Longitudinal link between moral disengagement and bullying among children and adolescents: A systematic review

ABSTRACT

Moral disengagement (MD) refers to social-cognitive distortions that allow individuals to sidestep the self-regulatory processes that normally prevent immoral conduct. MD has been linked to bullying in childhood and adolescence, but most studies have used a cross-sectional design. Less is known about the longitudinal link. Therefore, the aim of this review was to examine the longitudinal relationship between MD and bullying among school-aged children and adolescents. The focus included individual MD, classroom collective MD, MD loci, traditional bullying, and cyberbullying. Eligibility criteria resulted in a final sample of 16 peer-reviewed research articles. A large part of these studies examined and found that MD predicts subsequent bullying. Only a few studies have investigated whether bullying predicts subsequent MD, and the findings were mixed. Most studies used a short-term longitudinal approach and, in particular, longitudinal studies on the link between MD loci and bullying and the link between classroom collective MD and bullying were scarce.

KEYWORDS:

• Systematic review
• longitudinal
• bullying
• cyberbullying
• moral disengagement

Bullying is usually defined as unwanted aggressive behaviour repeatedly directed against an individual who is less powerful in relation to the perpetrator(s) (Hellström et al., 2021). A distinction can be made between traditional bullying (offline bullying), which occurs face to face and in physical real-life settings, and cyberbullying (online bullying), which is conducted via computers, mobile phones, and other electronic devices (Smith & O’higgins Norman, 2021). As can be deduced from its definition, bullying is undoubtedly an unfair and immoral behaviour (Romera et al., 2019). Even children themselves judge bullying to be a severe transgression and wrong, independent of school rules, by referring to the harm it causes the victim (e.g., Thornberg et al., 2017). Despite this, children can become bullies or engage in bullying perpetration for several reasons (Smith & O’higgins Norman, 2021). So the question remains: how can they do that if they condemn bullying in general? The prevalence of bullying in schools indicates a troublesome gap between moral standards and actions. To examine and explain this gap, a growing number of researchers have turned to Bandura’s (2016) social-cognitive theory of moral agency and, in particular, to his proposed concept of moral disengagement.

Social-cognitive theory of moral agency and moral disengagement

To understand how children, adolescents, and adults can engage in behaviours that run counter to moral standards and still maintain their self-respect and feel good about themselves, Bandura (1999, 2016) introduced the concept of moral disengagement (MD). This refers to social-cognitive distortions that allow individuals to sidestep the self-regulatory processes that normally prevent immoral conduct. MD is a multidimensional construct that includes four loci. The first locus is cognitive restructuring (behaviour locus), which means that the individual cognitively reconstrues the harmful behaviour or turns it into an acceptable or good behaviour through moral justification, euphemistic labelling, or advantageous comparison. The second locus is minimizing one’s agentive role (agency locus), which means that the individual obscures or detaches themselves from personal responsibility for the harmful behaviour through displacement or diffusion of responsibility. The third locus is distorting consequences (effects locus) and refers to minimizing, ignoring, or disputing the harmful effects of the immoral behaviour. The fourth locus is victim attribution (victim locus), which includes dehumanizing and blaming the victim.

MD is a phenomenon that not only occurs at the individual level but can also be widespread and shared across group members as a characteristic of a group. Collective MD refers to a group-level phenomenon of perceived shared beliefs of MD generated by the group dynamics (Bandura, 2016). In the school context, classroom collective MD refers to the classroom-level aggregation of student perceptions of the degree to which MD is shared among their classmates (classroom members) in the school class or classroom unit (Gini et al., 2015).

Moral disengagement and bullying perpetration

In accordance with social-cognitive theory, research has shown that children and adolescents who score higher in MD are more inclined to engage in traditional bullying perpetration (for meta-analyses, see Gini et al., 2014; Killer et al., 2019), and cyberbullying perpetration (for meta-analyses, see Chen et al., 2017; Killer et al., 2019; Zhao & Yu, 2021). From a developmental perspective, social-cognitive theory (Bandura, 2016) states that changes in MD and immoral behaviour occur as a gradual, reciprocal process over time. Children who initially bully others may increasingly engage in MD to avoid negative self-sanctions, which allows them to continue their bullying perpetration with fewer and fewer feelings of guilt and remorse. Thus, bullying perpetration may predict MD over time. Simultaneously, these children may continue to activate MD mechanisms while bullying others, which in turn allows them to increase their bullying behaviour in the future. Thus, MD may predict bullying behaviour over time. Therefore, this developmental aspect of MD and its association with bullying among children and adolescents is a relevant topic to study from a developmental psychology perspective.

A drawback of past research is that most studies have adopted a cross-sectional design. In their meta-analyses, Gini et al. (2014), Killer et al. (2019), and Zhao and Yu (2021) noted this as a limitation of the research field and in their meta-analyses. There is, however, a small but growing body of longitudinal studies on MD and bullying. Many of these studies have been published in very recent years and therefore were not available, or sufficient in number, for inclusion in past systematic reviews. Cricchio et al. (2021) conducted a recent systematic review on MD and cyberbullying roles and included 33 articles focusing on MD and cyberbullying perpetration. They found that most of these studies showed a positive cross-sectional association between the two variables, but emphasized that longitudinal studies are rare. Only one longitudinal study (Sticca et al., 2013) was included in their systematic review, addressing the link between MD and cyberbullying perpetration. Thus, close examination of longitudinal studies is missing.

The main rationale for the present systematic review is that research on the longitudinal link between MD and bullying is necessary to empirically test Bandura’s social-cognitive theory of MD. According to its so-called triadic codetermination, ‘human functioning is a product of the interplay of personal influences, the behavior individuals engage in, and the environmental forces that impinge on them’ (Bandura, 2016, p. 6). Personal influences include cognitive and emotional processes and states, goals, attitudes, and values (e.g., individual MD), and Bandura assumes a reciprocal influence between intra-psychological factors and behavioural engagement. Applied to the link between MD and bullying perpetration, if the theory is true regarding its assumption of triadic codetermination, not only should MD predict bullying behaviour, but bullying behaviour should also predict MD. In other words, the association should be bidirectional.

In addition, previous systematic reviews have neither examined how different MD loci are related to bullying, nor investigated the link between collective MD and bullying among children and adolescents (Chen et al., 2017; Gini et al., 2014; Killer et al., 2019; Zhao & Yu, 2021). Studies on MD loci and bullying, as well as studies on collective MD and bullying have not been sufficient to address these aspects in previous meta-analyses. Although MD is a multidimensional construct, the vast majority of studies have examined it as a global construct in relation to bullying perpetration. Still, Bandura’s (2016) argues that the subconstructs of MD can operate differently depending on the activity, behaviour, or aspects of life. Therefore, from a social-cognitive theoretical perspective, it is possible that only certain MD loci are longitudinally related with bullying behaviour. Research on this topic may advance and refine social-cognitive theory of MD and offer evidence to schools to guide them in tailoring their bullying prevention and interventions by targeting those MD loci that are relevant to combat bullying.

Further, only a few studies have examined classroom collective MD and shown that it is associated with a higher prevalence of bullying perpetration (Bjärehed, 2022; Bjärehed et al., 2021b; Kollerová et al., 2018; Thornberg et al., 2021; Thornberg, Wänström, & Hymel, 2019). The social-cognitive theory (Bandura, 2016) also assumes that MD can operate at both individual and collective levels. With reference to the triadic codetermination, classroom collective MD is a part of the environmental influences in school, and longitudinal studies examining the link between collective MD and bullying can further test the social-cognitive theory of MD.

Current study

The aim of the current study was to examine the longitudinal relationship between MD and bullying among school-aged children and adolescents. A significant advantage offered by longitudinal studies is that they allow developmental psychologists to study change, stability, and development regarding any psychological variable over time, but also to investigate the directionality of relationships between psychological variables and, thus, whether a certain psychological variable predicts another particular psychological variable over time. This makes it possible to test the social-cognitive theory of MD with reference to its assumption of triadic codetermination. In this article, the term bullying refers to bullying perpetration or behaviour, and excludes other possible involvement or participant roles in bullying, such as bullying victimization, defending, and other bystander behaviours. The present systematic review attempted to answer two major questions:

1. Does MD predict bullying?

2. Does bullying predict MD?

Taken together, these two questions explored directionality, including the possibility of bidirectionality, in the longitudinal association between MD and bullying. With reference to the triadic codetermination stated in the social-cognitive theory (Bandura, 2016), I expected that MD and bullying would predict each other over time, and thus a bidirectional association between the two variables. In accordance with the social-cognitive theory, MD and aggressive behaviour such as bullying are mutually affecting each other over time. Therefore, I expected that an increase in MD would be linked with an increase in bullying, while a decrease in MD would be associated with a decrease in bullying. Accordingly, an additional question is whether change in MD is linked with change in bullying over time. If longitudinal studies used MD loci as variables, or if they used collective MD as a variable, the questions above were applied to these variables as well.

Methods

To examine the questions above, I carried out a systematic review (Moher et al., 2009). The high degree of heterogeneity in longitudinal study design, MD and bullying constructs, measurements, and statistical procedures, in combination with partial study sample overlaps, lack of sample size information in some waves in a few studies, and the small final sample of empirical articles, ruled out a meta-analysis. A systematic review was, therefore, conducted in order to present key findings.

Eligibility criteria

Articles were included if they: (a) included measures of MD (as a uni- or multi-dimensional construct at the individual or collective level) and bullying perpetration (i.e., traditional bullying, cyberbullying, or a subcategory or form of traditional bullying), (b) examined longitudinal associations (changes or predictions) between MD and bullying, (c) included a sample of school-aged children or adolescents, (d) were written in English, and (e) were empirical articles published in peer-reviewed journals. Exclusion criteria were applied to articles without a longitudinal design, articles with a sample not consisting of school-aged children or adolescents, articles that did not measure MD, articles that did not measure bullying, articles that did not report longitudinal associations between MD and bullying, articles not written in English, and grey literature.

Search strategy

An initial search was conducted in a set of electronic databases (ERIC, PsycINFO, PubMed, Scopus, and Web of Science) using the search terms (‘moral disengagement’) AND (bullying OR cyberbullying) AND (longitudinal). The search was not limited in relation to a certain timespan; rather, a full database search was conducted. The database search was completed in November 2022 and resulted in 107 records. After eliminating duplicates, 40 articles remained.

In the next step, the articles were screened by reading their abstracts. Articles were only selected if their abstracts indicated a longitudinal design in which the link between MD and bullying had been examined. This step resulted in 16 articles. Examples of reasons for exclusion were: lack of longitudinal design; the use of broader constructs than – or other constructs instead of – bullying or cyberbullying, such as aggressive behaviour, violence, cyberaggression, antisocial behaviour, defending, pro-bullying bystander behaviour, passive bystanding, and bullying victimization; the use of other constructs instead of MD, and a sample involving young adults instead of children or adolescents.

The remaining empirical articles were read thoroughly, which resulted in one additional exclusion (MD was embedded in a broader construct termed ‘moral deficiencies’ and could not be parcelled out to investigate its longitudinal relationship with bullying). Thus, 15 studies met the inclusion criteria. Literature review section and reference lists of the studies were examined. One additional article (Obermann, 2013) met the inclusion criteria and, therefore, included in the review. It was not detected in the previous systematic search because the word longitudinal was not present in title, abstract or keywords. Thus, 16 studies met the inclusion criteria and constituted the final sample (Figure 1 shows the flow diagram of the search strategy and article selection).

Figure 1. Flow diagram of article selection process for review.

Results

General characteristics of the included studies

Nearly all of the 16 studies reported short-term longitudinal findings. The most common timespan was a period of one year. The longitudinal approach covered six months in one study, one year in 10 studies, 18 months in three studies, two years in one study, and three years in one study. Furthermore, eight studies included two waves or measurement time-points, six studies had three waves, and two studies had four waves. In terms of geographical coverage, four studies were conducted in Spain, four in Sweden, two in Italy, two in China, one in the U.S.A, one in Switzerland, one study in Denmark, and one in Cyprus.

In the majority of studies, MD was presented in the findings as a unidimensional construct at the individual level (14 studies). It was also presented as a unidimensional construct at the classroom level (i.e., classroom collective MD) in two studies, and as a multidimensional construct representing the four loci (i.e., cognitive restructuring, minimizing agency, distorting consequences, and victim attribution) at the individual level in three studies. In two studies, friends’ MD was included in the analyses.

Eleven studies measured and analysed traditional bullying as a distinct construct, four measured and analysed cyberbullying as a distinct construct, and one measured and analysed verbal bullying as a distinct construct. Two studies used a bullying variable that included but did not make a distinction between traditional bullying and cyberbullying (see Table 1). The latter was, however, much less represented in the measures: Georgiou et al. (2020) used a 10-item scale where only one item represented cyberbullying; Wang et al. (2017) used 5 items where only one of these items represented cyberbullying. I have therefore treated them as measuring traditional bullying in the analysis. All in all, 14 studies measured traditional bullying (11 with a distinct traditional bullying scale; two with a measure that mostly represented traditional bullying; and one with a measure that represented verbal bullying, which is a subcategory of traditional bullying), whereas only four studies explicitly measured cyberbullying.

Table 1. Characteristics of the 16 included longitudinal studies.

Reference	Design	Sociocultural characteristics of the sample	Sample size, gender, and age	Measurements	Results
Bjärehed, (2022)^a	1 year with 2 waves	Swedish sample; lower to upper-middle class; rural areas, small towns, medium-sized and large cities; 18–19% with immigrant background	The same 1,048 (54% girls) students in both waves. T1: M = 11.5 (SD = 0.3) T2: M = 12.6 (SD = 0.3)	MD: 18-item Moral Disengagement in Peer Victimization Scale (self-report) Collective MD: 18-item Classroom Collective Moral Disengagement Scale (student report) Traditional bullying: 11-item Linköping Bullying Perpetration Scale (self-report) Cyberbullying: 6-item Linköping Cyberbullying Perpetration Scale (self-report)	Correlation: MD at T1 was positively related with trad. bullying (r = 0.30***) and cyberbullying (r = 0.19***) at T2; trad. bullying at T1 was positively related with MD at T2 (r = 0.43***); cyberbullying at T1 was positively related with MD at T2 (r = 0.35***) (individual level). Collective MD at T1 was positively related with class mean of trad. bullying (r = 0.60*) and class mean of cyberbullying (r = 0.47*) at T2; class mean of trad. bullying at T1 was positively related with collective MD at T2 (r = 0.49*); class mean of cyberbullying at T1 was positively related with collective MD at T2 (r = 0.52*) (classroom level). Regression: Multilevel growth modelling: Individual MD (Est = 0.19***) and classroom collective MD (Est = 0.20***) at T1 were positively associated with trad. bullying across the grades. Change in individual MD was associated with change in trad. bullying (Est = 0.16***). Change in classroom collective MD was not significantly associated with change in trad. bullying. Individual MD (Est = 0.13***) and classroom collective MD (Est = 0.12***) at T1 were positively associated with cyberbullying across the grades. Change in individual MD was associated with change in cyberbullying (Est = 0.13***). Change in classroom collective MD was not significantly associated with change in cyberbullying.
Bjärehed et al. (2021a)^a	3 years with 4 waves	Swedish sample; lower to upper-middle class; rural areas, small towns, medium-sized and large cities; 18% with immigrant background	T1: 1,626 (48% girls), M = 10.55 (SD = 0.35) T2: 1,685 (49% girls), M = 11.55 (SD = 0.33) T3: 1,487 (50% girls), M = 12.58 (SD = 0.35) T4: 1,748 (53% girls), M = 13.65 (SD = 0.36)	MD: 18-item Moral Disengagement in Peer Victimization Scale (self-report) Verbal bullying: 3 items measuring verbal bullying from 11-item Linköping Bullying Perpetration Scale (self-report)	Correlation: For boys, MD at T1 correlated positively with verbal bullying at T2 (r = 0.16**), T3 (r = 0.17**) and T4 (r = 0.16**); MD at T2 correlated positively with verbal bullying at T3 (r = 0.29**) but not significantly at T4 (r = 0.08); MD at T3 correlated positively with verbal bullying at T4 (r = 0.24**). For girls, MD at T1 correlated positively with verbal bullying at T2 (r = 0.17**), T3 (r = 0.16**) and T4 (r = 0.13**); MD at T2 correlated positively with verbal bullying at T3 (r = 0.30**) and T4 (r = 0.28**); MD at T3 correlated positively with verbal bullying at T4 (r = 0.27**). For boys, verbal bullying at T1 correlated positively with MD at T2 (r = 0.31**), at T3 (r = 0.19**) and T4 (r = 0.14**); verbal bullying at T2 correlated positively with MD at T3 (r = 0.37**) and T4 (r = 0.19**); verbal bullying at T3 correlated positively with MD at T4 (r = 0.29**). For girls, verbal bullying at T1 correlated positively with MD at T2 (r = 0.27**), at T3 (r = 0.19**) and T4 (r = 0.21**); verbal bullying at T2 correlated positively with MD at T3 (r = 0.32**) and T4 (r = 0.21**); verbal bullying at T3 correlated positively with MD at T4 (r = 0.35**).
					Regression: Multilevel growth modelling: Across the four waves, students with higher MD were more inclined to engage in verbal bullying, and their verbal bullying trajectories were steeper. Changes in MD was also linked with concurrent changes in verbal bullying.
Caravita et al. (2014)^b	1 year with 2 waves	Italian sample; missing data from 25.8% regarding SES; of the remaining participants: 25.8% from low-middle SES, 50.8% from middle SES, 23.5% from middle-high SES	369 (42% girls) representing two age groups (late childhood, early adolescence): T1: 133 (42% girl), 9–10 years, M = 9.36 (SD = 0.52); 236 (41% girls), 11–14 years, M = 11.91 (SD = 0.73) T2: No information provided	MD: 14-item Moral Disengagement Scale for elementary school students; 24-item Moral Disengagement scale for middle school students (self-report) Traditional bullying: Participant Roles Questionnaire (peer nomination)	Correlation: Bullying at T1 was positively related with MD at T2, both in late childhood (r = 0.26**) and in early adolescence (r = 0.26**). RSiena: In early adolescence (not in late childhood), students became more similar to their friends’ level of MD, while their bullying behaviour at T1 did not affect change of MD over time.
Falla et al. (2022)^c	1 year with 3 waves	Spanish sample; sociodemographic information not provided	T1: 3,617 (49% girls), M = 13.64 (SD = 1.36) T2: 3,068 (50% girls), M = 14.08 (SD = 1.39) T3: 2,066 (51% girls), M = 14.01 (SD = 1.07).	MD: 24-item Moral Disengagement Scale (self-report) to measure the four loci om MD. Traditional bullying: 7-item bullying perpetration subscale from the Spanish adaption of the European Bullying Intervention Project Questionnaire (self-report).	Correlation: Cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T1 correlated positively with bullying at T2 (r = 0.34**, r = 0.19**, r = 0.21**, r = 0.28**) and at T3 (r = 0.30**, r = 0.16**, r = 0.24**, r = 0.24**). Cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T2 correlated positively with bullying at T3 (r = 0.27**, r = 0.20**, r = 0.23**, r = 0.23**). Bullying at T1 correlated positively with cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T2 (r = 0.37**, r = 0.20**, r = 0.29**, r = 0.32**) and T3 (r = 0.34**, r = 0.21**, r = 0.27**, r = 0.30**). Bullying at T2 correlated positively with cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T3 (r = 0.33**, r = 0.16**, r = 0.25**, r = 0.28**). Mediation modelling: When tested whether the four MD loci at T2 predicted bullying at T3 while controlling for victimization at T1, only cognitive restructuring significantly predicted greater bullying (β = 0.14***)^.d
Falla et al., (2021)^c	1.5 year with 3 waves	Spanish sample; sociodemographic information not provided	T1: 2,360 (50% girls), M = 13.58 (SD = 1.13) T2: 2,031 (51% girls), M = 13.97 (SD = 1.04) T3: 1,810 (51% girls), M = 14.50 (SD = 1.05).	MD: 24-item Moral Disengagement Scale (self-report) to measure the four loci om MD. Traditional bullying: 7-item bullying perpetration subscale from the Spanish adaption of the European Bullying Intervention Project Questionnaire (self-report).	Correlation: Cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T1 correlated positively with bullying at T2 (r = 0.29**, r = 0.21**, r = 0.25**, r = 0.25**) and at T3 (r = 0.26**, r = 0.17**, r = 0.18**, r = 0.20**). Cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T2 correlated positively with bullying at T3 (r = 0.32**, r = 0.21**, r = 0.26**, r = 0.26**). Bullying at T1 correlated positively with cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T2 (r = 0.35**, r = 0.16**, r = 0.25**, r = 0.29**) and T3 (r = 0.34**, r = 0.19**, r = 0.20**, r = 0.26**). Bullying at T2 correlated positively with cognitive restructuring, minimizing agency, distorting consequences, and victim attribution at T3 (r = 0.29**, r = 0.25**, r = 0.20**, r = 0.27**).
					Mediation modelling: The four MD loci (T2) were tested as mediators in two models with bullying (T1) as the predictor and cognitive empathy or affective empathy (T3) as the outcome. Of interest in this review are the links between bullying and MD loci. Bullying at T1 predicted cognitive restructuring (b = 0.30***/0.31***), minimizing agency (b = 0.15**/0.16***), distorting consequences (b = 0.23***/0.24***), and victim attribution (b = 0.27***/0.29***) at T2 in both models. However, the four MD loci at T1 and bullying at T2 were not included in the models^.d
Georgiou et al. (2020)	1 year with 3 waves	Cypriot sample; sociodemographic information not provided	T1: 558 (50% girls), 14–17 years, M = 15.3 (SD = 0.69) T2: no further information T3: no further information	MD: 13-item Moral Disengagement Scale (self-report). Only administrated in T2. Bullying: 10-item subscale on bullying perpetration from the Olweus Bully/Victim Questionnaire. Administrated in T1 and T3.	Correlation: Bullying at T1 correlated positively with MD at T2 (r = 0.27***), which in turn correlated positively with bullying at T3 (r = 0.34***). SEM: MD at T2 predicted bullying at T3 (b = 0.20*). However, MD at T3 and bullying at T2 were not measured and therefore, not included in the model.
Marín-López et al. (2020)	1 year with 2 waves	Spanish sample; sociodemographic information not provided	T1: 1,033 (48% girls), 11–17 years, M = 13.66 (SD = 1.64) T2: 534 (50% girls), 12–18 years, M = 14.10 (SD = 1.33)	MD: 16-item Moral Disengagement through Technology Questionnaire (self-report) Cyberbullying: 11-item cyberperpetration subscale from 22-item European Bullying Intervention Project Questionnaire (self-report)	ANOVA: MD at T1 was higher among students who were cyberbully/victims at T2 as compared to those who were uninvolved at T2. Among the four MD loci, cognitive restructuring and victim attribution were higher at T1 among cyberbully/victims than uninvolved at T2. Regression: Logistic regression: Among the four MD loci, only cognitive restructuring at T1 predicted being a cyberbully (OR = 1.22), while none of them predicted being a cyberbully/victim, one year later (T2)^.d
Obermann, (2013)	1 year with 2 waves	Danish sample; the majority of the students were from middle class families, some from higher-middle class and some from lower-middle class families; from schools in a medium-large city; schools located in town centre and in suburbs outside the centre	The same 567 (48% girls) students in both waves. T1: M = 12.6 T2: no further information	MD: 32-item Moral Disengagement Scale (self-report). Traditional bullying: Two measures: (a) a Danish short version of the Olweus Bully/Victim Questionnaire (self-report); (b) a peer-nomination bully/victim questionnaire.	Correlation: MD at T1 correlated positively with self-reported bullying (boys: r = 0.33**; girls: r = 0.35**) and peer-nominated bullying (boys: r = 0.22**; r = 0.21**) at T2. Self-reported bullying (boys: r = 0.24**; girls: r = 0.31**) and peer-nominated bullying (boys: r = 0.15**; r = 0.20**) at T1 correlated positively with MD at T2. Regression: self-reported bullying at T1 (β = 0.14*) and change in self-reported bullying (β = 0.23**) was positively associated with change in MD. Peer-nominated bullying at T1 and change in peer-nominated bullying were not linked with change in MD (all bullying variables were included in the same model). ANOVA: Four categories of students based on self-report bullying data. Stable bullies and new bullies increased in MD compared to nonbullies and desisted bullies (all p < .05). t-test with Bonferroni corrections: None of the four groups changed significantly in MD from T1 to T2.
Romera et al. (2021)^c	18 months with 4 waves	Spanish sample; sociodemographic information not provided	T1: 2,790 (49% girls), M = 13.15 (SD = 1.09) T2: 2,553 (50% girls), M = 13.61 (SD = 1.13) T3: 2,362 (51% girls), M = 14.03 (SD = 1.05) T4: 2,361 (50% girls), M = 14.55 (SD = 1.06)	MD: 24-item Moral Disengagement Scale (self-report). Traditional bullying: 7-item bullying perpetration subscale from the Spanish adaption of the European Bullying Intervention Project Questionnaire (self-report).	Correlation: MD at T1 was positively related with bullying at T2 (r = 0.29***), T3 (r = 0.30***), and T4 (r = 0.25***). MD at T2 was positively related with bullying at T3 (r = 0.27***) and T4 (r = 0.26***). MD at T3 was positively related with bullying at T4 (r = 0.31***). Bullying at T1 was positively related with MD at T2 (r = 0.35***), T3 (r = 0.32***), and T4 (r = 0.32***). Bullying at T2 was positively related with MD at T3 (r = 0.31***) and T4 (r = 0.30***). MD at T3 was positively related with bullying at T4 (r = 0.30***). SEM: Cross-lagged panel modelling: MD predicted subsequent bullying from T1 to T2 (b = 0.10***), from T2 to T3 (b = 0.10***), and from T3 to T4 (b = 0.10***), and concurrent correlations between MD and bullying within each wave (0.28–0.30***). Bullying did not significantly predict subsequent MD at any timepoint. Thus, the model did not support a bidirectional longitudinal association between the two variables.
					Random intercept cross-lagged panel modelling: Between-person effects showed that students high on MD tended to have high on bullying across the period (b = 0.73***). Whitin-person effects showed no significant longitudinal associations between MD and bullying. Instead, the model showed that MD predicted subsequent MD, and bullying predicted subsequent bullying as well as concurrently associated with each other within each wave.
Sijtsema et al. (2014)^b	1 year with 2 waves	Italian sample; missing data from 25.8% regarding SES; of the remaining participants: 25.8% from low-middle SES, 50.8% from middle SES, 23.5% from middle-high SES; 7.4% immigrant background	369 (42% girls) representing two age groups (late childhood, early adolescence): T1: 133 (42% girl), 9–10 years, M = 9.36 (SD = 0.52); 236 (41% girls), 11–14 years, M = 11.91 (SD = 0.73) T2: No information provided	MD: 14-item Moral Disengagement Scale for elementary school students; 24-item Moral Disengagement scale for middle school students (self-report) Traditional bullying: Participant Roles Questionnaire (peer nomination)	Correlation: Neither MD (r = 0.18) nor friends’ MD (r = 0.16) at T1 were significantly related with bullying at T2 in late childhood. MD at T1 was positively related with bullying at T2 in early adolescence (r = 0.15*). The link between friends’ MD at T1 and bullying at T2 was non-significant in early adolescence (r = 0.14). RSiena: Friends’ MD were linked to bullying in early adolescence, but the effect differs between gender. Boys were more prone to decrease in bullying over time when their friends had higher level of MD, while they were more likely to increase in bullying if their friends were low on MD. In contrast, girls were more inclined to decrease in bullying when their friends were low on MD.
Sticca et al. (2013)	6 months with 2 waves	Swiss sample; sociodemographic information not provided	T1: 835 (49% girls), M = 13.2 (SD = 0.64) T2: 820; no further information	MD: 2 hypothetical bullying scenarios with 5 and 6 follow-up items, separately Cyberbullying: 6-item cyberbullying scale (self-report) Traditional bullying: 6 items from a 12-item traditional bullying scale (self-report)	Correlation: MD at T1 was positively related with cyberbullying (r = 0.24***) and traditional bullying (r = 0.32***) at T2. MD at T2 was not measured. Regression: Logistic regression: MD at T1 did not significantly predict cyberbullying at T2 when controlling for traditional bullying, cyberbullying, and other variables (e.g., rule-breaking behaviours, cyber victimization and empathic concern) at T1. No regression models were performed to test predictions of traditional bullying and MD at T2.
Teng et al. (2020)	18 months with 3 waves	Chinese sample; 46–47% from rural areas, and the rest from cities and towns; variation in household income; 34–35% from ethnic minorities; around half from middle schools, and half from high schools.	T1: 2,997 (50% girls), M = 14.9 (SD = 1.53) T2: 2,781 (51% girls), M = 15.55 (SD = 1.54) T3: 2,413 (53% girls), M = 16.07 (SD = 1.55)	MD: 32-item Moral Disengagement Scale (self-report) Traditional bullying: A 6-item bullying perpetration subscale from a Chinese version of the Olweus Bully/Victim Questionnaire where cyberbullying items were excluded in this study	Regression: Adolescents’ bullying perpetration was positively predicted by MD at both between-person (SE = 0.25**) and within-person (SE = 0.12**) levels across the 18 months. The within-person effects indicate that longitudinal changes in MD within students were positively associated with changes in their bullying behaviour over time.
Thornberg, Wänström, & Hymel(2019)^a	1 year with 2 waves	Swedish sample; lower to upper-middle class; rural areas, small towns, medium-sized and large cities; 18% with immigrant background	The same 1,250 (52% girls) students in both waves. T1: M = 10.55 (SD = 0.34) T2: M = 11.55 (SD = 0.32)	MD: 18-item Moral Disengagement in Peer Victimization Scale (self-report) Collective MD: 18-item Classroom Collective Moral Disengagement Scale (student report) Traditional bullying: 11-item Linköping Bullying Perpetration Scale (self-report)	Correlation: MD at T1 was positively related with bullying at T2 (r = 0.22***); bullying at T1 was positively related with MD at T2 (r = 0.38***) (individual level). Collective MD at T1 was positively related with class mean of bullying at T2 (r = 0.37***); class mean of bullying at T1 was positively related with collective MD at T2 (r = 0.46***) (classroom level). Regression: Multilevel regression: Individual MD (Est = 0.19***) and classroom collective MD (Est = 0.10**) at T1 were positively associated with bullying across the grades. Change in individual MD was associated with change in bullying (Est = 0.22***). Change in classroom collective MD was not significantly associated with change in bullying.
Thornberg, Wänström, Pozzoli, & Hong (2019)^a	1 year with 2 waves	Swedish sample; lower to upper-middle class; rural areas, small towns, medium-sized and large cities; 19% with immigrant background	The same 1,354 (52% girls) students in both waves. T1: M = 10.54 (SD = 0.35) T2: M = 11.54 (SD = 0.32)	MD: 18-item Moral Disengagement in Peer Victimization Scale (self-report) Traditional bullying: 11-item Linköping Bullying Perpetration Scale (self-report)	SEM: MD at T1 predicted bullying at T2 (b = 0.07***), and bullying at T1 predicted MD at T2 (b = 0.20***), when controlling for both MD stability and bullying stability over time. These predictions also held for boys (b = 0.05* and b = 0.17***) and girls (b = 0.08*** and b = 0.21***).
Wang et al. (2017)	1 year with 3 waves ^e	U.S. sample; 80.2% White, 7.1% African American, 5.4% Hispanic, 2.4% Asian Americans, 3.2% missing values	T1: 1,180 (53% girls), M = 12.2 (SD = 1.29) T2: no further information T3: no further information	MD: 32-item Moral Disengagement Scale (self-report) Bullying: 5 items from the Pacific-Rim Bullying Measure (self-report)	SEM: Although all associations between MD and bullying was significant at the same wave, MD at T1 did not predict bullying at T2, while MD at T2 predicted bullying at T3 (d = 0.194***). Comparisons between models favoured the directionality where MD is a precursor to bullying.
Yang et al. (2022)	2 years with 3 waves	Chinese sample; 26.1% had a mother and 27.4% had a father with a bachelor’s degree or above; variation in household income	T1: 2,407 (50% girls), 11–16 years, M = 12.75 (SD = 0.58) T2: 2,103, no further info. T3: 1,871, no further info.	MD: 32-item Moral Disengagement Scale (self-report) Cyberbullying: 9-item Cyberbullying Questionnaire (self-report)	Correlation: MD at T1 was positively related with cyberbullying at T2 (r = 0.21***) and T3 (r = 0.20***); cyberbullying at T1 was positively related with MD at T2 (r = 0.22***) and T3 (r = 0.19***); MD at T2 was positively related with cyberbullying at T3 (r = 0.25***); cyberbullying at T2 was positively related with MD at T3 (r = 0.24***). SEM: Cross-lagged panel model: MD positively predicted cyberbullying from T1 to T2 (b = 0.11***) and from T2 to T3 (b = 0.08**). Cyberbullying at T1 positively predicted MD at T2 (b = 0.11***), but cyberbullying at T2 did not predict MD at T3.

Note. *p < .05, **p < .01, ***p < .001 (according to the reports in the studies).

a. These four studies are part of the same longitudinal project, and thus, there were some sample overlaps between them.

b. The same sample.

c. These three studies are part of the same longitudinal project, and thus, there were some sample overlaps between them.

d. In these studies, other labels have partially been used to term the four loci, but to avoid confusion, I have used the terms used in the introduction of this review in a consistent way.

e. The study includes three waves (Fall, Spring and Fall semesters) with an interval of 6 months between each wave. Thus, the total time period was approximately 1 year.

Key findings

Moral disengagement and traditional bullying

The longitudinal link between (individual) MD and traditional bullying was the most studied association. Eight of the 16 studies reported bivariate correlations between MD and subsequent traditional bullying. All of them reported positive correlations, including the 3-year longitudinal study on verbal bullying, showing significant positive correlations between MD and subsequent verbal bullying for both boys and girls, but with one exception for boys between MD at T2 and verbal bullying two years later at T4. In one of the studies, the correlation was only significant in early adolescence, but not in late childhood. Eight studies reported bivariate correlations between traditional bullying and subsequent MD. All of them found significant positive correlations (including the 3-year longitudinal study showing that verbal bullying was consistently related to subsequent bullying between all timepoints and for both boys and girls).

When the analyses became more complex and included a set of control variables, the general findings were less consistent. MD predicted itself, and traditional bullying predicted itself over time, which revealed a certain stability in the levels of MD and traditional bullying. School-aged children and adolescents who scored high in MD tended to continue to be high in MD, and those who scored high in traditional bullying were more inclined to engage in high levels of traditional bullying over time. Furthermore, the between-person results and the concurrent positive associations between MD and traditional bullying within waves indicate that children and adolescents who scored high in MD in any of the waves were also prone to displaying high levels of traditional bullying in the same wave.

The mixed results concerned whether MD and traditional bullying predict each other over time. Most studies using regression analyses found that MD predicted later traditional bullying or that changes in MD predicted changes in traditional bullying over time. In the studies using SEM statistics, MD and traditional bullying predicted one another bidirectionally in a Swedish sample involving pre-adolescents. However, bidirectional associations were not supported in either a Spanish sample or an American sample of adolescents. In their studies, MD predicted subsequent traditional bullying, while traditional bullying did not predict subsequent MD. In the Spanish study, when further controlling for the between-person effects, which showed that students high on MD tended to score high on traditional bullying across the period, MD neither predicted later traditional bullying, nor did traditional bullying predict later MD.

Moral disengagement loci and traditional bullying

Only two studies examined the longitudinal associations between MD loci and traditional bullying. Bivariate correlations showed that all four MD loci correlated positively with subsequent traditional bullying. At the same time, traditional bullying correlated positively with all four subsequent MD loci. When including all four MD loci in the same mediation models, only cognitive restructuring predicted traditional bullying, while traditional bullying predicted all four MD loci.

Moral disengagement and cyberbullying

Four studies examined the longitudinal association between MD and cyberbullying. Three of them reported bivariate correlations. All of them reported positive correlations between MD and subsequent cyberbullying. Two of the studies reported bivariate correlations between cyberbullying and subsequent MD. Both found positive correlations. In one study, a regression analysis revealed that change in MD was positively linked with change in cyberbullying over time. In another study, outcomes from an ANOVA showed that MD at T1 was higher among students who were cyberbully/victims at T2 compared to those who were uninvolved, which may suggest that MD predicted whether students were engaged in cyberbullying as both bullies and victims one year later. In a third study using SEM statistics, MD consistently predicted subsequent cyberbullying, while cyberbullying only partially predicted MD. According to the logistic regression model in a fourth study, MD at T1 did not predict cyberbullying at T2.

Moral disengagement loci and cyberbullying

There was only one study examining whether the four loci were longitudinally associated with cyberbullying. According to the ANOVA findings, cognitive restructuring and victim attribution were higher at T1 among cyberbully/victims than those uninvolved at T2. Findings from logistic regression showed that only cognitive restructuring predicted being a cyberbully one year later on.

Collective moral disengagement, traditional bullying and cyberbullying

Two studies were found that examined whether classroom collective MD was longitudinally related with bullying. One study investigated the link with traditional bullying, while the other included both traditional bullying and cyberbullying. Bivariate correlations at the classroom level showed that collective MD correlated positively with class means of subsequent traditional bullying and cyberbullying. At the same time, a higher class mean of traditional bullying and cyberbullying correlated positively with subsequent classroom collective MD. In regression models, initial classroom collective MD was positively associated with traditional bullying and cyberbullying across time, but changes in classroom collective MD were not linked with changes in either traditional bullying or cyberbullying.

Discussion

Most research on the link between MD and bullying among school-age children and adolescents in the literature is cross-sectional (Chen et al., 2017; Cricchio et al., 2021; Gini et al., 2014; Killer et al., 2019; Zhao & Yu, 2021). The common underlying assumption in these studies is that cognition influences behaviour, or more specifically that MD precedes and predicts bullying perpetration, since the latter is most often used as a dependent variable. However, due to the cross-sectional design, directionality cannot be determined. The number of longitudinal studies is currently still low, even though it has recently increased. The present systematic review identified 16 longitudinal studies with some sample overlaps.

In line with the underlying assumption in the cross-sectional studies, the longitudinal studies that adopted regression analyses either tested and confirmed that MD predicted bullying over time, or that change in MD was related to change in bullying over time. Together with some of the other longitudinal findings, these regression models not only support the underlying assumption that MD predicts bullying in previous cross-sectional studies but also confirms Bandura’s (2016) social-cognitive theory of MD: ‘People do not usually engage in harmful conduct until they have justified to themselves the morality of their actions’ (Bandura, 2016, p. 49). In this way, MD precedes and facilitates bullying behaviour by providing distorted moral cognitions that justify and rationalize the behaviour, and thus allowing children and adolescents to engage in such harmful actions without any feelings of guilt or remorse but with a maintained positive self-image.

According to the triadic codetermination built into the social-cognitive theory (Bandura, 2016), cognition will not only influence behaviour but will also be influenced by behaviour. The more children engage in harmful conduct, such as bullying, the stronger is their psychological need to develop the ability to use MD to explain away their negative behaviour, and thus to protect their positive self-image and to avoid self-sanctions. All studies that reported bivariate longitudinal correlations demonstrated that bullying perpetration was positively correlated with subsequent MD. However, only four studies using SEM statistics tested whether MD predicted bullying and whether bullying predicted MD at the same time, and the results were inconsistent. While MD usually predicted subsequent bullying, the outcomes of the investigations of whether bullying predicted subsequent MD were more mixed.

Possible explanations for these inconsistencies might be found in cultural, methodological, and age-related differences across the studies. For example, changes in MD and bullying behaviour might be a gradual, reciprocal process over time among children (Thornberg, Wänström, Pozzoli, & Hong, 2019), while the tendency to morally disengage, the inclination to bully, and the interrelation between MD and bullying may become more stable and less developmentally changeable in adolescence (Romera et al., 2021; Sticca et al., 2013). Nevertheless, empirical findings in the current review indicating that bullying behaviour predicts subsequent MD call for further theoretical development of social-cognitive theory to better understand and explain this possible developmental trajectory, how, when and why greater engagement in bullying perpetration results in an increase in MD over time.

Very few studies have explored whether different MD loci predict later bullying when they are included in the same model. The findings so far suggest that only cognitive restructuring predicts bullying (Falla et al., 2022; Marín López et al., 2020), while bullying seems to predict higher levels of all four MD loci (Falla, Romera, & Ortega-Ruiz, 2021). A possible explanation could be that, since cognitive restructuring involves moral justification (i.e., justifying negative behaviour by referring to higher purposes or worthy ends), bullying can then be interpreted by the perpetrator as serving a good purpose (Bandura, 1999, 2016). By means of cognitive restructuring, the perpetrator convinces themselves that their behaviour is socially acceptable, or even desirable, admirable, or good (Bandura, 2016), which might make this MD locus extremely powerful in deactivating moral self-regulation and self-sanctioning.

Only two studies in the review examined the longitudinal association between classroom collective MD and bullying. While the initial level of classroom collective MD was related to higher levels of bullying across the studied time period, changes in this classroom-level variable were not associated with changes in bullying over time (Bjärehed, 2022; Thornberg, Wänström, & Hymel, 2019). A possible explanation might be that collective MD had already been developed into less changeable and more stable group characteristics because the students had been together in the same classroom unit for a semester or longer by the first wave in these studies. However, longitudinal studies on the link between MD loci and bullying and the link between classroom collective MD and bullying are scarce.

Limitations and future directions

A first limitation is the study heterogeneity, which calls for caution when comparing and synthesizing their findings. There is a lack of consistency in measuring MD and bullying. Different self-report scales have been used to measure MD, also in terms of delimited MD to various domains (e.g., MD items regarding peer victimization, bullying, antisocial behaviour, cyber aggression, and hypothetical bullying scenarios). While bullying was mostly measured with self-report scales, different measures were adopted across the studies (with various degrees of validation), and only a few of the studies included peer reports. Self-reported data are vulnerable to a range of biases, including social desirability, careless marking, recall and perception biases. Since most studies only gathered self-reported data, their findings might have produced inflated variable associations due to shared method variance.

Further, there was a heterogeneity across the studies regarding research design and statistical analysis. Whereas MD and bullying were measured and analysed in all waves in some studies, that was not the case in other studies. Regression analysis allowed for examining one directionality (whether MD predicted bullying), or if change in MD was associated with change in bullying, but did not investigate the opposite directionality (whether bullying predicted MD), and could not answering the question of bidirectionality. When mediation modelling was adopted in two of the studies, the dependent variables were not controlled for by including previous values. Finally, four of the five studies using SEM statistics measured and analysed both MD and bullying in all waves, and were therefore able to truly test bidirectionality. Because of the low number of these studies and with reference to their overall mixed findings, the outcome of the present review is insufficient to draw valid conclusions about bidirectionality. More longitudinal studies designed to investigate bidirectionality between MD and bullying are needed.

A second limitation is that the review only identified and analysed 16 studies from seven countries, including sample overlaps between some of the studies, which clearly limits the generalizability of the findings. Despite the strength of a longitudinal design, very little research has examined the longitudinal link between MD and bullying. Thus, a higher proportion of future research should adopt a longitudinal approach in order to explore developmental processes, trajectories, stability, and changes in MD and bullying during childhood and adolescence, and whether MD and bullying predict each other over time.

A third limitation of this review is that almost all the examined studies adopted a short-term longitudinal design (one year or less in 11 of the 16 studies) with 2–3 waves. Longitudinal approaches that extend across several years of childhood and adolescence, and include additional waves or timepoints of measurement, would have allowed for a more extensive inquiry into developmental changes and trajectories. Future longitudinal studies on MD and bullying that run throughout early childhood and into late adolescence, and even beyond, would help us to better integrate longitudinal associations between MD and bullying with current theories of social and moral development (cf., Killen & Smetana, 2023; Smith, 2022). This will also require more studies investigating whether other individual and contextual variables moderate or mediate the longitudinal association between MD and bullying.

A fourth limitation is the surprisingly small number of longitudinal studies investigating MD loci and collective MD, considering that MD is a multidimensional construct and can operate at both the individual and collective levels (Bandura, 2016). Moreover, a further limitation is that only four of the 16 longitudinal studies included cyberbullying, which calls for further longitudinal studies on cyberbullying. Although there is a correlation and an overlap between traditional school bullying and cyberbullying in the literature (Estévez et al., 2020), the two phenomena are not identical and future studies should examine possible differences between them in how they are longitudinally related with MD, collective MD, and different MD loci. An additional limitation is that the systematic review was conducted by only one researcher and coder.

A final limitation is the eligibility criteria of this review that restricted the sample to articles written in English and published in peer-reviewed journals. Thus, significant research findings in the grey literature or published in other languages might have been overlooked, which further limits the generalizability. Nonetheless, the current systematic review is, to the best of my knowledge, the first to examine the literature on the longitudinal link between MD and bullying among school-aged children and adolescents. While it provides some insights, it underlines the need for further longitudinal studies.

Disclosure statement

No potential conflict of interest was reported by the author.