A cross-sectional study was conducted among university students in Shanghai, China, from September to December 2024. A stratified cluster random sampling method was adopted, with 5 universities serving as the sampling frame. Participants were stratified by academic discipline (medical sciences, science and engineering, and social sciences) and academic year (freshman to senior). Within each university, student groups (classes) were defined as clusters according to discipline and academic year. Subsequently, several clusters were randomly selected within each stratum, and all students in the selected clusters were invited to participate in the survey. The inclusion criteria were as follows: (1) full-time undergraduate students; (2) active social media users, defined as using social media platforms at least once per day; and (3) voluntary participation with informed consent.

In the preliminary stage of this study, a research team was established, consisting of the principal investigator and graduate student investigators. The team developed an investigator manual and conducted 2 systematic training sessions. The training covered the study objectives, ethical requirements, operational procedures, frequently asked questions, and emergency response plans. Before the formal survey, a pilot study involving 30 medical students was conducted to assess the average questionnaire completion time (approximately 25 minutes) and to refine the questionnaire’s layout and instructions based on participant feedback.

Following the sampling design, systematic recruitment procedures were implemented to ensure standardized data collection. Survey notifications and approvals were first coordinated through the student affairs departments. Recruitment primarily took place in designated class meetings. University counselors assisted the research team by introducing the study to the students and distributing the survey via QR codes. An informed consent form was presented on the first page of the online survey, clearly outlining the study purpose, the voluntary nature of participation, and privacy protection measures.

The professional version of the Wenjuanxing platform was used for questionnaire design and data collection. To facilitate systematic tracking across the strata, separate QR code links were generated for different grades and majors. During the data collection sessions, trained investigators were physically present (on-site) in the classrooms to provide guidance and address any immediate questions from the students, while 2 dedicated staff members handled any online inquiries. To ensure data quality, IP restrictions were implemented so that each device could complete the survey only once, and the acceptable completion time was limited to 15‐30 minutes. Mandatory response settings were applied to prevent missing critical information, and one logic-check item was included to monitor response attentiveness. The status of questionnaire returns was monitored daily, and any anomalies were addressed promptly.

The study protocol was approved by the Ethics Committee of Shanghai Jiao Tong University School of Medicine (no. SJUPN201907). Electronic informed consent was obtained from all participants prior to their inclusion in the study. To ensure privacy and confidentiality, all collected data were strictly anonymized, and no personally identifiable information was collected. Student participation was entirely voluntary, and therefore, no financial compensation was provided. Psychological counseling resources were made available, and high-risk cases identified during the screening process were promptly referred for further evaluation and intervention.

Data were presented as mean (SD) for normally distributed continuous variables and as median (IQR) for nonnormally distributed variables. Categorical variables were summarized using frequencies and percentages. Content validity of the initial SMCVS was assessed prior to the main psychometric analyses. A panel of 6 experts in the relevant field independently rated the relevance of each item. Item-level content validity indices (I-CVI) and the scale-level content validity index/average (S-CVI/Ave) were computed as the proportion of experts rating an item as relevant. Item analysis of the SMCVS included descriptive statistics, corrected item-total correlations, and Cronbach α if item deleted for all 13 items.

To evaluate construct validity, the total sample (n=1766) was randomly split into two subgroups for exploratory factor analysis (EFA; n=883) and confirmatory factor analysis (CFA; n=883). Chi-square tests were performed to compare demographic differences between the two groups. The suitability for factor analysis in the EFA subsample was assessed using the Kaiser-Meyer-Olkin test and Bartlett test of sphericity. Principal component analysis with Promax rotation was used to extract common factors, retaining components with eigenvalues greater than 1. Items with factor loadings below 0.4 and/or high cross-loadings were considered for removal or further investigation [31]. CFA was conducted using maximum likelihood estimation. Model fit was evaluated using multiple indices: chi-square goodness of fit (χ²/df), goodness-of-fit index (GFI), comparative fit index (CFI), normed fit index (NFI), Tucker-Lewis index (TLI), incremental fit index (IFI), root mean square error of approximation (RMSEA), and standardized root mean square residual (SRMR). Acceptable model fit was defined as χ²/df<5, GFI, CFI, NFI, TLI, and IFI>0.9, RMSEA<0.08, and SRMR<0.05 [32].

Measurement invariance of the SMCVS across gender and academic major was examined using multigroup CFA. For each grouping variable, a sequence of configural, metric, and scalar models was tested. Model fit was evaluated using χ²/df, CFI, and RMSEA, and changes between nested models were assessed with ΔCFI and ΔRMSEA. Values of ΔCFI≤0.01 and ΔRMSEA≤0.015 were taken as evidence of acceptable invariance across groups [33,34]. The criterion validity of the SMCVS was assessed using the PHQ-9 and GAD-7 as external criterion measures. Reliability was assessed through internal consistency reliability (Cronbach α coefficient), split-half reliability (Spearman-Brown coefficient), and test-retest reliability (intraclass correlation coefficient and correlation coefficient).

Receiver operating characteristic (ROC) curve analyses and the area under the curve (AUC) were used to evaluate the predictive ability of the SMCVS total and subscale scores for depression and anxiety, using established clinical cutoff points for the PHQ-9 and GAD-7 as external criteria. For each outcome, the optimal cutoff score for the SMCVS was determined by maximizing the Youden index (sensitivity + specificity − 1), and the corresponding sensitivity, specificity, positive likelihood ratio (LR+), and negative likelihood ratio (LR−) were calculated. All analyses were performed using SPSS 26.0, Amos 26.0, and R (version 4.3.2; pROC package for ROC analyses), with statistical significance set at P<.05 (2-tailed).

A total of 1766 valid questionnaires were collected, yielding a response rate of 92.4%. The sample included 660 participants from Shanghai Jiao Tong University School of Medicine, 473 from Shanghai Sanda University, 252 from the University of Shanghai for Science and Technology, 248 from Shanghai University of Political Science and Law, and 133 from Shanghai University of Traditional Chinese Medicine.

The demographic characteristics of the participants are presented in Table 1. Participants were aged 18‐24 years, with a mean age of 19.87 (SD 1.52) years. The participants consisted of 857 (48.5%) males and 909 (51.5%) females. Nearly half of the participants were enrolled in medical-related disciplines (n=793, 44.9%) and were only children (n=885, 50.1%). Most participants (n=1599, 90.5%) reported that their parents were married. The majority (n=1156, 65.5%) reported a medium family economic status. In addition, 45.5% (n=803) of fathers and 39.4% (n=695) of mothers had attained at least a college-level education. Overall, 37.4% (n=660) of participants reported good academic performance and 51.5% (n=910) reported moderate levels of academic stress. Furthermore, 7.1% (n=125) of participants reported smoking behavior and 24.8% (n=438) reported alcohol consumption.

Content validity of the initial 17-item SMCVS was evaluated by a panel of 6 experts in the field, who rated the relevance of each item. In line with commonly recommended standards for the I-CVI ≥0.78 [35], items meeting this criterion were retained and items below this threshold were deleted, resulting in a 13-item version of the SMCVS. The wording of several retained items was slightly revised to improve clarity and alignment with the target population. For these 13 items, the I-CVI values ranged from 0.833 to 1, and the S-CVI/Ave was 0.949, indicating acceptable overall content validity.

Table 2 presents the descriptive statistics, corrected item-total correlations, and Cronbach α if item deleted for all 13 items of the SMCVS. The mean scores for the items ranged from 1.41 to 3.01, with SDs between 1 and 1.53. Corrected item-total correlations ranged from 0.385 to 0.897, indicating moderate to strong item-total relationships. Notably, item 1 demonstrated the lowest corrected item-total correlation (r=0.385), whereas items 7 and 8 showed the highest correlations (r=0.897). The Cronbach α values if item deleted ranged from 0.953 to 0.969. The internal consistency of the scale would not be substantially improved by removing any single item.

As shown in Table 6, the SMCVS total score was strongly correlated with its 2 subscales, “reputation and privacy violation” (r=0.964, P<.01) and “information harassment” (r=0.856, P<.01). The 2 subscales demonstrated a moderate correlation with each other (r=0.688, P<.01). In terms of psychological indicators, the SMCVS total score showed moderate positive correlations with depression (r=0.484, P<.01) and anxiety (r=0.389, P<.01). Both subscales exhibited similar correlation patterns with these psychological indicators, with correlation coefficients ranging from 0.356 to 0.457 (all P<.01).

The internal consistency reliability of the SMCVS, as indicated by Cronbach α, was 0.959, with Cronbach α coefficients for the two subscales of 0.982 and 0.828, respectively. The split-half reliability, measured using the Spearman-Brown coefficient, was 0.973, and the split-half reliability coefficients for the two dimensions were 0.986 and 0.856, respectively. The test-retest reliability of the SMCVS was assessed in 37 medical students with a 3-week interval between measurements. The correlation coefficient between the two measurements was r=0.86 (P<.001), and the intraclass correlation coefficient was 0.9 (95% CI 0.805‐0.948).

ROC analyses were conducted to assess the predictive accuracy of the SMCVS and its subscales for depression and anxiety, as shown in Figure 3. For depression screening, the AUC for the SMCVS total score was 0.738 (95% CI 0.709‐0.767), while the AUC values for the “information harassment” and “reputation and privacy violation” subscales were 0.721 (95% CI 0.692‐0.750) and 0.692 (95% CI 0.663‐0.722), respectively, indicating moderate predictive ability. The optimal cutoff score for the SMCVS total score was 19.5, which yielded a sensitivity of 0.646, a specificity of 0.724, LR+ of 2.339, and LR− of 0.489. For anxiety screening, the AUC for the SMCVS total score was 0.739 (95% CI 0.704‐0.775), with the “information harassment” and “reputation and privacy violation” subscales showing AUC values of 0.727 (95% CI 0.691‐0.762) and 0.7 (95% CI 0.663‐0.736), respectively, which also reflected moderate discriminative accuracy. The optimal cutoff score for the SMCVS total score was 22.5, with a sensitivity of 0.610, a specificity of 0.763, LR+ of 2.570, and LR− of 0.511. The AUC values and diagnostic indices for the subscales are also summarized in Table 7.

This study adapted and validated a 2D SMCVS for use among Chinese college students. Our findings support the 2-factor structure, comprising “information harassment” and “reputation and privacy violation,” and demonstrate generally sound psychometric properties and meaningful associations with psychological outcomes.

The psychometric properties of the SMCVS support its utility as a measurement tool. The high factor loadings (>0.70) and absence of cross-loadings indicate clear dimensional differentiation. The cumulative variance explained (82.2%) exceeds typical thresholds in scale development research, suggesting comprehensive coverage of the construct [36]. The confirmation of this structure through CFA with satisfactory fit indices further supports the scale’s structural validity. The emergence of 2 distinct dimensions reflects the evolving nature of cyberbullying in the social media era [37]. Information harassment, characterized by direct aggressive communications, represents traditional forms of cyberbullying adapted to social media contexts. Reputation and privacy violation, involving the unauthorized sharing, distortion, or public exposure of personal information and damaging content, captures the unique affordances of social media platforms that enable new forms of victimization targeting both one’s online image and privacy. This dimensional structure advances our understanding of CV by distinguishing between direct aggressive acts and privacy-related violations. In addition, multigroup CFA supported measurement invariance across gender, indicating that factor structure, factor loadings, and item intercepts are comparable and that latent scores can be meaningfully compared between male and female students. For academic major, configural and metric invariance were supported, but scalar invariance was not achieved; therefore, while the structure appears stable across majors, direct mean comparisons of latent scores between different majors should be interpreted with caution, and future studies are needed to further examine and improve cross-major measurement invariance [38].

Some psychometric findings warrant cautious interpretation. Item analysis showed that item 1 had a lower corrected item-total correlation (0.385) than the other items, slightly below the commonly cited cutoff of 0.4 [31]. This item was retained because it captures a broad and relatively mild form of negative experience on social media that may precede or coexist with more explicit cyberbullying behaviors, thereby contributing to content validity. Empirically, item 1 still showed an acceptable loading on its intended factor in the EFA (0.826), and deleting it led to only a negligible change in Cronbach α for the total scale. In addition, although the total and subscale α coefficients were very high (all >0.8), which is often viewed as excellent reliability, such values may also indicate some item redundancy. In particular, the factor loadings for factor 1 were extremely high (0.9‐0.947), suggesting a high degree of homogeneity among the factor 1 items and potential overlap in the information they capture. Future studies could therefore consider removing or rewriting some factor 1 items to reduce redundancy and improve the informational efficiency of the scale. Finally, the CFA yielded a χ²/df value of 6.0, slightly above the commonly used lenient criterion. This may partly reflect the well-known sensitivity of chi-square to large samples and model complexity [39]; however, given that other fit indices (eg, CFI, TLI, RMSEA, and SRMR) were within acceptable ranges, the overall model fit is considered adequate, while potential local item dependence should be examined in future research.

Scholars have been actively involved in the development of measurement scales for cyberbullying and victimization in recent decades. International measurement scales included the European Cyberbullying Intervention Project Questionnaire, designed for use across 6 European countries [40]; the Cyberbullying Triangulation Questionnaire, created for Spanish students [41]; the Cyber Victimization Experiences and Cyberbullying Behaviors Scales, tailored for British teenagers [42]; the Bullying and Cyberbullying Scale for Adolescents, developed for Australian adolescents [43], and so on [15]. Chinese researchers were also dedicated to developing measurement tools tailored to local culture and national conditions. Lam and Li [16] developed the first scales specifically for CV among Chinese adolescents, known as the E-Victimization Scale. The E-Victimization Scale consisted of only 5 items and was validated as a unidimensional construct. However, the scale was limited to specific social networking platforms, such as email, texting, short messages, and websites like Renren, and focused on a narrow range of cyberbullying behaviors [16]. Rao and colleagues [44] developed a self-designed questionnaire that identified 7 forms of cyberbullying and investigated the prevalence of these 7 types of cyberbullying behaviors among Chinese middle and high school students. Unlike these earlier scales, which primarily targeted adolescents in high school, the scale developed in this study specifically addressed CV among university students. The SMCVS not only removed restrictions on specific social networking platforms but also expanded the scope of behaviors it measured, including harassment, denigration, defamation, sexual harassment, teasing, intrusion, privacy violations, and betrayal. This study conceptualized social media CV as a 2D construct, distinguishing “information harassment” from “reputation and privacy violation.” Although our data do not permit a direct comparison of predictive accuracy between the 2D SMCVS and Lam and Li’s [16] unidimensional model, the differential associations of the two SMCVS dimensions with depression and anxiety suggest a more nuanced characterization of psychological risk. Compared with 7-category behavioral classifications, this 2-factor structure is more parsimonious while still capturing core forms of social media cyberbullying, though future work is needed to examine whether additional, more specific subdimensions further improve the explanation of mental health outcomes.

The differential associations between SMCVS dimensions and psychological outcomes provide important insights into the nature of CV. In contrast to traditional bullying, cyberbullying does not rely on direct interpersonal contact and often provides perpetrators with complete anonymity. The rapid dissemination of harmful messages and images across online platforms can expose victims to a wider audience, intensifying their feelings of vulnerability. Moreover, the pervasive nature of the digital environment makes it significantly harder for victims to evade cyberbullying [44]. Studies have shown that many victims of cyberbullying are also victims of traditional bullying [45]. Psychological health issues among cyber-victims are widespread and may even be more severe than those resulting from face-to-face bullying [46]. The study found an association between CV and depression and anxiety among college students, which was consistent with previous research findings [47,48]. Multivariate logistic regression revealed that participants with depression (odds ratio [OR] 1.40, 95% CI 1.17‐1.67) and anxiety (OR 1.67, 95% CI 1.38‐2.03) were associated with a higher risk of experiencing cybervictimization [48]. General Strain Theory is one of the most commonly applied frameworks for examining the outcomes of cyberbullying [49]. The theory posits that negative interactions or relationships with others—such as being involved in cyberbullying or experiencing CV—can create strain, which may result in adverse emotional reactions (eg, depression or anxiety) and maladaptive coping strategies (eg, substance use, academic underperformance, or deviant behaviors) [50]. The ROC analyses demonstrated acceptable but moderate discriminative accuracy for both depression (AUC=0.738) and anxiety (AUC=0.739), suggesting that the SMCVS may serve as a supplementary screening tool, rather than a standalone diagnostic instrument, for screening students at risk for psychological distress.

Adaptation and validation of the SMCVS specifically tailored for Chinese college students holds significant theoretical and practical value. Given the unique cultural, social, and technological context in China, such a scale can provide a more accurate assessment of cyberbullying experiences and behaviors in this population and may be used in contexts such as academic research, campus mental health services, and policy-related discussions. For instance, it may facilitate the identification of at-risk individuals and the evaluation of prevention and intervention programs, as well as examinations of the prevalence and characteristics of cyberbullying in Chinese higher education institutions. In this context, the ROC analyses provide preliminary support for the screening potential of the SMCVS: the total score showed moderate accuracy in identifying students with depression and anxiety, with optimal cutoff scores of 19.5 and 22.5, respectively, offering a reasonable balance between sensitivity and specificity in this sample and thereby helping to flag students who might benefit from further psychological assessment. Nevertheless, the SMCVS as a screening tool should be used cautiously—higher sensitivity may increase false positives, whereas higher specificity may increase false negatives, meaning that some distressed students could still be missed. Moreover, these cutoff points are sample- and prevalence-dependent estimates derived from cross-sectional data and should not be regarded as definitive; their performance and practical implications need to be re-evaluated in settings with different base rates of depression and anxiety and, ideally, within prospective designs. In clinical and service settings, a 2-stage strategy may be preferable, whereby the SMCVS is used for initial screening followed by more comprehensive clinical or psychological assessment for individuals who screen positive.

Several limitations should be noted. First, the cross-sectional design of this study precludes causal inferences about the relationships between CV and psychological outcomes. Longitudinal studies are needed to establish the directionality and underlying mechanisms of these associations. Second, the study was limited to Chinese college students, and the SMCVS was adapted and validated within the mainland Chinese social media ecosystem, where platforms such as WeChat, QQ, and Xiaohongshu are predominant, which may restrict the generalizability of the findings to other cultural contexts. As patterns of use and platform features may differ across social media environments, the applicability of the SMCVS to other platforms and cultural contexts remains to be further examined. Cross-cultural and cross-platform validation studies including participants from more varied backgrounds would help enhance the scale’s utility and the universality of the results. Third, the reliance on self-report measures may introduce recall bias, as participants may not accurately remember past experiences, and social desirability effects, where they may underreport or overreport certain behaviors to present themselves in a favorable light. Using multimethod assessments, such as peer reports or observational data, could help address these issues. Fourth, although this study did evaluate the test-retest reliability of the SMCVS, this analysis was conducted in a relatively small subsample of 37 medical students. The high correlation coefficient therefore provides only preliminary evidence regarding the temporal stability of the scale and may not be fully representative of the broader college student population or different academic majors. Future research should replicate the test-retest assessment in larger and more diverse samples to further confirm the robustness and generalizability of the scale’s stability over time.

This study establishes the SMCVS as a psychometrically sound and culturally relevant instrument for assessing social media cyberbullying victimization among Chinese college students. By validating its 2D structure, this research captures the specific manifestations of cyberbullying within the contemporary Chinese social media ecosystem. Consequently, it provides researchers with a reliable tool to track victimization trends and evaluate the effectiveness of future preventive programs. Furthermore, the significant associations found between CV and adverse psychological outcomes highlight an urgent need for action. These findings strongly advocate for the integration of targeted cyberbullying screening and coping strategies into university mental health initiatives, ultimately fostering a safer digital environment and supporting the overall psychological well-being of young adults.