The dynamic interplay of cyberbullying victimization and self-injury: a longitudinal investigation of rumination and friendship quality in Chinese vocational students

Abstract

Objective

This longitudinal study examined the relationship between cyberbullying victimization (CV) and non-suicidal self-injury (NSSI) among Chinese vocational college students, testing the mediating role of ruminative contemplation (RC) and the moderating effect of friendship quality (FQ).

Methods

A six-month longitudinal design was implemented with 2,312 vocational students (M_age(T1) = 19.06; 35.55% CV prevalence, 19.26% NSSI prevalence at baseline). Data were collected at two waves (T1: December 2024; T2: June 2025) using validated scales: Cyberbullying Inventory for College Students (CICS), Adolescent Self-Harm Questionnaire, Event-Related Rumination Inventory (C-ERRI), and Friendship Quality Questionnaire (FQQ-C). Cross-lagged panel models (CLPM) and moderated mediation analyses were conducted via Mplus 8.3, controlling for gender, only-child status, and residence.

Results

CV directly predicted increased NSSI over time. RC mediated the CV→NSSI link. FQ significantly buffered the direct CV→NSSI pathway, with maximal protection observed at FQ scores = 212.3 (Johnson-Neyman analysis).

Conclusions

Cyberbullying victimization heightens NSSI risk both directly and indirectly through ruminative contemplation. High-quality friendships mitigate this pathway, offering a critical protective factor for vocational students. Interventions should target RC reduction and FQ enhancement to disrupt the CV→NSSI trajectory.

Cyberbullying victimization (CV) refers to the experience of being intentionally, repeatedly, and aggressively harmed by others through digital platforms such as social media, instant messaging applications, and SMS [1]. From the victim’s perspective, this experience is typically characterized by perceived hostile intent, exposure to repeated harmful acts, and a power imbalance that impedes easy defense or escape. Such behaviors often manifest as harassment, threats, and public humiliation, inflicting profound damage on psychosocial functioning [1, 2]. Globally, CV is recognized as a significant public health issue among adolescents and young adults, with prevalence rates varying across countries and cultural contexts [3].

Within the Chinese context, vocational college students represent a particularly vulnerable subgroup. Compared to university peers, vocational students face compounded stressors, including lower academic thresholds, intensified employment pressure, and social marginalization - potentially amplifying both susceptibility to CV and maladaptive coping [3]. Consequently, CV prevalence in this population reaches alarming levels: 63% of Chinese vocational students report victimization [4], significantly exceeding rates among general university students (7.82% − 11.49%) [5, 6]. Critically, this disparity extends to self-injurious outcomes, positioning CV as a key precipitant of non-suicidal self-injury (NSSI) - deliberate bodily harm without suicidal intent [7]. Meta-analyses reveal an NSSI prevalence of 16.6% among Chinese college students [8], compared to 23.45% among vocational college students [9].

A growing body of research demonstrates a robust association between CV and NSSI. Bullied adolescents are more than twice as likely to engage in NSSI compared to non-victims [10], with approximately one-third of cyberbullying victims reporting persistent fear and debilitating negative emotions [11]. These findings are particularly concerning because NSSI is one of the strongest predictors of suicidal behavior [12–16]. Understanding the mechanisms linking CV to NSSI, such as cognitive and emotional processes, is therefore critical for developing targeted interventions that can interrupt this trajectory before it escalates to life-threatening outcomes. Recent systematic reviews have called for longitudinal studies that examine mediators and moderators of this relationship to inform prevention efforts [16, 17].

China’s unique sociocultural context, characterized by rapid digitalization, collectivist values, and a stratified education system, offers a critical lens to examine how cyberbullying victimization operates in non-Western settings [3]. As CV is a transnational phenomenon, understanding its dynamics in Chinese vocational students, a population facing heightened stressors, can inform cross-cultural theories of stress and coping. Moreover, findings from this underrepresented group may reveal culturally specific protective factors (e.g., friendship quality) that could be integrated into global intervention strategies.

Theoretical frameworks and hypotheses development

To elucidate the mechanisms linking CV to NSSI, this study integrates three complementary theoretical perspectives. Each theory directly informs specific pathways in our proposed moderated mediation model .

General Strain Theory (GST)

GST posits that exposure to chronic stressors generates negative affect (e.g., anger, frustration, depression). In the absence of effective coping resources, this increases the likelihood of maladaptive behaviors, including self-harm [18, 19]. Within our model, CV constitutes a significant digital strain–characterized by goal blockage and identity erosion, key dimensions of strain in GST that directly increases NSSI risk. Thus, GST provides the theoretical rationale for the direct path from CV to NSSI (Hypothesis 1).

Emotional Cascade Model (ECM)

ECM proposes that ruminative processes create recursive cognition-affect loops, each iteration intensifying negative emotions and depleting cognitive resources, ultimately overwhelming adaptive regulation capacity [20, 21]. NSSI may then serve as a maladaptive escape mechanism [22, 23]. In our model, ruminative contemplation (RC), defined as the tendency to repetitively and passively think about the causes and consequences of distress–is hypothesized to be the cognitive mechanism through which CV-related distress into NSSI. ECM therefore underpins the mediating role of RC in the CV→NSSI relationship (Hypothesis 2).

Buffering model of social support

This model suggests that high-quality social relationships can mitigate the negative impact of stressors on mental health by providing emotional sustenance, cognitive reappraisal, and practical assistance [24].

Simultaneously, friendship quality (FQ), reflecting the level of support, companionship, and trust within peer relationships, has been shown to be a stable protective factor against the negative outcomes of bullying [25] and to negatively predict NSSI among adolescents [26, 27]. In our model, friendship quality (FQ), reflecting the level of support, companionship, and trust within peer relationships - is hypothesized to attenuate the direct effect of CV on NSSI. The Buffering Model thus provides the theoretical foundation for the moderating role of FQ (Hypothesis 3).

Integrative framework

By integrating these three perspectives, our study proposes a unified longitudinal model (Fig. 1) in which: (a) CV functions as a strain (GST) that directly predicts NSSI; (b) RC serves as the cognitive mechanism that amplifies this relationship through an emotional cascade (ECM); and (c) FQ acts as an external protective factor that buffers the direct pathway (Buffering Model). This integration addresses the critical gaps identified in prior research by simultaneously examining mediation and moderation processes within a longitudinal design.

Fig. 1.

Proposed theoretical model. Note: GST underpins H1:T1 CV → T2 NSSI (direct path, controlling for autoregressive effects); T1 CV → T2 RC → T2 NSSI (mediation path); The direct path T1 CV → T2 NSSI is moderated by FQ (i.e., CV × FQ interaction → T2 NSSI)

Methods

Participants

This longitudinal study received ethical approval from the Ethics Committee of The Third People’s Hospital of Qujing. The study procedures adhered to the principles of the Declaration of Helsinki.

A longitudinal study was conducted over six months using a cluster convenience sampling method at two vocational colleges in Southwest China and one in Central China. College administrators were contacted to obtain permission for research. Prior to survey administration, written informed consent was obtained from all participants. For participants under 18 years of age, written assent was obtained alongside parental consent. The first survey wave (T1) was administered in December 2024, in classroom settings under the supervision of trained research assistants, collecting data on demographics, cyberbullying victimization, NSSI, and ruminative contemplation. The second wave (T2) was administered in June 2025, adding the friendship quality.

T1 yielded 4,465 questionnaires (initial response rate: 98% of approached students). Due to some participants being unavailable for follow-up (e.g., off-campus internships), 2,927 questionnaires were distributed at T2. After preliminary screening (excluding short response times, family history of self-harm, current/past substance abuse/dependence) and matching T1 and T2 data, 2,312 valid longitudinal responses were retained (final response-rate:78.99%). At T1, participants’ mean age was 19.06 years (SD = 0.94); 1,695 were male, 617 female; 307 were only children, 2,005 were not; 2,065 resided in rural areas, 247 in urban areas. CV screening prevalence was 35.55%; NSSI screening prevalence was 19.26%.

Measures

Cyberbullying Inventory for College Students (CICS)

We utilized the Chinese version of the Cyberbullying Inventory for College Students (CICS) translated and published by Tang et al. [28]. It comprises 18 items across two subscales: cyberbullying perpetration (9 items) and cyber victimization (9 items). This study used the victimization subscale. Items (e.g., “Someone spread rumors about me online”) are rated on a 5-point frequency scale (1 = never, 5 = multiple times a week). For the purpose of our six-month longitudinal analysis, the instruction at T1 assessed experiences “in the past year,” establishing a baseline. At T2, the instruction was modified to assess experiences “in the past six months (since the last survey)” to capture incidents occurring during the study interval and minimize temporal overlap. Higher scores indicate more severe victimization. Cronbach’s α was 0.803 (T1) and 0.826 (T2). Confirmatory Factor Analysis (CFA) indicated good fit: T1: χ²/df = 2.11, CFI = 0.91, TLI = 0.92, RMSEA = 0.05, SRMR = 0.043; T2: χ²/df = 2.28, CFI = 0.946, TLI = 0.928, RMSEA = 0.07, SRMR = 0.036.

Adolescent self-harm questionnaire

This self-report measure, developed by Feng [29] based on Graze’s Deliberate Self-Harm Scale and Zheng’s Middle School Student Behavior Questionnaire, consists of 19 items. To suit the longitudinal design and assess NSSI occurring within the study period, we adapted the timeframe. Participants reported on the frequency and severity of each NSSI behavior occurring in the six months prior to each assessment (T1 and T2), rather than over their lifetime. NSSI score = frequency × severity. Frequency (0 times, 1 time, 2–5 times, > 5 times) is scored 0–3. Severity (none, mild, moderate, severe, extremely severe) is scored 0–4. Item 19 (open-ended) was excluded, only items 1–18 were used. Higher scores indicate more severe NSSI. Sample item: “Deliberately scratch your skin with glass, knife, etc.” Cronbach’s α was 0.834 (T1) and 0.829 (T2). CFA indicated good fit: T1: χ²/df = 2.37, CFI = 0.977, TLI = 0.921, RMSEA = 0.041, SRMR = 0.069; T2: χ²/df = 2.84, CFI = 0.946, TLI = 0.924, RMSEA = 0.048, SRMR = 0.05.

Simplified Chinese version of the Event-Related Rumination Inventory (C-ERRI)

We employed the published Simplified Chinese Version of the Event-Related Rumination Inventory (C-ERRI), which was translated and revised by Dong [30] from Cann’s original. The 20-item scale has two dimensions: ruminative contemplation (items 1–10) and deliberate rumination (items 11–20). Rated on a 4-pointscale (0–3). Higher scores indicate higher rumination. Sample item: “Other things keep me thinking about that experience.” Cronbach’s α was 0.979 (T1) and 0.984 (T2). CFA indicated good fit: T1: χ²/df = 4.62, CFI = 0.918, TLI = 0.903, RMSEA = 0.026, SRMR = 0.072; T2: χ²/df = 4.43, CFI = 0.917, TLI = 0.905, RMSEA = 0.042, SRMR = 0.055.

Friendship Quality Questionnaire for College Students(FQQ-C)

Developed by Wen [31] based on Chinese university students’ interpersonal characteristics. The 40-item scale has a first-order six-factor structure measuring: Intimacy, Help and Support, Relationship Maintenance, Appreciation and Trust, Relationship Resilience, Shared Interests. Participants rated statements on a 6-point Likert scale (1 = Completely Disagree to 6 = Completely Agree) regarding a specific close friend. Higher total scores indicate better friendship quality. Sample item: “My close friend helps me confide worries when I am troubled.” Administered at T2, Cronbach’s α was 0.905. CFA indicated good fit: χ²/df = 4.93, CFI = 0.952, TLI = 0.944, RMSEA = 0.064, SRMR = 0.032.

Data analysis

First, longitudinal measurement invariance of CV, NSSI, and RC across T1 and T2 was tested using Confirmatory Factor Analysis (CFA) in Mplus 8.3. Sequential models were tested: (1) Configural invariance (baseline model); (2) Weak invariance (equal factor loadings); (3) Strong invariance (equal factor loadings and intercepts). Invariance was supported if ΔCFI < 0.01 [32]. Second, bivariate correlations among variables were analyzed using SPSS 26.0. Third, a cross-lagged panel model (CLPM) between CV and NSSI was constructed. Fourth, a CLPM incorporating CV, NSSI, and RC was constructed. Finally, the moderating role of FQ within the model was tested. Simple slope analysis was performed using the Johnson-Neyman (J-N) technique in R 4.4.3. Gender, only-child status, and family residence were included as covariates in all models.

Results

Measurement invariance testing

When testing invariance in large samples, ΔCFI is a more robust criterion than absolute χ² values due to the latter’s oversensitivity to sample size [32]. As shown in Table 1, strong measurement invariance was established for CV, NSSI, and RC (all ΔCFI < 0.01), indicating equivalent factor structures over time.

Table 1.

Measurement invariance testing for CV、NSSI、IR

Measurement Model	χ2 (df)	ΔCFI	CFI	TLI	RMSEA [90% CI]	SRMR
CV
Configural	587.01 (125)	—	0.923	0.906	0.040 [0.037, 0.043]	0.036
Weak	933.26 (134)	0.001	0.922	0.910	0.040 [0.038, 0.042]	0.038
Strong	1014.99 (142)	0.009	0.913	0.908	0.050 [0.049, 0.052]	0.039
NSSI
Configural	2231.96 (575)	—	0.887	0.861	0.035 [0.034, 0.037]	0.072
Weak	2178.58 (593)	0.002	0.885	0.865	0.034 [0.032, 0.036]	0.076
Strong	2240.82 (610)	0.004	0.881	0.867	0.034 [0.033, 0.036]	0.078
RC
Configural	1233.08 (719)	—	0.910	0.901	0.080 [0.078, 0.081]	0.049
Weak	1356.01 (739)	0.006	0.904	0.903	0.079 [0.078, 0.080]	0.050
Strong	1567.79 (758)	0.002	0.902	0.898	0.079 [0.077, 0.080]	0.050

CV  Cyberbullying Victimization, NSSI  Non-Suicidal Self-Injury, RC Ruminative Contemplation, CI  Confidence Interval

Descriptive statistics and correlations

Correlation analysis results (see Table 2) indicated significant positive correlations between CV, NSSI, and RC at both T1 and T2. T2 FQ was significantly negatively correlated with T1 CV and T2 NSSI.

Table 2.

Descriptive statistics and correlations among variables(n = 2312)

	M	SD	1	2	3	4	5	6	7	8	9
1.T1CV	10.23	2.37
2.T2CV	10.50	3.63	0.23***
3.T1NSSI	1.60	3.24	0.12***	0.12***
4.T2NSSI	1.35	3.15	0.19***	0.19***	0.41***
5.T1RC	8.17	12.36	0.34***	0.10***	0.30***	0.15***
6.T2RC	6.67	12.06	0.16***	0.30***	0.19***	0.24***	0.25***
7.T2FQ	149.77	30.82	−0.04*	−0.14***	-0.02	-0.05*	0.01	0.08***
8.Gender	-	-	-0.01	0.04*	0.06**	0.04	0.07**	0.05*	0.14***
9.Only child	-	-	-0.02	0.01	-0.03	-0.04***	-0.05*	-0.02	-0.03	0.05*
10.Family residence	-	-	-0.07**	-0.02	-0.06**	-0.08***	-0.08***	-0.01	-0.05**	-0.10	0.27***

CV Cyberbullying Victimization, NSSI Non-Suicidal Self-Injury, RC Ruminative Contemplation, FQ Friendship Quality; Gender: 1 = Male, 2 = Female; Only Child: 1 = Yes, 2 = No; Residence: 1 = Rural, 2 = Urban. ^*p < 0.05, ^**p < 0.01, ^***p < 0.001

Cross-lagged model between cyberbullying victimization and NSSI

A CLPM between CV and NSSI was constructed using Mplus 8.3 (Fig. 2). Controlling for gender, only-child status, and family residence, the model fit was good: χ²/df = 2.06, p < 0.05, CFI = 0.951, TLI = 0.918, RMSEA = 0.021, 90% CI [0.009, 0.033], SRMR = 0.016. T1 CV significantly predicted T2 NSSI (β = 0.07, p < 0.001), while T1 NSSI did not significantly predict T2 CV (β = 0.05, p > 0.05). These results indicate that cyberbullying victimization positively predicts subsequent NSSI. Among covariates, gender negatively predicted T2 CV (β = -0.05, p < 0.05), and family residence negatively predicted T1 NSSI (β = -0.08, p < 0.05). Only-child status had no significant predictive effect.

Fig. 2.

Cross-lagged model between Cyberbullying Victimization (CV) and Non-Suicidal Self-Injury (NSSI)

Longitudinal mediating role of ruminative contemplation

The longitudinal mediating role of RC in the relationship between CV and NSSI was tested using Mplus 8.3. Controlling for covariates and stability paths, the model fit was good (χ²/df = 1.92, p < 0.01, CFI = 0.911, TLI = 0.904, RMSEA = 0.06, 90% CI [0.056, 0.064], SRMR = 0.047). Results are shown in Fig. 3. Controlling for T1 RC and T1 NSSI, T1 CV significantly predicted T2 RC (β = 0.31, p < 0.05). Controlling for T1 RC and T1 NSSI, T1 CV significantly predicted T2 NSSI (β = 0.06, p < 0.05). Controlling for T1 CV and T1 NSSI, T1 RC significantly predicted T2 NSSI (β = 0.02, p < 0.05). This indicates that cyberbullying victimization leads to higher levels of ruminative contemplation, which in turn lead to increased NSSI. Among covariates, gender negatively predicted T1 RC (β = 0.07, p < 0.01) and T2 CV (β = -0.05, p < 0.01); family residence negatively predicted T1 NSSI (β = -0.05, p < 0.05); only-child status had no significant effect.

Fig. 3.

Longitudinal mediation model: RC mediating the relationship between CV and NSSI

Moderating role of friendship quality in the longitudinal mediation model

Given the significant T2 RC mediation effect, the moderating role of T2 FQ on the direct path from T1 CV to T2 NSSI within the mediation model was tested. A moderated mediation model was constructed in Mplus 8.3. Controlling for covariates, the model fit was good (χ²/df = 1.68, p < 0.05, CFI = 0.928, TLI = 0.910, RMSEA = 0.04, 90% CI [0.033, 0.047], SRMR = 0.022). The direct effect of T1 CV on T2 NSSI remained significant (β = 0.04, p < 0.001). Crucially, the interaction term between T1 CV and T2 FQ significantly negatively predicted T2 NSSI (β = -0.21, p < 0.01), indicating that friendship quality significantly buffers the direct effect of CV on NSSI.

To clarify the nature of the moderation, a simple slope analysis was conducted using the Johnson-Neyman (J-N) technique in R 4.4.3. Results are shown in Fig. 4. The J-N plot shows the effect size (slope) of T1 CV on T2 NSSI across the range of T2 FQ scores, with curved lines representing the 95% confidence interval (CI). The 95% CI upper bound included zero at a T2 FQ score of 368.76. When T2 FQ was greater than 368.76, the CI no longer contained zero. More importantly, the 95% CI lower bound included zero at a T2 FQ score of 212.3. When T2 FQ was less than 212.3, the CI did not contain zero, indicating a significant positive effect of T1 CV on T2 NSSI in this range. As the maximum possible FQ score is 240, the value of 368.76 has limited practical meaning (extrapolation). Therefore, the critical threshold of 212.3 is meaningful. The buffering effect of FQ is significant across virtually the entire sample range, as the Johnson-Neyman point of 212.3 approaches the scale maximum (240). This indicates that FQ provides robust protection against the CV→NSSI pathway for nearly all students, with only those possessing exceptionally high friendship quality (FQ > 212.3) showing no significant buffering effect.

Fig. 4.

Simple slope analysis: buffering effect of Friendship Quality (FQ) on the CV→NSSI path

The Johnson-Neyman analysis indicated that the protective buffering effect of friendship quality (FQ) was statistically significant across almost the entire observed range of FQ scores (i.e., for FQ scores below 212.3). The positive effect of T1 CV on T2 NSSI weakened as FQ increased. Thus, friendship quality effectively buffers the direct effect of CV on NSSI.

Discussion

Relationship between CV and NSSI

The longitudinal findings confirm CV as a critical predictor of NSSI among Chinese vocational college students, extending General Strain Theory (GST) by demonstrating that digital strain operates through distinct mechanisms in this population. Unlike traditional bullying, CV’s 24/7 accessibility and potential for anonymous perpetration create a uniquely inescapable stressor, what we term “digital entrapment.” This finding advances GST by suggesting that the permanence and publicness of digital attacks may amplify their strain-inducing potential beyond what face-to-face victimization typically achieves [18, 19]. For vocational students specifically, who already navigate academic marginalization and employment precarity, CV represents an additional layer of identity threat that compounds existing strains, making them particularly susceptible to maladaptive coping.

Longitudinal mediating role of RC

The resulting negative affect - shame, anger, and a profound sense of relational devaluation, creates the affective raw material for self-harm. Our findings reveal that it is the ruminative processing of this affect that propels the individual toward self-injury. This pattern directly supports the Emotional Cascade Model’s core proposition: recursive cognition-affect loops, rather than discrete emotional experiences, drive dysregulated behavior [33]. Notably, the magnitude of the mediation effect (β = 0.02) suggests that rumination functions as a “cognitive amplifier” rather than a simple transmitter - a distinction with important clinical implications. This process is supported by neurocognitive evidence: rumination is associated with sustained activation in brain regions linked to self-referential processing and emotional distress (e.g., default mode network, amygdala), concurrent with diminished activity in prefrontal regions responsible for executive control, cognitive reappraisal, and inhibitory regulation. This neural profile effectively traps the individual, depleting the very cognitive resources needed to break the cycle [20, 34]. When the amplified distress surpasses a threshold of tolerance, NSSI emerges as a desperate, concrete action to disrupt the unbearable cognitive-affective loop. It serves a paradoxical regulatory function, shifting focus from psychic pain to physical sensation, thereby providing temporary relief from the rumination-driven emotional overload [35].

Interventions targeting rumination may therefore be more effective at disrupting the CV→NSSI trajectory than those targeting negative affect directly, as they address the amplification mechanism rather than its input.

The alarming prevalence rates in our sample − 35.55% for CV and 19.26% for NSSI - reflect vocational students’ unique vulnerability. These figures starkly exceed rates among general university students [5, 8]. This disparity underscores how vocational stressors - academic marginalization, employment anxiety, and social stigma [3] amplify digital victimization’s impact, creating a high-risk context where ECM’s emotion-cognition-behavior cascades readily ignite.

Moderating role of FQ

The protective buffering effect of FQ on the direct CV→NSSI path robustly supports the Buffering Model of social support [24]. High-quality friendships do not eliminate CV but fundamentally alter its psychological impact through multiple synergistic pathways [31]:

• Cognitive-Affective Buffering: Trusted friends provide immediate emotional validation (“That was wrong, and your feelings are understandable”), which counteracts the isolation and self-blame CV often instills. More importantly, they offer alternative cognitive appraisals, helping the victim reframe the event (e.g., “This says more about the bully’s issues than yours”) and challenging distorted, global self-evaluations fueled by rumination. This directly interrupts the cognitive component of the emotional cascade.

• Behavioral Diversion & Problem-Solving: Friends engage the individual in shared positive activities (reflecting the Shared Interests and Companionship dimensions of FQ), which serves as active behavioral distraction, pulling attention away from ruminative loops. They may also assist in practical problem-solving, such as reporting abuse or blocking perpetrators, thereby enhancing perceived control.

• Provision of a Secure Base: The dimensions of Intimacy, Trust, and Relationship Resilience create a psychological safety net. Knowing one has unwavering support reduces the perceived threat of the stressor, lowers physiological arousal, and fosters resilience. This secure base makes the individual less likely to resort to drastic, solitary coping mechanisms like NSSI.

The Johnson-Neyman analysis indicating significant buffering for nearly the entire sample underscores that even moderately good friendships offer substantial protection. The mechanism is not all-or-nothing but operates along a continuum: stronger friendship bonds provide more resources for validation, reappraisal, and distraction, thereby more effectively leaking pressure from the GST-ECM pathway before it culminates in self-injury.

Theoretical contributions and practical implications

Theoretical contributions

This study makes three substantive contributions to the literature. First, by integrating GST, ECM, and the Buffering Model within a single longitudinal framework, we demonstrate that these theories are not competing explanations but complementary perspectives operating at different levels of analysis. GST explains the initiation of distress (strain → affect), ECM explains its amplification (affect → rumination → dysregulation), and the Buffering Model explains its attenuation (social support as external regulator). This integrative approach moves the field beyond single-theory tests toward a more nuanced understanding of how risk and protective factors interact across time.

Second, our finding that RC mediates the CV→NSSI relationship provides the longitudinal evidence for ECM’s core mechanism in the context of digital victimization. The modest but significant mediation effect (β = 0.02) suggests that rumination functions as a “cognitive amplifier” - a small but persistent effect that, over time, may accumulate into clinically meaningful outcomes. This temporal perspective has been largely absent from prior cross-sectional tests of ECM.

Third, the J-N analysis revealing buffering effects across almost the entire FQ range (significant for FQ < 212.3) challenges assumptions about threshold effects in social support research. Rather than a “critical minimum” of support required for protection, our findings suggest a continuous gradient: every increment in friendship quality confers additional protective benefit. This has important implications for how we conceptualize and measure protective factors.

Practical implications

For mental health practitioners serving vocational students, our findings suggest three intervention targets. First, CV prevention programs should address the unique features of digital victimization (e.g., 24/7 accessibility, public humiliation) that make it particularly damaging. Second, rumination-focused interventions, such as mindfulness-based cognitive therapy or rumination-focused CBT, may be more effective in disrupting the CV→NSSI trajectory than general emotion regulation training. Third, friendship quality enhancement should be integrated into intervention design, as even modest improvements in peer relationships provide protective benefits. This suggests that school-based social skills programs and peer mentoring initiatives are worthwhile investments.

Conclusion

This longitudinal study provides empirical support for an integrated theoretical model linking CV to NSSI among Chinese vocational students. CV functions as a digital strain that, when amplified through ruminative contemplation, increases NSSI risk - a process that high-quality friendships can significantly buffer. These findings underscore the value of integrating strain-based, cognitive, and interpersonal perspectives to understand self-harm in vulnerable populations. Future interventions should consider simultaneously targeting digital victimization, ruminative processes, and peer relationship quality to disrupt the trajectory from online harassment to self-injury.

Authors’ contributions

X M: Funding acquisition, Writing—original draft. L H and M Y: Investigation. S L: Methodology, Formal analysis. L G: Formal analysis. L Y: Data curation. J X: Project administration. X Y: Writing—review and editing. All authors read and approved the final manuscript.

Funding

This work was supported by Scientific Research Fund Project of Education Department of Yunnan Province (No.2025J1484).

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of The Second People’s Hospital of Honghe Prefecture. The study procedures were conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants prior to their enrolment in the study. The participants’ responses were confidential and were only used in this study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.