knowledge on why bullying occurs in school, considering that children in general condemn bullying by referring to the harm it causes (Thornberg, 2010;Thornberg et al., 2016Thornberg et al., , 2017.
The current study adopted the social cognitive framework (Bandura, 1997(Bandura, , 2000(Bandura, , 2002(Bandura, , 2016 and focused on classroom collective moral disengagement (Bandura, 2016;Gini et al., 2015). The framework assumes that behaviours are situated in social contexts through an interplay or a reciprocal influence between behaviour, personal factors, and external environment (the so-called triadic codetermination process; Bandura, 1997Bandura, , 2016. In accordance with our hypotheses, the bivariate correlation analyses showed that students who belonged to classrooms with less collective moral disengagement were indeed less likely to be victimised by bullying or engaged in bullying perpetration, and vice versa. Moreover, when controlling for gender and age at the individual level and including collective moral disengagement in the same model at the classroom level, gender and collective moral disengagement contributed significantly to explain the variance of bullying perpetration. In contrast, when this model was tested with bullying victimisation as the dependent variable, only collective moral disengagement significantly explained the variance in victimisation. Regarding classroom collective moral disengagement, the current findings were consistent with the few studies that have found that aggression , bullying perpetration (Bj€ arehed et al., 2019;Kollerov a et al., 2018;Thornberg et al., 2019) and bullying victimisation (Kollerov a et al., 2018) were more frequent in school classes characterised by higher levels of collective moral disengagement (i.e. where peer group members collectively tended to justify bullying behaviour, minimise its consequences, or blame the victim). Thus, the present study contributed important empirical support for the extremely small but growing body of research showing how collective moral disengagement at the classroom level is linked to bullying perpetration and victimisation. Together with the previous findings from Bj€ arehed et al. (2019), Gini et al. (2015), Kollerov a et al. (2018) and Thornberg et al. (2019), our findings suggested that peer aggression/bullying is situated in the social context as classroom collective moral disengagement undermines the moral standards among the students, and thus facilitates peer aggression and bullying.
Regarding gender, the current findings were also in line with the literature by demonstrating that boys were more engaged in bullying perpetration than girls (Cook et al., 2010;Mitsopoulou & Giovazolias, 2015), whereas there were no significant gender differences in bullying victimisation (Cook et al., 2010). To our knowledge, the present study is the first to test whether collective moral disengagement at the classroom level interacts with gender at the individual level to help explain the variances of bullying perpetration and victimisation among schoolchildren. No cross-level interaction effects were found, suggesting that the main effects of classroom collective moral disengagement on bullying perpetration and victimisation may not be influenced by the gender of the student who is bullying or the student who is bullied.

Limitations and future directions
Some limitations of this study should be noted. First, the cross-sectional design does not allow us to draw definitive conclusions about the direction of the effects. Therefore, the multilevel models in the present findings need to be further tested with longitudinal data. Given social cognitive theory's assumptions about the interplay between environmental, individual, and behavioural influences (Bandura 1997(Bandura , 2016, future research needs to examine possible bidirectional relationships between classroom collective moral disengagement and bullying perpetration as well as between classroom collective moral disengagement and bullying victimisation over time. A second limitation was the reliance on self-report data in the current study, since this might result in underreporting when students are asked questions about their experiences of bullying. The term 'bullying' is a negative value-loaded word (Felix et al., 2011), and research has shown that students tend to underreport their bullying experiences when the questionnaire is asking global questions of general bullying (e.g. 'How often have you been bullied') as compared with specific questions of concrete behaviours (Huang & Cornell, 2015;Vaillancourt et al., 2010). In order to decrease the risk of underreporting, we used a bullying perpetration and victimisation scale that included aggression, power imbalance and repetition but without mentioning the term 'bullying', and by using specific behavioural items instead of global questions. Still, self-report data are vulnerable to careless responses, memory distortion, responses influenced by social desirability, and intentionally exaggerated responses (Cornell & Bandyopadhyay, 2010). Future research may adopt a peer nomination procedure to test the current multilevel models further.
The victimisation and bullying scales were skewed with most participants reporting few instances of victimisation and bullying. This skewness was expected as this is typical for bullying and victimisation variables (e.g. Breivik & Olweus, 2015). It may have affected significance tests. Our sample was fairly large, however, which may indicate that this is not a big concern. While the low variability in the scores on the bullying scale decreased statistical power, which would have made more it difficult to detect differences between participants, this also enhances confidence in the significant associations that were obtained.
Finally, the sample was not diverse in terms of ethnicity (only 6% of the students had a foreign background) and grade level. Although it represented a wide range of socio-geographic and socio-economic locations, the sample was confined to Sweden. As a result, these findings cannot be generalised to other grade levels, ethnic groups, or cultural contexts. Future research should examine these research questions with students at different grade levels and in different ethnic, cultural and geographical settings.

Practical implications
The current results added to the recent literature showing the importance of considering the role of class-level characteristics in the bullying process (for a review, see Saarento et al., 2015). Relatedly, growing evidence is also available that including group variables into school-based anti-bullying programs can add value to educational and preventive efforts and has potential to increase the likelihood of being successful. For example, some anti-bullying programs (e.g. KiVa, see Herkama & Salmivalli, 2018;Salmivalli, & Poskiparta, 2012;NoTrap! program, see Palladino et al., 2016) aim at changing not only the individual behaviour, but also the group dynamics and norms that may support or counter bullying perpetration, as well as the general climate in which pupils live and learn.
More specifically, the fact that collective moral disengagement at the class level may influence the likelihood of bullying perpetration and victimisation within a classroom should lead school psychologists, teachers, and other school staff to raise students' awareness of these disengaging mechanisms and help them recognise and possibly avoid their activation. Too often these mechanisms are shared within the peer group (Bandura, 2016;White et al., 2009), and thus linked to negative behaviours such as bullying. For example, moral justification is often based upon a group's or friends' valued goal, and euphemistic labelling is easily detectable in the common language early adolescents use to talk about bullying. Moreover, diffusion of responsibility is stronger if many students are present and support the bullying or remain passive, and blaming the victim can sometimes be so common among group members to result in scapegoating. Therefore, actively involving students who are bystanders and making them aware of their role in shaping the group moral norms and climate may be an effective intervention strategy to lessen the negative impact of collective moral disengagement within a classroom.

Disclosure statement
No potential conflict of interest was reported by the author(s).