Self-harm, also known as non-suicidal self-injury (NSSI), refers to the intentional act of harming oneself without suicidal intentions, including cutting, punching, burning, scratching bleeding areas, and interfering with wound healing [8]. It has been proven that NSSI is particularly prevalent during adolescence and peaking in early and late adolescence, with college students mainly in this period [9]. Additionally, a meta-analysis conducted in China found that the prevalence of self-harm behaviors among Chinese undergraduates was as high as 16.6% [1]. Evidence has shown that self-harm behaviors have emerged as a major global public health issue that significantly impact individual psychological development [10]. These behaviors are strongly associated with a range of mental problems (e.g., depression, anxiety, and posttraumatic stress disorder) [9] and exhibit greater predictability in relation to suicide compared to other risk factors [11]. Therefore, exploring the psychological mechanisms as well as the risk factors underlying self-harm behaviors among college students is essential for informing the development of targeted intervention and prevention strategies.

Similar to NSSI, aggressive behavior is also considered as one of the common maladaptive behaviors among college students [12]. Empirical evidence has shown that 27.7% of college students exhibited moderate or even higher levels of aggression tendencies [13]. However, with the development and popularization of the internet, interpersonal interactions have increasingly shifted from face-to-face to online. Consequently, aggressive behavior has also taken on new derivative forms, namely online aggression [14].

Online aggression is defined as aggressive behavior executed via a broad range of information and communication technologies, such as social media platforms, email, chat apps, or other messaging tools [15]. Different to traditional aggressive behaviors, aggression occurring in the virtual cyber environment possesses distinct manifestations and characteristics, such as the absence of time and space limitations, the potential for repeated harm, and anonymity [16]. Consequently, online aggressive behaviors have a broader influence and a higher prevalence. Existing evidence indicated that nearly 57% of college students reported initiating online aggression [2]. Moreover, online aggression can seriously influence youth development. A large number of studies have demonstrated that online aggression may result in serious negative impacts on victims, including depression, anxiety, lower academic performance, poor self-esteem, and even suicide [17], as well as the perpetrators [3,18]. Consequently, it is necessary to reveal the antecedents of online aggression, effectively preventing and intervening this behavior.

As for the relationship between self-harm and online aggression, although self-harm and aggression may initially seem distinct, several researches have suggested that self-harm and aggression often co-occur [19]. Recently, the concurrent presence of self-harm behaviors and aggressive behaviors within the same individual has been termed “dual harm.” [20] Individuals manifesting dual-harm behaviors are highly likely to be a high-risk cohort characterized by distinctive attributes and atypical patterns of harmful behaviors [21]. Especially, a hospital-based retrospective study has confirmed that adolescents worrying about maintaining their social connections and reputations are more prone to engage in dual harm [22], indicating the possible predicting effect of SA. However, previous research mainly focused on either self-harm or aggression, neglecting the combination of these two harmful behaviors. Furthermore, very few studies testify to the possible role of SA or clarify the mechanism. To fill the gap, the current study aimed to investigate the underlying mechanisms and risk factors of dual harm behaviors, as well as effective interventions to address this complex issue.

SA refers to the negative emotional experiences, including anxiety, nervousness, and shyness, that individuals encounter in social interactions [23]. These emotions often result in significant distress and functional impairments, adversely affecting social activities and interpersonal relationships [24]. On the one hand, according to interpersonal theory [25], thwarted belongingness and perceived burdensomeness are key factors that increase the likelihood of self-harm behaviors. When individuals experiencing SA, they may overly concern about others’ evaluations, increasingly the susceptibility and fear for social interactions [25]. Unable to effectively express and regulate these negative emotions, individuals may turn to maladaptive coping strategies, such as engaging in self-harm behaviors [26]. Several studies have also demonstrated the close relation between SA and self-harm behaviors [24,27]. For instance, in the investigation of the predictors of self-harm behaviors among college students, SA has been identified as the sole significant predictor. 28 Another study involving 2,992 undergraduates also supports this perspective, revealing the significant predictive role of SA in self-injury behaviors [24].

On the other hand, SA is also one of the factors that can be associated with online aggressive behaviors. The general aggression model (GAM) posits that an individual’s behavior is shaped by the combined influence of personal and environmental factors. Inputs from these factors affect cognition, emotion, and arousal, which are then integrated into the processes of assessment and decision-making [29]. Within this framework, SA functions as a dispositional factor that heightens individuals’ sensitivity to perceived interpersonal threats or negative evaluations in online contexts [30]. This heightened internal reactivity may impair emotional regulation, thereby elevating the risk that impulsive reactions, such as posting hostile comments or engaging in retaliatory behavior, escalate into acts of online aggression [31]. Additionally, SA has been empirically shown to significantly predict online aggressive behaviors [32,33]. For instance, in one study, individuals with SA were found to respond aggressively online [32]. Additionally, longitudinal studies have provided further support for this relationship, revealing that higher levels of SA at baseline predict increased online aggression over time [33].

Therefore, we make the Hypothesis 1 and 2 as below:

Hypothesis 1: Social anxiety could be positively associated with online aggressive behaviors.

Hypothesis 2: Social anxiety could be positively associated with self-harm behaviors.

In the age of information, social networks have profoundly transformed the way humans communicate, offering opportunities for instant interaction and extensive connectivity. However, the excessive use of social networks is gradually evolving into a widespread mental health issue, commonly referred to as SNS addiction [34]. As the prodromal stage of internet addiction, internet addictive tendency includes the core symptoms of internet addiction including withdrawal, salience, relapse [35] and its associated psychological and behavioral consequences such as insomnia, loneliness, depressive affect [36]. While SNS addiction is not officially classified as a psychological disorder in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) [37], its symptoms and negative consequences closely resemble those of other behavioral addictions, such as gambling and video game addiction [38]. The development of social network addiction is influenced by a variety of factors. According to the cognitive-behavioral addiction theory model (CBM) [39], these factors can be categorized into distal and proximal influences. Distal influences, such as environmental stressors and psychopathological vulnerabilities, play indirect role in the occurrence of SNS addictive tendency. In contrast, proximal factors (e.g., maladaptive cognitive processes), including dysfunctional beliefs about the self and interpersonal relationships, may directly promote compulsive use of SNS. Individuals with SA often demonstrate cognitive impairments, such as biased attention to negative social cues and persistent negative self-evaluation [40,41]. These dysfunctional cognitions may intensify avoidance of face-to-face interactions and reinforce maladaptive online behaviors that serve as compensatory strategies. Consequently, SA may lead individuals to rely more heavily on social media platforms as perceived safer and more controllable social environments, thereby increasing the risk of developing SNS addiction [7,42,43].

Moreover, emerging evidence has identified SNS addictive tendency as a significant predictor of self-harm and online aggressive behaviors. According to the Problematic Psychosocial Predisposition Model, uncontrollable internet use, especially excessive engagement in online social interaction, can undermine individuals’ socially adaptive functioning, leading to a range of internalizing and externalizing problems, including self-injury and aggressive behaviors [44]. Specifically, continuous access to social networks may heighten exposure to bullying, unfavorable social comparisons, and persistent criticism. These factors are closely linked to elevated levels of anxiety, depression, low self-esteem, and maladaptive emotional regulation strategies [45]. When left unresolved, these negative emotional states may drive individuals to engage in self-injurious behaviors as a means of affect regulation, or to display aggression online as a maladaptive strategy for expressing frustration or reclaiming a sense of social power [46-48].

Given the above, it is reasonable to believe that SA may influence individuals’ self-harm and online aggressive behaviors by enhancing their tendency to engage in SNS addiction. Therefore, within the integrated of theoretical frameworks and empirical evidence, we propose the Hypothesis 3, 4, and 5 as below:

Hypothesis 3: Social anxiety could be positively associated with SNS addictive tendency.

Hypothesis 4: SNS addictive tendency is positively related to self-harm behaviors and could mediate the relation between social anxiety and self-harm behaviors.

Hypothesis 5: SNS addictive tendency is positively related to online aggressive behaviors and could mediate the relation between social anxiety and online aggressive behaviors.

Despite the theoretically established relationship between SA and SNS addictive tendency, not all individuals with high levels of SA will exhibit excessive social network use [49]. Thus, it is essential to explore the factors that may weaken the strength of the relation between SA and SNS addictive tendency. Social role theory (SRT) suggests that individuals’ internalized sex role can significantly influence their responses to social and psychological stress [50]. Sex roles refer to the societal and cultural expectations for male and female behavior, which in turn affect how individuals cope with stressors. Influenced by these social norms, males are generally expected to exhibit independence, decisiveness, and the ability to handle stress without showing vulnerability, whereas females are more likely to depend on interpersonal support and exhibit more emotional reactions [51].

However, this societal expectation for men to suppress emotional vulnerability may discourage them from seeking face-to-face emotional support, thereby prompting them to pursue more controllable and less confrontational alternatives, such as online platforms. As a result, men experiencing high levels of SA may be more likely to engage in SNS use as a compensatory strategy to manage stress and maintain a socially desirable self-image. Supporting this perspective, prior research has demonstrated that socially anxious men are more likely to use SNS for initiating new connections, whereas women tend to use SNS to maintain existing relationships [52]. Further reinforcing this view, a longitudinal research conducted in 2,103 Korean college students showed that men with SA tend to spend more time engaging in online social interactions in order to maintain their self-image [53].

Therefore, based on SRT and empirical evidence, we formulate the Hypothesis 6 that:

Hypothesis 6: The relation between SA and SNS addictive tendency is stronger among males than females.

In summary, based on the interpersonal theory, GAM, CBM, SRT, and empirical research, the current study aims to investigate the mediating role of SNS addictive tendency and the moderating role of sex in the relation between SA and self-harm behaviors, as well as online aggression. Specifically, the present study proposes the following hypothesized model in Figure 1.

The data collection process happened in Jiangsu province of China in October 2023 and we used the convenient sampling to recruit college students from four universities. The data collection was based on an online survey platform named “Wenjuanxing” (www.wenjuanxing.com). Before the survey began, participants were presented with informed consent in which the aim and content of the online survey were introduced clearly. All participants were informed that they could freely choose to participate in the survey or not and had the right to quit at any moment. In addition, we used G*power (version 3.1.9.7; Heinrich Heine University) to estimate the needed sample size for verifying the hypothesized model and it is suggested that at least 107 samples are needed with the effect size setting to 0.15.

Among 3,157 college students who received the online survey link, 3,005 of them consented to participate and completed the survey. Noticeably, to prevent any possible negative impacts caused by the survey, participants who reported being at the onset stage for severe mental disorders were not included and provided with the resources for mental health services. Based on participants’ responses to two attention test questions, we identified 1,435 participants as careless responders for not passing the attention test and removed them from the data set. A total of 1,608 participants (females=1,245, 77.43%; Mean_age=18.95 years) were included in further data analysis.

All participants and their parents were fully informed about the objectives of the study and provided electronic informed consent. This study was approved by the ethics Committee of Brain Hospital of Nanjing Medical University (reference number: 202305290090).

The data analysis was conducted using R 4.3.1 [61] and Mplus 8.0 (Muthén & Muthén). All variables were standardized before further analysis. The reliability and validity of the measuring instruments were assessed using the reliability analysis and factor analysis. Meanwhile, the Haman’s single-factor test was used in the study to test the existence of the common method bias [62]. The descriptive statistics and Pearson correlations between variables were conducted to examine the relation among online aggressive behaviours, social network addiction, SA, self-harm behaviors, sex, and age.

Then, we examined the mediating effect of social network addiction on the relation between SA-online aggressive behaviors and SA-self-harm behaviors. The indirect effects were tested using the bootstrap approach and were considered significant if the 95% confidence intervals (CIs) did not cover 0. After that, we explore the moderating effect of sex on the mediating process by conducting a multi-group analysis which could testify if there exist significant differences in the coefficients in the model across sexes. In detail, we compare the unconstrained model (allowing freely estimating the coefficients in the model of two sexes) with the constrained model (setting the coefficients in the model of two sexes to be equal). Finally, we used the Wald χ² test to compare the specific path coefficients between sexes. In the model, adolescents’ age was controlled.

Before we conduct further analysis, we tested the reliability and validity of the scales used in the current study. All instruments employed in the current study showed good reliability and validity with Cronbach’s α >0.80 and Kaiser-Meyer-Olkin values >0.85. Therefore, the data collected by these measuring instruments could be included in further study.

In addition, the result of Haman’s single-factor test shows that the eigenvalues of 10 factors were higher than 1 in unrotated factor analysis, and the first factor accounted for 26.05% of the total variance, which was less than the critical standard of 40% [62].

The mean values, standard deviation, and correlation between variables are shown in Table 1.

As demonstrated in Table 1, all research variables are significantly correlated with each other, which suggests that further analysis of the mediation effect is available. Meanwhile, it is also shown in Table 1 that sex is significantly related to both SA and online aggressive behaviours. Indeed, we conducted a t-test to compare levels of SA, SNS addictive tendences, online aggressive behaviours, and self-harm behaviours among two sex groups. The results are shown in Table 2.

As depicted in Table 2, the level of SA among females is significantly higher than that among males (t=-3.33, p<0.001), while males show more online aggressive behaviours than females (t=3.38, p<0.001).

The mediation model in which SNS addictive tendence mediates the relation between SA and online aggressive behaviour as well as the relation between SA and self-harm behaviours was conducted. The model fit indices suggest that the model fit well on the current data, with Root Mean Square Error of Approximation equal to 0.049, comparative fit index equal to 0.994 and Tucker-Lewis index (TLI) equal to 0.965. The results of regression indicate that SA could significantly predict online aggressive behaviours (β=0.075, p<0.01), self-harm behaviours (β=0.112, p<0.01), and SNS addictive tendence (β=0.710, p<0.001). Meanwhile, SNS addictive tendence can also significantly predict both online aggressive behaviours (β=0.183, p<0.001) and self-harm behaviours (β=0.073, p<0.001).

Moreover, the indirect effect from SA to online aggressive behaviours was equal to 0.130 (p<0.001) and the indirect effect from SA to self-harm behaviours was equal to 0.052 (p<0.05). Apart from that, the results of bootstrap (times=1,000) also indicate significant indirect between SA and online aggressive behaviours (95% CIs: 0.077-0.191) as well as SA and self-harm behaviours (95% CIs: 0.000-0.104). The results of the mediation analysis are shown in Table 3 and Figure 2 to better illustrate the mediation model.

Considering that there exist significant differences in the levels of SA and online aggressive behaviours, we further analyzed whether sex plays a moderated role in the mediation model. Hence, the Wald χ² test was employed in the paths of the mediation model to explore the difference between sexes. Specifically, the difference in model fit index was significant (ΔTLI=0.021), indicating a significant difference between the unconstrained and the constrained model [63]. In other words, there is significant difference between the group models of females and males.

The results of the Wald χ² test are shown in Table 4. As shown in Table 4, the predicted effect of SA on SNS addictive tendency differed significantly between sexes (χ²=4.585, p<0.05). In other words, sex moderates the relation between SA and SNS addictive tendency. The mediation models of both sexes are depicted in Figure 3, respectively.

Results in the current study declare that SNS addictive tendency mediates the relation between SA and self-harm solely among females. This may indicate that more attention is warranted among females facing a higher risk of SA and showing SNS addictive tendency to prevent the possible self-harm behaviours. Meanwhile, the current study may suggest intervention targeting at equipping individuals with more adaptive social interacting skills could be beneficial in preventing self-harm behaviours.

Several limitations in the current study need to be mentioned. First and foremost, the number of males and females in the current study is highly unbalanced. Therefore, the results of the current study need to be explained with caution. Moreover, the definition of problematic social media use or SNS addictive tendency did not reach an agreement. Hence, a more standardized and universal definition of the problematic use of social media is needed. Last but not least, when assessing SNS addictive tendency, more objective indices, such as the time spent on social media could be further included.

The current study used the structural equation model to testify to the mediating effect of SNS addictive tendency on the relation between SA and self-harm behaviours, online aggressive behaviours. By adopting the Wald χ² test, the study verified the moderating effect of sexes on the mediating effect of SNS addictive tendency. The results demonstrated that SNS addictive tendency plays a significant mediating role in the relation between SA and self-harm behaviours, online aggressive behaviours and sex could moderate the relation between SA and SNS addictive tendency. Our findings suggested that intervention targeted at improving socially anxious individuals’ ability to develop and maintain the interpersonal relationship in real life could help decrease the risk for SNS addiction as well as buffer self-harm behaviours and online aggressive behaviours.