Bullying was first defined by Olweus (1994) as a form of school violence inflicted on a student by one or more of his or her peers, recurrently over an extended duration. School bullying is characterized by a significant power imbalance between the aggressor and the victim (e.g., physical strength, social status, or perceived authority) and it comprises a complex social dynamic of victims, bullies, and bystanders. Aggressors or bullies seek to establish dominance and control over their victims to enhance their social standing and reputation among their peers. Depending on their response towards bullying, bystanders have been conceptualized as aggressor-supporter, pro-victim (e.g., defender), and those that adopt a passive position or outsiders (Salmivalli, 2010).

Numerous studies consistently highlight the adverse long-term effects of bullying on all individuals involved. Victims often endure feelings of isolation, diminished self-acceptance, emotional trauma, and depression and are at increased risk of attempting suicide. For bullies, the positive reinforcement they receive from peers may encourage them to evaluate aggressive responses favorably, in ways that may impair their long-term abilities to recognize moral transgressions, reduce their capacity for compassion, and perpetuate involvement in abusive dynamics as valid ways to relate to others, potentially leading to the development of conduct disorders (Gini & Pozzoli, 2013; Moore et al., 2017; Schoeler et al., 2018; Swearer & Hymel, 2015). Being involved as a bystander of bullying contributes to greater psychological distress, but compared to aggressor supporters, defender bystanders appear to display victimization-like consequences such as greater depressive and social anxiety levels (Midgett & Doumas, 2019; Wu et al., 2016).

Understanding the psychological mechanisms that underlie the different roles of students in bullying situations is crucial for designing effective anti-bullying strategies. Drawing from social cognitive theory (Bandura et al., 1996), it is proposed that cognitive mechanisms such as moral disengagement (MD) may neutralize self-censorship in the face of moral transgressions like bullying. Indeed, several reports (Gini et al., 2014; Killer et al., 2019) consistently demonstrate a significant positive correlation between MD and bullying perpetration in cross-sectional studies. Additionally, an analysis of longitudinal data conducted by Thornberg (2023) suggests that the utilization of MD mechanisms predicts subsequent involvement in bullying perpetration, rather than bullying leading to later MD. Interventions aiming to reduce MD have been linked with a decline in transgressive behavior (Tolmatcheff et al., 2022; Wang et al., 2017).

The use of MD in victims and bystanders is not as clear as in the case of perpetrators. Meta-analytic findings in Killer et al. (2019) indicated a significant relationship between victimization and MD, though the effect size was so small (i.e., r = 0.08) that the association lacked practical significance. On the other hand, no relationship was detected between bystanding and MD, but greater MD was associated with decreased defending behavior in bystanders. Given the significance of bystander responses in mitigating school bullying, it makes sense to intervene MD mechanisms to foster pro-victim responses as part of anti-bullying intervention strategies (Midgett & Doumas, 2019).

The study of MD has been central to understanding and intervening in transgressive behaviors. However, the type of MD mechanisms that underlie such behaviors is restricted by the lack of consensus in measuring this construct. The MD Scale (MDS) is hypothesized to follow a bifactor model that includes eight MD mechanisms: moral justification, euphemistic language, advantageous comparison, displacement of responsibility, diffusion of responsibility, distortion of consequences, attribution of blame, and dehumanization. In later presentations of the theory and practice, the mechanisms were clustered into four greater categories as moral justification, euphemistic language, advantageous comparison, and minimization of one’s agentive role (Bandura et al., 1996).

To date, confirmatory factor analyses (CFA) across the literature report a range of factor structures to the MDS, including eight (Qi, 2019; Romera et al., 2023), six (Boardley & Kavussanu, 2007), four (Marín-López et al., 2019; Newton et al., 2016; Romera et al., 2021), and three (García Vázquez et al., 2019; Rubio-Garay et al., 2017), while other studies favor a single-structure model as the best fit (Çapan & Bakioğlu, 2016; Caprara et al., 2009; Concha-Salgado et al., 2022; Ettekal & Ladd, 2020; Luo & Bussey, 2022; Paciello et al., 2008; Pelton et al., 2004). Furthermore, the distribution and number of items recommended within each subconstruct varies across studies; some studies discard the original 32-item MDS and suggest a brief version with 10 items (Concha-Salgado et al., 2022), 11 items (García Vázquez et al., 2019), 15 (Çapan & Bakioğlu, 2016), 16 (Luo & Bussey, 2022; Marín-López et al., 2019), while others propose 24 (Romera et al., 2023) or 22 items (Newton et al., 2016). Refer to Table 1 for a summary of CFA findings.

Table 1 Summary of psychometric validations of the MDS
Full size table

The variability of factor structures and distribution of items that compose the MDS limits the interpretation of findings regarding the specific MD mechanisms represented in the scale. Furthermore, this variability makes it challenging to compare findings across studies and build upon existing knowledge in the field. Various solutions may shed light on external factors that could interfere with the measurement of these mechanisms. The stability of the construct over time, linguistic issues in item adaptations across languages, and cultural conditions may be affecting the way MD mechanisms are perceived across different samples. Moral constructs are highly influenced by cultural differences, and it is possible that these cultural variations are reflected in the psychometric differences (Ardila, 2005).

By focusing on the cognitive processes that allow individuals to detach from the moral consequences of their aggressive actions, intervention programs can challenge and reshape the underlying beliefs and justifications behind bullying actions in various settings, such as schools, community, and online environments (Tolmatcheff et al., 2022). To ensure a robust instrument for measuring MD, this study pursued two objectives of a statistical and theoretical nature. Firstly, it aimed to test the psychometric properties of the MDS in a sample of Colombian youth. No specific hypothesis was formulated to allow for a statistical exploration of the best-fit structure for this culturally specific sample. Secondly, the study aimed to compare the use of MD mechanisms across the roles that partake in bullying within school contexts, namely, victims, bullies, and non-participating students. It was hypothesized that compared to victims and non-participating students, bullies would exhibit significantly greater use of MD mechanisms. Pro-victim bystanders were anticipated to use significantly less MD mechanisms than pro-aggressor or outsider (passive) bystanders.

Method

Participants

This study employed a convenience sampling method involving 375 Colombian adolescents, with 49% of them being females, consistent with the sample size used in the initial validation of the MDS (Bandura et al., 1996). The participants were enrolled in public schools located in the Caribbean region of Colombia and had an age range spanning from 11 to 17 years (M = 13.3; SD = 1.69). The questionnaire was administered in paper form to students from grades 7th to 11th during school hours. Out of the 420 questionnaires received, 45 were discarded due to incomplete responses, leaving 375 valid responses for analysis. The sample size was guided by the N:q rule of thumb (Kyriazos, 2018), suggesting a ratio of 10:1 to 20:1 (participants per items in the scale) for CFA.

This study obtained approval from the Ethical Committee of Universidad del Norte, as documented in Minutes of Meeting Number 141. The Ethical Committee functions as an Institutional Review Board (IRB) responsible for overseeing research involving human participants. All adolescents and their legal guardians provided informed consent before participating in this study. The research strictly adhered to established ethical guidelines, encompassing contact procedures, the solicitation of informed consent, ensuring participant anonymity, and affording individuals the prerogative to withdraw from the study, as recommended in the Declaration of Helsinki (revised in Taipei in 2016).

Measures

Sociodemographic

Participants provided information about their age, sex, and school grade.

Moral Disengagement

The 32-item Spanish version of the Moral Disengagement Scale (MDS; Rubio-Garay et al., 2017) assessed participants' tendencies to disengage from moral self-sanctions regarding transgressive behavior. Respondents rated items on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). Total scores were calculated by averaging item responses, with higher values indicating a greater propensity for moral disengagement. The 32 statements comprising the scale measure eight moral disengagement mechanisms grouped into four clusters, as previously mentioned. The psychometric analyses of the MDS first presented a one-dimensional structure that explained 16% of the variance of the data, with good reliability score or Cronbach's alpha coefficient (α) in a sample of American adolescents α = 0.82 (Bandura et al., 1996). A second validation of the scale by the same research group of authors (Caprara et al., 2009) confirmed the one-dimensional structure (that explained 27% of the variance), associated to an excellent reliability score in a sample of Italian adolescents (α = 0.92). This study found a unidimensional factor structure with an adequate scale reliability score of α = 0.66 and ω = 0.68.

Bullying and Bullying Roles

The bullying questionnaire used in the Spanish national commission report on school violence in secondary schools (del Barrio et al., 2008; Van der Meulen et al., 2019) is based on the Olweus Bully-Victim Questionnaire (Solberg & Olweus, 2003) and aims to measure bullying and social exclusion. This instrument starts off with a brief introduction about peer-on-peer aggression, and the different roles adopted by individuals: perpetrators of the transgression (i.e., bullies), the targets of the maltreatment (i.e., victims), and non-participating students that may have taken the role of bystanders during bullying situations. Subsequently, participants are asked to rate on 4-point Likert scale (0 = never; 3 = always) how frequently they had engaged in bullying and cyberbullying situations as perpetrators (e.g., “I insult other students,” “I send insulting e-mails”), and how often they had been the target of bullying themselves (e.g., how often “my peers insult me,” “other pupils send me insulting e-mails”) since the school year started. Total scores are calculated by averaging the items portraying traditional bullying and cyberbullying perpetration (13 items and nine items respectively) and those covering bullying and cyberbullying victimization (13 and nine items, respectively). The scale reliability scores in this study were good, both for traditional (α = 0.85) and cyberbullying (α = 0.89) perpetration, and for traditional (α = 0.80) and cyberbullying victimization (α = 0.88).

The bullying questionnaire by del Barrio et al. (2008) permits the identification of student’s roles within bullying scenarios based on reported frequency of perpetration and victimization. Pupils with a score ≥ 4 in the victimization items are labeled as “victim.” Similarly, scores ≥ 4 in bullying perpetration indicate the role of “bully.” Respondents coded as both “victim” and “bullies” are considered “bully-victims” and scores ≤ 4 in the two perspectives (i.e., bullying perpetrator and bullying victimization) are coded as non-participating students (Van der Meulen et al., 2019). An additional item asking participants what they did when they witnessed bullying situations in their school (i.e., What do you do when another student is being consistently bullied?) was included to assess their role as bystanders. Responses were coded in two groups: (i) pro-victim responses included defending the victim or calling a teacher for help, and (ii) outsiders or pro-aggressor, who were students that take no action when witnessing bullying situations or are supportive of the aggressor.

Statistical Analysis

The psychometric properties of the MDS were assessed through confirmatory factor analysis (CFA) following the Hair et al. (2014). During this analysis, CFA was conducted on the entire sample, utilizing the rival model approach to identify the optimal factorial solution in line with the findings of the literature review, with the aim of mitigating any potential factorial biases. Four different models were included in the analysis (Table 2). The first model (model 1) was a higher-order factor based on Bandura and colleagues (1996). The second model (model 2) followed the three-factor solution proposed by Rubio-Garay et al. (2017). The third model (model 3) depicted a four-factor solution, as suggested by Romera and colleagues (2023). Finally, a fourth model (model 4) featured an eight-factor solution as per the research by Qi (2019).

Table 2 Confirmatory factor analysis
Full size table

To determine the most suitable model, we considered a range of absolute, incremental, and parsimony indices, including I2 (Chi-square), mean squared error of approximation (RMSEA), goodness of fit index (GFI), adjusted goodness of fit index (AGFI), comparative fit index (CFI), and Tucker-Lewis index (TLI). The evaluation was carried out based on the cutoff criteria proposed by Hu and Bentler (1999). Additionally, the internal consistency of the selected model was assessed by calculating composite reliability using both Cronbach’s alpha (α) and McDonald’s omega (ω) coefficients (Peters, 2014).

Following the CFA, an exploratory factor analysis (EFA) was employed to gain insights into the scale’s distribution and stability within our sample. This EFA process involved various critical steps, such as assessing the determinant (d) in the inter-item correlation matrix, calculating the Kaiser-Mayer-Olkin adequacy index (KMO), and conducting Bartlett’s test of sphericity, all of which were essential for evaluating the factor structure. The EFA allowed to explore three factorial structures generated through mechanical procedures, psychometric indices, and statistical methods. The reason for generating different models arises from the need to reduce the bias of factorial interdependence in the exploration. Among the statistical tests, we considered the eigenvalues yielded by the elbow drop test, Horn’s parallel analysis (PA), optimal coordinates (OC), and the minimum average partial (MAP) test (O’connor, 2000; Pearson et al., 2013; Ruscio & Roche, 2012). The various employed techniques ensure the robustness and reliability of the factor analysis results.

Initially, the Kaiser criteria (elbow method) recommends retaining only factors with eigenvalues greater than 1 while discarding the rest. This approach ensures that factors retained explain more variance than individual variables alone. The advantage of this criterion is its ability to produce concise solutions, reducing the number of factors to a minimum while capturing most of the data’s variance. This helps prevent overfitting and the creation of numerous, challenging-to-interpret factors. However, a disadvantage is that with extensive datasets, it might overestimate the number of factors and retain some insignificant ones. Subsequently, more advanced methods than the elbow drop test were applied within the EFA framework. Horn’s parallel analysis (PA) utilizes statistical simulation to estimate the optimal number of factors through an optimization algorithm. Its primary advantage is the ability to compare observed eigenvalues with generated eigenvalues, reducing the risk of over-extraction of factors. Additionally, the optimal coordinate (OC) technique combines the elbow drop test with Horn’s parallel analysis. It involves calculating a series of linear equations to determine whether observed eigenvalues exceed predicted values, thus maximizing model performance. Finally, the minimal average partial (MAP) method, was implemented. It is based on the matrix of partial correlations and determines the number of factors to retain by identifying the point at which the minimum mean of partial correlations is achieved. Importantly, this method appears to be less influenced by sample size and offers high accuracy in optimizing factor selection, making it the preferred technique for our analyses. The EFA was conducted under the ULS (unweighted least squares) method (Lloret-segura et al., 2014; Mîndrilă, 2010) with VARIMAX rotation. Factors were retained after a minimum of three items, and factor loadings ≥ 0.40 (Matsunaga, 2010).

To address the second objective of this study, which involved comparing the use of MD mechanisms among bullying roles, we conducted ANOVA (analysis of variance) followed by Bonferroni post hoc comparisons (p < 0.05). This enabled the comparison of mean scores for a selected MDS model across various student groups, namely those who assumed different roles as bullies, victims, and non-participating students. To compare mean MDS scores between two groups of bystanders, namely, pro-victims and neutral or pro-supporters, we employed a t-test. All the statistical analyses were performed with the software R in its version 4.1.1 (R Core Team, 2021), using the “fa” and “cfa” functions, which are available in the “psych” packages (Revelle, 2021) and “lavaan” (Rosseel, 2012). For modeling and estimating the models, we selected the “DWLS” parameter, as it provides better adjustments when dealing with categorical data and does not penalize variable non-normality (Li, 2016).

Results

Confirmatory Factor Analysis (CFA)

All four models that were tested exhibited acceptable fit indicators in alignment with existing literature guidelines (Hair et al., 2014; Hu & Bentler, 1999). Model 1, model 2, model 3, and model 4 consistently achieved incremental fit indices exceeding the threshold of > 0.90 and showed low values in terms of absolute fit indicators, such as RMSEA. These results collectively affirm that each of the models explored possesses a robust fit and a high level of validity with respect to the factor structures delineated in the existing literature (Table 2). Only a more detailed comparison of goodness-of-fit indices among these models revealed that the four-dimensional solution (model 3) slightly outperformed the others in the following indices, including X2(246) = 305.26, CFI and TLI = 0.94, RMSEA = 0.02, GFI = 0.92, AGFI = 0.90, and ECVI = 1.11.

Exploratory Factor Analysis (EFA)

As aforementioned, an EFA was subsequently performed to understand the scale in this study’s specific sample. This analysis confirmed the underlying factorial structure (d < 0.001) and tests of sphericity and sampling adequacy revealed that the scale was suitable for further factorial analyses, KMO = 0.81, Bartlett’s χ2(496) = 2400.69, p < 0.05. A model proposing a higher-order factor that encloses eight mechanisms as per Bandura’s theory (1996) was discarded because it exhibited notably low factor loadings, no significant correlations, and a factor configuration in which each factor had fewer than three associated items (Lloret-Segura et al., 2014; Matsunaga, 2010).

The number of factors, as determined by both the parallel analysis (PA) and Kaiser criteria (elbow method), pointed towards a four-factor structure, encompassing 13 items (model a). In contrast, the optimal coordinates (OC) criterion suggested a 19-item solution (model b), while the minimum average partial (MAP) criterion advocated for a single-factor solution with 13 items (model c) (see Table 3 for details). It is worth noting that the four-factor structure, as confirmed by both PA–Kaiser and optimal coordinates (OC) criteria (models a and b), retained consensus regarding the number of factors; however, these models exhibited variations in item configurations. Given the reduced influence with respect to sample size, and the high accuracy in optimizing factor selection offered by the MAP criterion, the model with 13 items (model c) and unidimensional factor was deemed as more appropriate for explaining the data.

Table 3 EFA models and distribution of items
Full size table

Comparison of MD Mechanisms

Findings showed a moderate and positive correlation between MD scores of the single-factor solution with 13 items (model c) and bullying perpetration in traditional (r = 0.27, p < 0.001) and cybernetic forms (r = 0.31, p < 0.001). Victimization of traditional and cyberbullying also showed significant and positive correlations with MD scores (r = 0.28, p < 0.001; r = 0.24, p < 0.001, respectively). See details in Table 4. Most participants (n = 155) stated that in the face of a bullying situation they engage in pro-victim actions such as defending the victim or seeking help from teachers, while the remaining students would remain passive (n = 23) or support the aggressor (n = 14). Mean comparisons of MD scores did not show significant differences between pro-victim bystanders and outsiders or pro-aggressor ones (p > 0.05).

Table 4 Correlation matrix (N = 192)
Full size table

Forty-eight (n = 48) participants met the criteria scores for “bullies” or “bully-victims,” a further 43 were identified as “victims,” and the remaining 101 as “non-participating students.” Mean comparisons of MD across three groups of bullying roles (e.g., bullies and bully-victims, victims, and non-participating students) indicated that MD levels were significantly greater in students that were bullies and bully-victims (M = 1.05, SD = 0.75), compared to those who fell under the category of “victims” (M = 0.55, SD = 0.56) or “non-participating students” (M = 0.49, SD = 0.53), and this difference was significant, F(2, 191) = 14.92, p < 0.00, η2 = 0.13. Further Bonferroni post hoc comparisons confirmed that bullies and bully-victims had a significantly greater use of MD mechanisms than victims and non-participating students (p < 0.001), but there was no significant difference in MD scores between non-participating students and victims (p > 0.05).

Discussion

This study aimed to investigate the psychometric properties of the MDS in Colombian youth and to compare the use of MD across bullying roles. While previous reports of the psychometric characteristics of the MDS support a higher-order factor model with eight MD mechanisms, as suggested in the original model of MD (e.g., Qi, 2019; Romera et al., 2023) CFA results also indicated acceptable fit indicators for three-factor, four-factor, and eight-factor structures. Furthermore, results of an EFA suggested that, for the studied sample, a shortened 13-item MDS with a unidimensional factor solution offered the best statistical fit for measuring MD in a broad sense. An analysis of the three roles in bullying situations showed that, compared to victims and non-participating students, bullies used significantly more MD mechanisms, which may suggest that intervening in the way pupils cognitively neutralize moral self-sanctions may help reduce harmful relational behaviors. Mean MD scores did not differ between bystanders who were pro-victim and those who did not intervene or were supportive of the aggressor.

EFA allowed to compare among various factor solutions in order to inquire structures suitable for the studied Colombian data. In this sense, the EFA served as a valuable complement to CFA especially when there was uncertainty about the best model solution for a culturally specific sample. An explanation for single-factor solution with 13 items suggested by the EFA could be that distinctive types of MD mechanisms were clustered together due to the perceived redundancy of some items, leading to overlapping interpretations of different dimensions within the scale. For example, items 1, 9, 10, and 17 all address hitting classmates to protect friends or family. While items 1, 9, and 17 belong to the mechanism of moral justification, item 10 theoretically falls under euphemistic language. A similar overlap may occur with items related to the attribution of guilt (items 8, 16, 24, 32) and the displacement of responsibility (items 5, 13, 21, 29), which justify immoral situations by attributing blame to external circumstances or individuals (e.g., parents, the school). These overlapping meanings may have resulted in high correlations among these items, making it challenging to identify distinct subconstructs within the factor structure of the instrument. Using measures of MD different to the MDS (Hymel et al., 2005; Thornberg & Jungert, 2014) or new culturally tailored MD scales could offer a more accurate depiction of this construct in different contexts.

In this study, the 13-item version of the scale proved highly effective in capturing the moral disengagement processes employed by Colombian children and adolescents to rationalize and justify aggressive behaviors comprehensively. The retained items shed light on how these young individuals perceive such behaviors as necessary or acceptable in specific situations. Notably, for this sample, the most robust items included statements that justified physical aggression as a valid means to defend or protect others (items 1, 9, 17), attributing wrongdoings to external factors such as parenting or school performance (items 8, 12, 13, 29, 32), and the belief that some people inherently deserve to be treated poorly based on who they are (items 7, 15, 31). The remaining items (27, 21) described situations in which individuals morally disengage by comparing their actions to those of others (e.g., “Taking from a store is not as bad as other things people do”). Among the items comprising the Moral Disengagement Scale (MDS), these particular ones appear to provide a more nuanced reflection of the culturally specific attitudes towards moral transgressions among the participants in this study (Ardila, 2005). The complete questionnaire and the validated short version can be found in English and Spanish as an Appendix to this section.

The short 13-item MDS showed a good scale reliability score, and positive association with measures of bullying perpetration and bullying victimization (del Barrio et al., 2008). Bullies showed greater use of MD mechanisms than their peers who were victims or non-involved students, thus supporting previous reports showcasing these differences (Gini & Pozzoli, 2013; Pozzoli et al., 2012; Thornberg et al., 2022). Contrary to expectations, students in pro-victim bystander roles did not show a lower use of MD than their peers that were outsiders or supporters of the aggressor (Midgett & Doumas, 2019). This result may be related to the broad classification of bystander roles used in the study. A more detailed depiction of bystanding roles would be necessary to further examine MD in relation to specific bystanding behaviors (e.g., defenders vs calling for help).

This study had several limitations that must be acknowledged. First, the convenient sampling strategy restricts the representativeness of these results to all Colombian adolescents since it did not consider factors like socioeconomic status or representation from rural areas. Future research would likely benefit from more rigorous sampling strategies to ensure a more diverse representation of the population. A second limitation was the reliance on self-report measures for both MDS and bullying. It would be valuable to incorporate additional sources of data, such as input from teachers and parents, or direct observations of bullying behaviors. Thirdly, the measure of bullying in del Barrio et al. (2008) used arbitrary cutoff points to distinguish between victims, bullies, and non-involved students. Furthermore, it did not differentiate bystanders as such, but rather as students that were not victims nor bullies. Future operationalization efforts may benefit from empirically supported criteria or a more precise delimitation of bullying roles, for instance by implementing cluster analyses to identify the specific profiles of students involved in bullying situations. Finally, the small sample size in the study did not provide sufficient statistical power to perform a more detailed comparison of MD scores across types of bystanders. Future research would be enriched by sampling strategies that can capture detailed depiction of how MD mechanisms are used by bystanders in different roles.

Despite the abovementioned limitations, the present study makes theoretical, and methodological contributions to attest the role of MD in transgressive behaviors: in the first place this study has provided valuable insights into the possible factorial structures of the Moral Disengagement Scale and offers a more precise instrument to assess adolescents’ justifications of bullying. However, the variability in the presence of certain mechanisms across different samples requires further investigation, with cultural factors likely playing a significant role in shaping moral disengagement patterns. Second, the study confirmed that in the context of school bullying, individuals engaging in bullying perpetration use more cognitive strategies to neutralize moral self-sanctions than their peers in the role of bystanders or victims. There were no differences in MD levels between bystander roles.

Considerations for Practice

Insights from this study can inform comprehensive recommendations for interventions addressing bullying in educational settings. Practitioners and policymakers within schools can enhance anti-bullying efforts by focusing on specific mechanisms identified, such as justifying physical aggression for defense, attributing wrongdoing to external factors, and subscribing to the belief that certain individuals inherently deserve mistreatment based on comparative assessments.

Fostering awareness among students, families, and community about the rational cognitive processes behind aggressive behaviors is also crucial. Open debates and group discussions can contribute to creating a school culture that values kindness, inclusivity, and respect, thereby reducing the likelihood of bullying instances (Swearer & Hymel, 2015). In this context, incorporating instruction on moral neutralization mechanisms serves as a complementary addition to educational content on conflict resolution and social skills in school curricula.

Educational institutions also play a pivotal role in contributing to research by implementing monitoring procedures for bullying incidents. As demonstrated by this study, it is imperative to conscientiously consider the perspectives of victims, bullies, and different types of bystanders. Such a meticulous approach is indispensable. Interventions, in turn, require prioritizing the cultivation of empathy, emotional regulation, and conflict resolution skills for individuals engaged in bullying behaviors and potential outsider bystanders. In the case of pro-victim bystanders and victims, interventions should strategically focus on fortifying support systems and empowering them to transition into active upstanders. This proactive paradigm exhibits considerable potential in disrupting the recurrent nature inherent in bullying dynamics.