It is essential to understand the role of specific social circumstances in cyberbullying among college students. Studies on the predictive factors of cyberbullying have shown that socioeconomic status, peer norms, and parental educational levels are significant influences (Bullo and Schulz, 2022; Lapierre and Dane, 2020; Uludasdemir and Kucuk, 2019). However, research is lacking on the risk factors of social circumstances related to cyberbullying behavior among college students (Peterson and Densley, 2017; Shaikh et al., 2021). A considerable number of young people are likely to face significant social challenges, such as perceived discrimination, due to being in the transitional phase of their social roles (Wu et al., 2022). For example, the rising cost of higher education is forcing many young people to rely on student loans to cover tuition fees and increasing their financial burden (Avery and Turner, 2012; Britt et al., 2017; Xiao et al., 2020). Meanwhile, as socioeconomic inequalities and structural barriers impede social mobility, the job market is becoming increasingly competitive, making it difficult for many students to secure satisfactory employment (Changjun and Liping, 2019; Chu et al., 2020). These factors contribute to a perception of social class discrimination among young people and lead them to a sense of greater inequality and differential treatment in their lives and career development. Facing severe social challenges, young people are more likely to use social networks covertly to express their dissatisfaction and compensate for what they feel is lacking in their offline social activities.

Frustration-aggression hypothesis (Berkowitz, 1989) provides theoretical support to understand the relationship between perceived discrimination and cyberbullying. Berkowitz (1989) argued that frustration can generate a tendency towards aggressive behavior because it instils aversive feelings and restricts people from achieving their goals. Even in situations not involving direct and personal discrimination, such as perceived disadvantages associated with socioeconomic status or race, people may still exhibit aggressive behavior. These arguments are supported by numerous empirical findings that perceived racial discrimination is positively related to aggressive behavior among youth (Burt et al., 2012; Hautala and Sittner, 2019). An empirical study by Bedrosova et al. (2022) demonstrated that both individual-based discrimination (e.g., related to physical characteristics such as height or weight) and group-based discrimination (e.g., linked to family origin) were significantly associated with increased cyberbullying behaviors among adolescents. Moreover, a longitudinal study indicated that socioeconomic status disadvantage can predict higher aggressive behavior 1 year later (Santiago et al., 2011). However, to the best of our knowledge, there is no direct empirical evidence for a longitudinal relationship between perceived discrimination and cyberbullying. To address this gap in the literature, the current study explores the association between perceived discrimination and cyberbullying, proposing the following hypothesis:

Studies have indicated that frustration, such as that stemming from discriminatory experience, can produce compliance with social norms rather than aggression (Leander et al., 2020), and others have suggested that the mechanism of the relationship between frustration and aggression needs further exploration (Berkowitz, 1989; Hautala and Sittner, 2019). The self-evaluation maintenance theory suggests that individuals strived to maintain a positive self-concept, and when their self-image was threatened (e.g., by perceived discrimination), they may resort to externalizing behaviors to restore their sense of self-worth (Tesser, 1988). As a global evaluation of a person’s own value and capacity, self-esteem may serve as a potential pathway to understand the complex relationship between perceived discrimination and cyberbullying. Low self-esteem is one of the outcomes most commonly associated with discrimination (Alsawalqa, 2021; Zhang et al., 2023). Moreover, Baumeister et al. (1996) suggested that aggression occurs when an individuals’ self-evaluation is threatened by negative evaluations from others, such as through discrimination. Self-esteem reflects an individual’s subjective assessment of their important social relationships. Respect and a sense of belonging from others were fundamental psychological needs (Kruglanski et al., 2023; Liu et al., 2024). When these needs are frustrated, particularly by external rejection or insult, the individual’s self-esteem was threatened, triggering feelings of anger and hostility towards others. Thus, individuals perceive a threat to their self-esteem, they may adopt defensive or aggressive behaviors to protect and recover it.

Research has indicated that perceived discrimination can damage self-esteem and then increase online bullying behavior (Lei et al., 2020). However, the relationship between self-esteem and cyberbullying is complex (Shi et al., 2017). On the one hand, studies have indicated a negative correlation between self-esteem and cyberbullying (Ding et al., 2018). On the other hand, other researchers contend that this correlation is positive, suggesting that individuals with higher self-esteem are more likely to engage in cyberbullying (Fan et al., 2019). As a trait-like individual characteristic, self-esteem comprises both stable and unstable components (Braun et al., 2021). This implies that the relationship between self-esteem and cyberbullying may be affected by other factors (Lei et al., 2020). Thus, the present study further explores the role of self-esteem in the relationship between perceived discrimination and cyberbullying, specifically by examining when threats to self-esteem become linked to cyberbullying. We propose the following hypothesis:

Self-compassion has been found to weaken the association between a frustrating situation and cyberbullying perpetration (Geng and Lei, 2021). Self-compassion, defined as compassion for one’s own suffering, is an effective approach to managing distressing thoughts and emotions and can promote mental and physical well-being (Neff, 2023). According to stress coping theory, individuals adopt different coping styles when confronted with stressful situations (Lazarus, 1993). As a positive coping strategy, self-compassion can mitigate aggression resulting from discrimination. College students can experience self-indifference and self-doubt due to unfair treatment (Wondra and McCrea, 2022), but self-compassion can help mitigate the negative effect of discriminatory experiences and protect their emotional well-being in tough social situations (Li et al., 2022). Accordingly, when the degree of negative emotion stimulated by a frustrating experience (e.g., discrimination) is reduced, an individual’s tendency towards aggression may also weaken. Furthermore, the association between damaged self-esteem and cyberbullying can also be moderated by self-compassion. Recent empirical studies have shown that self-compassion can decrease cyberbullying perpetration among college students and enhance the psychological functioning of young people with low self-esteem who encounter stigma or discrimination (Langford et al., 2022). In line with these previous findings, we propose that self-compassion buffers the direct association between perceived discrimination and cyberbullying and moderates the indirect association between perceived discrimination and cyberbullying through self-esteem.

Furthermore, the moderating role of different components of self-compassion remains underexplored in the literature. Recent studies have introduced a new framework that divides self-compassion into two components that distinguish the positive and reduced negative aspects of self-compassion: compassionate self-response (CS) (self-kindness, common humanity, mindfulness) and reducing uncompassionate responses (RUS) (self-judgment, isolation, over-identification) (Neff et al., 2018). Although CS and RUS have been shown to alleviate mental health problems and enhance healthy psychological functioning (Germer and Neff, 2013; Raes, 2011), RUS is more strongly linked to self-evaluation than CS (Neff et al., 2018). Individuals with low self-esteem often struggled with social connections, leading to poor social adaptation and increased aggression (Hirschi, 1969). Uncompassionate self-responding, characterized by dehumanization, allowed individuals to minimize the severity of their bullying behavior and avoid self-sanctions, increasing the likelihood of online aggression (Runions et al., 2019). Reducing uncompassionate self-responding is crucial for mitigating the impact of low self-esteem on online bullying.

Reducing uncompassionate responses is more likely to help break the vicious cycle of negative self-evaluation and self-criticism, which in turn deceases an individual’s aversive feelings and consequently reduces cyberbullying (Ding et al., 2018). An empirical study showed that CS and RUS differed in their moderating effects on the relationship between perceived discrimination and psychopathological outcomes (Li et al., 2022). To summarize, the present study examines the moderating effects of SC, RUS, and CS on the indirect and direct relationships between perceived discrimination and cyberbullying (the proposed model is shown in the Figure 1).

We distributed 1,000 questionnaires to eligible participants, and received 909 (90.9%) completed responses at the first survey. During the T2 data collection, 17 students failed to return their answers. Finally, data were collected from 892 university students (414 females and 478 males) across two survey waves with a one-year interval. We recruited eligible participants (full-time undergraduates aged 18 years and over) through the student affairs officers at a university in Guangdong Province, China. We excluded students who were on leave of absence or unable to provide informed consent. These criteria were chosen to ensure the study population represented actively enrolled undergraduates while minimizing potential biases associated with irregular academic status. The average age of the participants at the baseline survey was 20.47 years (SD = 1.37). The educational attainment of 26.7% of the participants’ fathers and 46.9% of their mothers was primary school or below. Their average family monthly income was RMB 6522. The research procedures performed in this study involving human participants were approved by the Institutional Review Board of the XXX University.

Participants were recruited through convenience sampling, with student affairs officers distributing invitations during university events to ensure accessibility. Retention strategies included clear instructions, regular follow-ups, practice credits for academic purposes, and assurances of confidentiality. Data collection occurred in September 2020 (T1) and September 2021 (T2), allowing for a 1-year interval to capture meaningful changes in variables like self-esteem while minimizing bias and fatigue. Data collection was conducted face-to-face during designated university activities for two waves (T1 and T2). To address missing data, we conducted a demographic and baseline analysis comparing participants who completed all three waves with those who dropped out, finding no significant differences. In present study, missing data were handled using full information maximum likelihood (FIML) estimation.

The demographic variables (age, gender, parents’ educational level, and household income) were assessed in the baseline survey. Moreover, the subjective socioeconomic status (SSES) of participants was measured by a visual scale with 10-rung ladders (Cheng et al., 2013).

Harman’s single factor test was conducted to examine whether common method bias was a serious issue in the present study. The results of an exploratory factor analysis showed that the extracted first component accounted for 23.57% of the total variance, which indicated that the common method bias was not severe.

Descriptive statistics were generated and correlation analyses performed to describe the demographic information of the participants and examine the bivariate correlations among the variables of interest. To further investigate the hypothesized moderated mediation model, a series of analyses were performed using the SPSS PROCESS macro (Hayes, 2013).

First, the mediating role of self-esteem in the relationship between perceived discrimination and cyberbullying was examined using the PROCESS macro (Model 4) and bootstrapping approaches (Hayes and Preacher, 2014). This model was chosen because it directly evaluates the indirect effect of the perceived discrimination on the cyberbullying via the self-esteem. The indirect relationship between perceived discrimination and cyberbullying through self-esteem was considered significant if the bootstrapped 95% confidence intervals (boot 95% CIs) did not include zero (Aiken et al., 1991).

Second, the moderated mediation effect was examined using the PROCESS macro (Model 15). This model was selected because it allows for the investigation of whether the mediation effect was conditional on a moderator. Demographic variables (age, gender, health, SSES) at T2 were controlled in the analysis. The conditional indirect effect analysis to examine whether the direct and indirect effect of perceived discrimination on cyberbullying significantly differed at high (mean + 1 standard deviation) and low (mean - 1 standard deviation) levels of each moderator (CS, RUS, SCS). Additionally, a simple slope analysis was conducted to explore the nature of the moderation models.

The descriptive information and bivariate correlations among the key variables are presented in Table 1. Perceived discrimination (PD) at T1 was positively associated with cyberbullying (CB) at T2 and negatively associated with self-esteem (S-E), CS, RUS, and SC at T2. The result suggests that individuals who perceive higher levels of discrimination are more likely to experience threats to their self-esteem and engage in cyberbullying.

Moreover, CB at T2 was negatively associated with S-E, CS, RUS, and SC at T2, while was not significantly associated with age and SSES. The finding indicates that higher levels of cyberbullying behaviors were associated with lower self-esteem and lower self-compassion. Additionally, the results of Spearman correlation analysis showed that CB was positively associated with gender (r = 1.115, p < 0.001; male = 1). The finding indicates that males were more likely to engage in cyberbullying than females.

Table 2 presents the results of the mediation analysis. After controlling for age, gender, health, and SSES, the results of Model 1 showed that PD at T1 negatively predicted self-esteem at T2 (B = −0.173, SE = 0.020, Boot 95% CI [−0.212, −0.135]), and the results of Model 2 showed that S-E at T2 negatively predicted CB at T2 (B = −0.061, SE = 0.020, Boot 95% CI [−0.100, −0.021]). The findings indicated that the total effect of PD at T1 on CB at T2 was significant (total effect = 0.028, SE = 0.001, Boot 95% CI [0.004, 0.051]). Moreover, the indirect effect of PD at T1 on CB at T2 through S-E at T2 was significant (indirect effect = 0.009, SE = 0.003, Boot 95% CI [0.003, 0.019]), although the direct effect of PD at T1 on CB at T2 was non-significant (direct effect = 0.017, SE = 0.012, 95% CI [−0.007, 0.041]).

The PROCESS macro (Model 15) was used to examine the hypothesized moderated mediation model, and the results are shown in Table 3. First, the interaction between RUS at T1 and PD at T1 had a significant effect on CB at T2 (B = −0.045, SE = 0.019, 95% CI [−0.082, −0.008]), indicating that RUS at T1 significantly moderated the direct association between PD at T1 and CB at T2. Moreover, the interaction between RUS at T1 and S-E at T2 had a significant effect on CB at T2 (B = −0.081, SE = 0.033, 95% CI [−0.146, −0.015]), indicating that RUS at T1 significantly moderated the indirect association between PD at T1 and CB at T2 through S-E at T2. According to the result of the conditional direct effect analysis, compared with individuals with low levels of RUS at T1 (i.e., the level of RUS at M - 1SD), the indirect effect of PD at T1 on CB at T2 through S-E at T2 was weaker among individuals with high levels of RUS at T1 (i.e., the level of RUS at M + 1SD). The result of simple slope analysis is shown in Figure 1.

Second, the results of Model 5 are shown in Table 3. The interaction between SC at T1 and PD at T1 had a significant effect on CB at T2 (B = −0.076, SE = 0.031, 95% CI [−0.137, −0.014]), indicating that SC at T1 significantly moderated the direct association between PD at T1 and CB at T2. Moreover, the interaction between SC at T1 and S-E at T2 had a significant effect on CB at T2 (B = −0.103, SE = 0.044, 95% CI [−0.189, −0.017]), indicating that SC at T1 significantly moderated the indirect association between PD at T1 and CB at T2 through S-E at T2. According to the results of the conditional direct effect analysis, compared with individuals with low levels of SC at T1 (i.e., the level of SC at M - 1SD), the indirect effect of PD at T1 on CB at T2 through S-E at T2 was weaker among individuals with high levels of SC at T1 (i.e., the level of SC at M + 1SD). The simple slope analysis is shown in Figures 2, 3. Additionally, the results of Model 4 indicated that the moderating effect of CS at T1 on the direct and indirect association between PD at T1 and CB at T2 through decreased S-E was non-significant (Table 3).