Adolescents Experiencing Both Bullying and Childhood Trauma: Longitudinal Patterns,Psychosocial Networks,and Risk Prediction

Abstract

Adolescents who experience both bullying at school and maltreatment within families are at heightened risk for psychological harm, yet little is known about how these dual adversities co-develop and influence psychosocial functioning over time. This three-wave longitudinal study followed 1,246 middle school students in central China to identify joint trajectories of bullying victimization and childhood trauma, examine their psychosocial network characteristics, and predict high-risk membership. Using KmL3D clustering, three distinct developmental trajectories were identified: (a) lower dual trauma declining, (b) high childhood trauma fluctuating, and (c) higher dual trauma declining. Network analyses revealed that protective ties among teacher–student relationships, peer attachment, gratitude, sense of control, and emotional regulation self-efficacy were strongest in the low-risk group but progressively weakened under chronic adversity, while depression and psychache became central nodes of vulnerability. To predict trajectory membership, six machine learning models were tested; the Random Forest model achieved the highest accuracy (Area Under the Receiver Operating Characteristic Curve [AUC] = 0.97), highlighting emotional distress and poor peer attachment as dominant early warning indicators. Findings underscore that poly-victimization trajectories are heterogeneous and embedded in distinct psychosocial systems. Early identification of emotional pain and relational deficits, combined with trauma-informed, school-based interventions, may help prevent re-victimization and long-term psychological harm among adolescents facing dual adversity.

Keywords

poly-victimization bullying childhood trauma joint trajectories psychosocial networks prediction

Introduction

Victimization in school settings and trauma within the family are two major adverse experiences that critically shape adolescent mental health. Systematic reviews and meta-analyses indicate that bullying victimization prospectively predicts later depression and suicidality, and is associated with broader adverse health outcomes (Gini & Pozzoli, 2013; Moore et al., 2017; Ttofi et al., 2011; van Geel et al., 2014, 2022). Ecological systems theory (Bronfenbrenner, 1979) conceptualizes family and school as proximal microsystems, suggesting that the co-occurrence of bullying victimization and family trauma may result in compounding risk. Empirical studies support this view, indicating that adolescents who experience childhood trauma are more likely to be bullied by peers (Finkelhor et al., 2009; Radford et al., 2013). Moreover, poly-victimized adolescents—those exposed to both school bullying and childhood trauma—display significantly higher risks for depression, anxiety, self-injury, and suicidal behaviors (Debowska et al., 2018; Suliman et al., 2009). However, much of the existing literature relies on cross-sectional or single-context designs, overlooking how these adversities jointly evolve over time. Life-course theory (Elder, 1998) argues that early adversity accumulates developmentally and interacts across life domains, contributing to persistent vulnerability patterns. Despite growing interest in multi-victimization (Finkelhor et al., 2009), there remains a lack of longitudinal evidence on the joint developmental trajectories of bullying victimization and childhood trauma.

Joint Trajectories of Bullying Victimization and Childhood Trauma

Bullying victimization is not a static experience but a developmentally dynamic phenomenon that evolves across childhood and adolescence. Longitudinal studies have identified several distinct victimization trajectories: some individuals are bullied only during early childhood and recover over time; others begin to experience victimization in late adolescence; a minority endure persistent bullying from childhood through adolescence (Brendgen et al., 2023; Cheng et al., 2025a). These variations highlight substantial individual heterogeneity and underscore the need for longitudinal approaches to capture these divergent patterns. According to ecological systems theory (Bronfenbrenner, 1979) and life-course theory (Elder, 1998), early adversity may become embedded through interactions across family, school, and peer microsystems, reinforcing chronic victimization over time. Although prior research has primarily focused on school-based bullying trajectories (Zhou et al., 2022), the parallel development of childhood trauma—especially emotional and physical abuse within the family—remains underexplored. Cross-sectional studies have shown high co-occurrence rates between bullying and childhood trauma, with typologies such as “dual-high,” “dual-low,” and “single-domain high” emerging across populations (Salmon et al., 2023). However, there is a lack of dynamic evidence to verify whether and how these patterns evolve over time. Therefore, it is necessary to track the joint evolution of school bullying victimization and childhood trauma within the same longitudinal framework to identify the most vulnerable profiles.

Previous studies have mainly used Latent Class Growth Analysis and Growth Mixture Modeling to depict developmental trajectories of childhood trauma and school bullying victimization (Bravo & Gómez, 2024; Huang & Lawrence, 2024; M. Wang & Smith, 2024). While informative, these approaches require researchers to predefine the number and shape of trajectories and are typically restricted to univariate outcomes. To address these limitations, the present study adopts KmL3D, a nonparametric clustering algorithm that identifies multivariate longitudinal patterns without parametric assumptions (Genolini et al., 2015). This method has been successfully applied in prior research—for example, Chow et al. (2022) used KmL3D to identify five joint trajectories of bullying perpetration and victimization in a European cohort—demonstrating its flexibility and empirical robustness. Building on this strength, we apply KmL3D to examine the longitudinal co-development of childhood trauma and school bullying victimization in early adolescence.

Risk Predictors of Joint Victimization Trajectories

To move beyond describing joint victimization trajectories and toward prevention, it is essential to identify factors that prospectively differentiate adolescents’ likelihood of following distinct joint victimization pathways. Accordingly, the present study considers a set of theoretically and empirically grounded predictors that may increase or decrease the likelihood of experiencing joint victimization. These predictors span three domains: external support, internal resources, and proximal psychological states.

At the contextual level, ecological systems theory (Bronfenbrenner, 1979) posits that adolescents’ family and school microsystems jointly shape developmental outcomes. Supportive ties with teachers and peers can buffer against adversity, yet existing studies rarely examine whether the absence of such ties forecasts vulnerability to co-occurring victimization across multiple contexts (Boulton et al., 2014; Li et al., 2024). We therefore test teacher–student relationships and peer attachment as contextual predictors of joint victimization risk.

At the individual resource level, self-determination theory (Ryan & Deci, 2017) and the control–value theory of achievement emotions (Pekrun, 2006) highlight the importance of autonomy, competence, and relatedness. Gratitude, sense of control (SoC), and regulatory emotional self-efficacy capture these psychological needs. While these resources are known to foster resilience, their predictive role in explaining vulnerability to multiple victimization has received little empirical attention (Raines et al., 2022; Rey et al., 2019; M. Wang et al., 2018).

At the proximal psychological-state level, developmental psychopathology emphasizes that mental-health states such as depressive mood and psychache may function as early markers of dysregulation, signaling heightened vulnerability and potential escalation in adverse experiences (Cicchetti & Toth, 2016). Conversely, meaning in life reflects protective processes that mitigate the negative psychological effects of adversity (Cheng et al., 2025a, 2025b; Zhang et al., 2023). By treating these indicators as predictors rather than mere outcomes, the present study extends prior research on adolescent mental health.

Network Analysis and Machine-Learning Prediction

Methodologically, traditional regression approaches have emphasized explanation while neglecting prediction. Yet, theoretical frameworks underscore that multiple risk and protective factors operate in combination. To address this gap, network analysis is used to map the structural interrelations and identify the most central predictors, whereas machine-learning models assess predictive accuracy, rank relative importance, and detect nonlinear interactions. Network analysis conceptualizes measured psychosocial variables as nodes and their conditional associations as edges, and regularized partial-correlation network models are commonly used to obtain more parsimonious and interpretable structures in psychological data (Epskamp & Fried, 2018). Complementarily, machine-learning approaches emphasize predictive performance—often evaluated via out-of-sample validation—and can flexibly accommodate nonlinearities and higher-order interactions, aligning with calls to prioritize prediction alongside explanation in behavioral research (Breiman, 2001; Yarkoni & Westfall, 2017). Consistent with emerging work in the bullying literature, psychological network approaches have been used to delineate symptom-/item-level links between victimization experiences and depressive symptoms and to identify key “bridge” features in these associations (Ren et al., 2023), and machine-learning studies have begun to predict bullying victimization risk at scale and to rank multi-domain risk/protective factors (Qiu et al., 2024; Yan et al., 2023). Together, these approaches translate theoretical claims about multifactorial risk into empirically testable models. Whereas prior work has largely focused on single-domain victimization or single outcomes, the present study employs network analysis to reveal the core features of joint victimization trajectories and applies machine-learning models to evaluate the relative predictive power of psychosocial factors, thereby identifying which indicators most effectively forecast adolescents’ risk of joint victimization.

The Current Study

Building on the above foundations, the present study seeks to clarify how school bullying victimization and childhood trauma jointly unfold in adolescence and which psychosocial factors predict these processes. To this end, we use KmL3D to identify heterogeneous longitudinal trajectories, providing novel evidence on the co-evolution of bullying victimization and childhood trauma. We then apply psychological network analysis to examine the structural configurations of external supports, internal resources, and psychological states, thereby advancing understanding of the mechanisms that differentiate vulnerable profiles. Finally, we employ machine-learning models to evaluate the relative predictive importance of these psychosocial factors, offering a predictive framework that identifies early indicators of joint victimization risk and informs timely intervention in school and family contexts.

Method

Participants

This study was based on a 1-year, three-wave longitudinal design. Data were collected from seventh- and eighth-grade students at a middle school in Henan Province between 2020 and 2021, with equal intervals of 6 months between each wave. With approval from the Ethics Committee of the first author’s University (ZZUIRB2020-021), participants were recruited using a cluster sampling method in a single junior middle school. Specifically, grade levels served as the sampling clusters, and all eligible students in Grades 7 and 8 in the selected school were invited to participate. After obtaining consent from school principals and homeroom teachers, written informed consent was obtained from the parents of 1,414 students.

At each wave, questionnaires were administered on site by trained graduate students in psychology, who provided standardized instructions and emphasized that participation was voluntary and that students could withdraw at any time. After completion, a brief psychological debriefing or support was offered to reduce any potential discomfort; students indicating marked distress were identified for follow-up and referred to school mental health staff for further support and intervention.

A total of 1,246 students (51.77% female; M age = 13.58 years, SD = 0.75) completed all three waves of data collection. The remaining 168 students did not complete the full assessment due to illness, withdrawal, school transfer, or other reasons. Little’ s Missing Completely at Random (MCAR) test indicated that the missing data were completely at random, χ²(11) = 12.80, p = .31. Gender distribution did not differ between completers and dropouts, χ²(1) = .005, p = .946. For continuous baseline variables, group differences were generally nonsignificant (t = −1.62 to 1.75, ps = .081 – .883). In addition, we fitted a logistic regression model to examine whether baseline characteristics predicted attrition. In the multivariable model including all baseline characteristics, none of the predictors significantly predicted attrition (ORs = 0.98 – 1.06; 95% CIs spanning 1.00; ps = .175 – .998).

Measures

Bullying victimization was measured using the Delaware Bullying Victimization Scale, and childhood trauma was assessed with the Childhood Trauma Questionnaire. External support factors were measured using the Teacher–Student Relationship Scale and the Peer Attachment Scale. Internal resources and coping capacity were assessed using the Gratitude Questionnaire (CGQ-6), the SoC Scale, and the Emotional Regulation Self-Efficacy Scale. Mental health outcomes were measured using the Psychache Scale, the Center for Epidemiologic Studies Depression Scale (CES-D), and the Meaning in Life Questionnaire. Detailed information about these instruments is provided in the Supplemental Appendix.

Data Analysis

All analyses were conducted in R (version 4.4.1; R Core Team, Vienna, Austria) and SPSS (version 25.0; IBM Corp., Armonk, NY, USA). Missingness was evaluated using Little’s MCAR test including all variables used in the primary analyses: demographics (gender), repeated measures of bullying victimization and childhood trauma across the three waves, and baseline psychosocial covariates (psychache, depression, SoC, emotion regulation self-efficacy, peer attachment, teacher–student relationship, gratitude, and meaning in life). Missing data were handled using multiple imputation by chained equations (MICE; m = 5, 50 iterations, seed = 500). Imputation was performed at the item level: item responses were imputed first, and scale scores were computed after imputation according to the scoring rules. The imputation model included all items from the focal measures (bullying victimization and childhood trauma across waves) as well as baseline demographics and baseline psychosocial covariates as predictors in the chained equations.

To identify the joint developmental trajectories of bullying victimization and childhood trauma, we applied the KmL3D clustering algorithm, a non-parametric k-means approach that clusters individuals based on the similarity of their joint longitudinal trajectories across repeated measurements. Prior to clustering, we enabled the built-in standardization option to place bullying victimization and childhood trauma on a comparable scale for Euclidean-distance clustering. In each iteration, participants were assigned to the nearest cluster centroid (trajectory center) and centroids were updated until convergence, thereby minimizing within-cluster dissimilarity in joint trajectories. To improve robustness and reproducibility, we performed 20 random redrawings and set a fixed random seed (set.seed[123]). Solutions with two to six clusters were tested, and the optimal model was selected based on a combination of fit indices (Calinski–Harabasz, Bayesian information criterion [BIC], Akaike information criterion [AIC], Ray–Turi, Davies–Bouldin), posterior probabilities (≥.70), and class size (≥5%). KmL3D was considered suitable for the present study because it does not require specifying a parametric growth form (e.g., linear/quadratic), and it directly captures heterogeneity in the joint evolution of two processes (bullying victimization and childhood trauma) over time.

Baseline differences across trajectory groups were then examined using one-way ANOVAs and chi-square tests on demographic, psychosocial, and mental-health variables. To further explore interdependencies within each group, we estimated psychological networks with the EBICglasso procedure and compared them across groups using the Network Comparison Test (NCT).

Finally, to predict trajectory group membership, we first randomly split the data into a training set (70%) and a testing set (30%). To avoid information leakage, ROSE resampling was applied only within the training set to mitigate class imbalance, and the untouched testing set was used for final performance evaluation. Six classifiers (multinomial logistic regression, random forest [RF], support vector machine [SVM], naïve Bayes, k-nearest neighbors [KNN], and LightGBM) were implemented, and their hyperparameters were tuned using 10-fold cross-validation conducted on the training data. During cross-validation, we retained out-of-fold (OOF) predicted probabilities for each participant and computed both discrimination and calibration metrics from these OOF predictions to avoid optimistic bias. Specifically, we evaluated discrimination using multiclass AUC (along with sensitivity, specificity, predictive values, Cohen’s kappa, and balanced accuracy) and assessed calibration using the multiclass Brier score and log loss (cross-entropy), where lower values indicate better probability accuracy and calibration. After selecting the optimal hyperparameters, the final model was refit on the full resampled training set and evaluated on the untouched testing set to report final performance. To further interpret the machine-learning models, we examined variable importance and partial dependence plots to identify the most influential predictors and their interactions.

Results

Joint Growth Trajectories of Bullying Victimization and Childhood Trauma

This study tested longitudinal measurement invariance across three time points for childhood trauma and bullying victimization. The results showed that changes in the Comparative Fit Index (CFI) and the Root Mean Square Error of Approximation (RMSEA) across models met the recommended thresholds (ΔCFI ≤ 0.01, ΔRMSEA ≤ 0.015, see Table S1), indicating that the assumption of measurement invariance was satisfied, thus allowing for subsequent analyses.

Based on model fit indices (Table S2) and classification quality (Table S3), a three-cluster solution was selected. The Calinski–Harabasz index was high for this model, while BIC and AIC values were lowest, indicating the best balance of fit and parsimony. Ray–Turi and Davies–Bouldin indices also supported clear separation and compactness. Posterior probabilities exceeded .70 and class sizes were all above 5%, confirming that the three-cluster solution provided the most robust and interpretable structure.

The study identified three joint trajectories of bullying victimization and childhood trauma (see Figure 1): lower dual trauma declining, high childhood trauma fluctuating, and higher dual trauma declining. The lower dual trauma declining group (71.75%) was characterized by gradually decreasing levels of bullying victimization that approached zero over time, alongside consistently low and declining levels of childhood trauma (bullying = 0.13, 0.02, 0.00; trauma = 0.72, 0.64, 0.63 at T1–T3). The high childhood trauma fluctuating group (19.98%) showed a decline in bullying victimization from T1 to T2, followed by an increase from T2 to T3 (bullying = 0.31, 0.15, 0.23 at T1–T3). By contrast, childhood trauma decreased from T1 to T2, but then rose to a peak at T3 (trauma = 1.28, 1.22, 1.50 at T1–T3). The higher dual trauma declining group (8.27%) experienced a sharp drop in bullying victimization after peaking at T1, while childhood trauma levels remained high initially but declined steadily over time (bullying = 1.45, 1.15, 0.83; trauma = 1.48, 1.39, 1.31 at T1–T3). Using the scaled means to index coupling or decoupling, we computed Δ_scaled (bullying − trauma). The lower dual trauma declining group showed the smallest gaps (Δ_scaled = −0.59 to −0.63), whereas larger gaps emerged in the high childhood trauma fluctuating group (Δ_scaled = −0.97 to −1.27) and the higher dual trauma declining group (Δ_scaled = −0.03 to −0.48), with the latter diverging over time.

Characteristics of Joint Trajectory Groups

Baseline characteristics and network structures were compared across the three joint trajectory groups. As shown in Table S4, significant group differences emerged in gender: the higher dual trauma declining group (C) had the highest proportion of males (68.93%), exceeding group A (46.42%), group B (46.18%), and the full sample (48.23%). Differences were also found in external support, internal resources, and mental health indicators. Group A (lower dual trauma declining) showed the most favorable profile, with higher support and resources and lower distress and depression. Group B (high childhood trauma fluctuating) had the lowest meaning in life, whereas group C had the weakest support and resources, along with the highest distress and depression.

Network analysis (Figure 2, Figure S1) revealed clear group-specific patterns when considering expected influence (EI). In the lower dual trauma declining group (A), protective factors such as teacher–student relationship (TSR) and emotional regulation self-efficacy (RES) showed the highest EI, while depression (DEP) exerted a negative influence, suggesting a resilient network configuration. In the high childhood trauma fluctuating group (B), protective nodes like SoC and RES had reduced EI, whereas psychache (PD) and depression emerged as more influential, reflecting a risk-shifted structure. In the higher dual trauma declining group (C), TSR and SoC retained relatively high EI, but meaning in life (MIL) and gratitude (GRAT) showed diminished influence, and depression continued to anchor the network negatively, indicating both resource depletion and heightened vulnerability. Network connectivity differed descriptively across groups (global strength: A = 2.887, B = 2.288, C = 2.082). However, NCT indicated no significant between-group differences in network structure (M: 0.115–0.264, all p ≥ .161) or global strength (S: 0.111–0.463, all p ≥ .092). Gender-related NCT results were non-significant in Groups A and C; in Group B, males showed higher global strength (S = 1.077, p = .046) without a structural difference (M = 0.288, p = .165).

Evaluation of Machine Learning Models

Six machine learning models were tested to classify the joint trajectories of bullying victimization and childhood trauma (see Figure S2). Model performance varied considerably: Random Forest (RF) and SVM achieved the best results (AUC ≈ 0.978), significantly outperforming Logistic Regression (0.963), Naïve Bayes (0.966), KNN (0.840), and LightGBM (≈ 0.500). As shown in Table S5, RF exhibited the most favorable calibration (lowest Brier and log loss), closely followed by SVM and Naïve Bayes (OOF Brier/log loss: RF = 0.161/1.741; SVM = 0.168/2.989; Naïve Bayes = 0.181/5.270; selected hyperparameters are also reported in Table S5).

Based on class-specific performance (Table S6), both models demonstrated strong discrimination, but RF showed a modest overall advantage. At the overall level, RF achieved a higher balanced accuracy than SVM (0.93 vs. 0.92), while the two models exhibited the same Kappa coefficient (.85), indicating comparably high agreement. At the class level, RF yielded higher sensitivity in groups A and B (0.92 vs. 0.91; 0.85 vs. 0.83), supporting better detection in these classes, whereas SVM showed only a slight sensitivity advantage in group C (0.95 vs. 0.93). Specificity was largely similar across models (A: 0.95 vs. 0.93; B: 0.94 vs. 0.94; C: 0.95 vs. 0.96), and positive predictive value (PPV) followed a comparable pattern, with RF higher in A (0.91 vs. 0.88), identical in B (0.88 vs. 0.88), and SVM marginally higher in C (0.93 vs. 0.91). NPV was uniformly high and identical for both models across all classes (A: 0.95; B: 0.92; C: 0.97). Taken together, evidence from discrimination (AUC), calibration (OOF Brier/log loss), and class-wise operating characteristics (Table S6) supported RF as the primary model for subsequent interpretation and reporting.

Variable Importance

Using the Random Forest model, we examined the importance and interactions of predictors for classifying the three joint trajectory groups. Variable importance analysis (Figure 3) identified psychache (PD) and depression (DEP) as the strongest predictors, followed by peer attachment (PA). SoC, RES, GRAT, and MIL showed moderate contributions, while gender and TSR were the least important.

Univariate analyses (Figure S3) confirmed that higher PA predicted membership in the low-risk trajectory (A), whereas lower PA increased the likelihood of the high-risk trajectory (C). For PD and DEP, low levels were associated with class A, moderate levels with class B, and high levels markedly increased the probability of class C.

Interaction analysis (Figure S4) further showed that individuals with low PD and DEP were most likely in class A, moderate and fluctuating levels characterized class B, and high levels of both indicators strongly predicted membership in class C.

Discussion

This study examined how bullying victimization at school and childhood trauma within families co-develop during early adolescence by integrating longitudinal clustering, psychosocial network analysis, and machine learning. We identified three joint trajectories—lower dual trauma declining (A), high childhood trauma fluctuating (B), and higher dual trauma declining (C)—highlighting that dual adversity is heterogeneous rather than a single uniform risk pattern (Finkelhor et al., 2009; Olweus, 2013). This heterogeneity is consistent with prior longitudinal evidence that victimization unfolds in distinct developmental courses, including stable-low, fluctuating, and persistently elevated patterns (Bravo & Gómez, 2024; Brendgen et al., 2023; Huang & Lawrence, 2024).

The findings advance developmental psychopathology in three ways. First, modeling bullying and childhood trauma jointly extends ecological and life-course views by demonstrating that risks across key microsystems (family and school) can co-vary into differentiated pathways rather than accumulating monotonically (Bronfenbrenner, 1979; Cicchetti & Toth, 2016; Elder, 1998). Second, the network results suggest that resilience is not simply the presence of protective factors, but the organization of external support and internal resources into a cohesive system—where supportive ties and self-regulatory resources are densely connected and buffer distress (Epskamp & Fried, 2018; Lachman & Weaver, 1998; McCullough et al., 2002). Third, the prediction results illustrate the value of complementing explanation with risk stratification: machine learning can identify a small set of high-yield indicators for early detection, aligning with the “prediction over explanation” perspective in psychological science (Breiman, 2001; Yarkoni & Westfall, 2017) and prior bullying-risk prediction work (X. Wen et al., 2025; Yan et al., 2023).

Together, these results support an “identify–explain–predict” framework: (a) identify heterogeneous joint pathways; (b) explain pathway-specific psychosocial systems; and (c) predict membership using a parsimonious set of warning indicators. Importantly, the framework links developmental patterns to actionable leverage points within school and family contexts, which is a key goal in prevention-oriented developmental psychopathology (Cicchetti & Toth, 2016; Olweus, 2013).

Diversity and Turning Points in Joint Trajectories of Bullying Victimization and Childhood Trauma

The three trajectories suggest meaningful diversity in timing, intensity, and potential turning points of dual adversity. Group A resembles an “accumulated advantage” pathway, likely reflecting relatively stable protective contexts; this aligns with evidence that sustained low victimization is common and associated with better psychosocial adjustment (Brendgen et al., 2023; Huang & Lawrence, 2024). Group B combines high family trauma with fluctuating bullying, which is consistent with the idea that early maltreatment can heighten vulnerability to peer victimization through disruptions in emotion regulation and interpersonal trust, while school-context changes may produce temporary protection or renewed risk (Cicchetti & Toth, 2016; Finkelhor et al., 2009). Group C begins with high adversity in both domains and then declines sharply, suggesting contextual relief or adaptive coping over time; however, life-course theory cautions that early high-intensity exposure may leave enduring psychosocial “imprints” even when later exposure decreases (Cicchetti & Toth, 2016; Elder, 1998).

Observed gender patterns warrant cautious interpretation but are broadly compatible with gendered expressions of distress and vulnerability processes. For instance, prior work suggests that girls’ distress may be more tightly coupled with relational contexts, whereas boys may show more externalizing or behaviorally expressed difficulties under adversity (Connell, 2012; S. Wang et al., 2024). This implies that prevention should remain sensitive to potential gender-differentiated needs without assuming fully distinct mechanisms in all groups.

Psychosocial Network Configurations Across Joint Trajectories

Network differences across groups provide a systems-level account of why similar levels of adversity may lead to different outcomes. In Group A, external support (teacher–student relationships, peer attachment) and internal resources (gratitude, SoC, emotion regulation self-efficacy) formed a cohesive protective structure, consistent with resilience models emphasizing the buffering role of relational safety and self-regulatory capacity (Cicchetti & Toth, 2016; Lachman & Weaver, 1998; McCullough et al., 2002). Distress indicators (depression, psychache) remained peripheral, echoing evidence that supportive peer and school experiences are linked with better emotional adjustment in early adolescence (Boulton et al., 2014; Li et al., 2024).

In Groups B and C, trauma exposure coincided with weakened integration among supports and resources, while depression and psychache became central nodes. This pattern is consistent with theories of developmental psychopathology in which chronic or intense adversity undermines regulatory systems and increases the salience of internalizing processes (Cicchetti & Toth, 2016). The centrality of psychache is also conceptually meaningful: psychache captures subjective psychological pain beyond depressive symptoms and has been validated as a distinct distress dimension (Holden et al., 2001). When relational ties are depleted, psychological pain and depressive affect may become mutually reinforcing, potentially maintaining vulnerability even if external adversity later declines (Cicchetti & Toth, 2016; Moore et al., 2017).

Although overall gender network structures were not significantly different, the stronger negative linkage between peer support and depression among girls in high-trauma groups suggests that relational disruptions may be particularly consequential for girls’ emotional functioning (Connell, 2012; Li et al., 2024). For boys, a tendency toward externalized distress (e.g., disengagement) has been documented in trauma-linked risk profiles, indicating that interventions may need to broaden beyond internalizing-only screening (S. Wang et al., 2024).

Predicting Developmental Trajectories From Early Risk and Protective Factors

Machine learning highlighted psychache, depression, and peer attachment as the most informative indicators for classifying joint trajectories. This aligns with longitudinal evidence that bullying victimization is associated with later internalizing problems (Moore et al., 2017; van Geel et al., 2014, 2022) and with work showing that peer support and attachment buffers the victimization–depression link (Li et al., 2024). Importantly, the predictive profile suggests a dual-process pathway: distress indicators (psychache and depression) amplify vulnerability, whereas peer attachment acts as a protective redirector toward a low-risk course (Boulton et al., 2014; Holden et al., 2001; Li et al., 2024).

From an applied standpoint, the findings support two priorities. First, schools could incorporate brief screening for depressive symptoms and psychological pain to flag adolescents who may be at risk for sustained dual-adversity pathways, consistent with prevention frameworks emphasizing early detection (Olweus, 2013; Yarkoni & Westfall, 2017). Second, strengthening peer attachment through structured peer mentoring, classroom prosocial routines, and engagement-rich extracurricular contexts may be particularly protective, echoing prior evidence that peer support can buffer internalizing risk following victimization (Li et al., 2024). Notably, Group C may require more intensive, relationship-rebuilding supports because its network indicates depleted interpersonal resources despite declining exposure.

Policy and Practice Implications

Three implications follow for policy and school practice. (a) Adopt tiered, trauma-informed prevention: universal school climate and bullying prevention (Olweus, 2013) should be paired with trauma-informed staff training and referral pathways for students exposed to family trauma, given the cross-context co-development observed here (Bronfenbrenner, 1979; Cicchetti & Toth, 2016). Whole-school trauma-informed models (e.g., HEARTS) emphasize staff capacity-building and organizational routines that reduce retraumatization and improve relational safety, which maps onto the “support–resource integration” pattern seen in the low-risk network (Avery et al., 2020; Dorado et al., 2016). (b) Screen and respond to distress signals: brief assessments of depression and psychache may serve as high-yield early warning indicators, enabling targeted support before vulnerabilities consolidate (Holden et al., 2001; Moore et al., 2017). (c) Evaluate implementation and equity: evidence syntheses emphasize that trauma-informed school approaches vary widely and require careful evaluation of implementation quality and outcomes; schools should monitor whether benefits reach the highest-risk subgroups identified by trajectory screening (Avery et al., 2020; Maynard et al., 2019).

Limitations and Future Directions

Several limitations should be noted. First, reliance on self-reports may inflate associations via shared method variance; future studies should incorporate multi-informant reports (parents or teachers) and, where feasible, objective indicators (e.g., disciplinary records, counselor contact). Second, generalizability is limited by sampling from a single region; replication across diverse socio-economic and cultural contexts is needed. Third, the observational design precludes causal inference; future work could apply longitudinal causal methods or intervention designs to test whether strengthening peer attachment and school support causally shifts adolescents away from high-risk joint trajectories (Cicchetti & Toth, 2016).

Conclusion

Bullying victimization and childhood trauma co-develop into distinct joint trajectories during early adolescence, and these pathways are embedded in qualitatively different psychosocial systems. Cohesive integration of external support and internal resources characterizes the low-risk group, whereas depleted relational ties coincide with distress-central networks in higher-risk groups. Psychache, depression, and peer attachment emerged as key early indicators that can guide targeted, trauma-informed prevention and support in school settings, with the ultimate goal of reducing re-victimization and downstream psychological harm.

Supplemental Material

sj-docx-1-jiv-10.1177_08862605261426585 – Supplemental material for Adolescents Experiencing Both Bullying and Childhood Trauma: Longitudinal Patterns, Psychosocial Networks, and Risk Prediction

Supplemental material, sj-docx-1-jiv-10.1177_08862605261426585 for Adolescents Experiencing Both Bullying and Childhood Trauma: Longitudinal Patterns, Psychosocial Networks, and Risk Prediction by Zhongjie Wang, Kaiyuan Lu, Fuqu Liu, Xinyang Xu and Xuezhen Wang in Journal of Interpersonal Violence

Footnotes

ORCID iDs

Zhongjie Wang

Kaiyuan Lu

Xuezhen Wang

Ethical Considerations

The study was conducted in accordance with the Declaration of Helsinki and approved by the Ethics Committee of Zhengzhou University (ZZUIRB2020-021,10 December 2020). All patients provided written informed consent.

Author Contributions

Zhongjie Wang conceived of the study, participated in its design and coordination, and drafted the paper; Kaiyuan Lu conceived of the study, participated in its design and coordination, and drafted the paper; Fuqu Liu, Xinyang Xu, and Xuezhen Wang helped to draft the paper.

Funding

The authors disclosed receipt of the following financial support for the research and/or authorship of this article: This work was supported by the National Social Science Fund of China (Grant No. 24BSH116).

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the authorship and/or publication of this article.

Data Availability Statement

The datasets generated and/or analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

Supplemental Material

Supplemental material for this article is available online.

Author Biographies

Zhongjie Wang, PhD, is a faculty member at the School of Education, Zhengzhou University. Her research examines adolescent development, school climate, and the prevention of interpersonal violence in educational settings. She has led or contributed to studies on bullying involvement, teacher–student relationships, and trajectories of youth mental health, with findings published in peer‑reviewed journals. Methodologically, her work uses longitudinal designs, person‑centered approaches, and multivariate modeling to clarify how risks and protective factors co‑evolve in early adolescence.

Kaiyuan Lu is a graduate student in Mental Health Education at Zhengzhou University. His work focuses on co‑occurring bullying victimization and childhood trauma, integrating longitudinal modeling, psychological network analysis, and machine‑learning prediction. He has experience with multi‑wave school‑based surveys, data management and imputation, and classification modeling to identify high‑risk developmental pathways. His broader interests include risk–resilience processes and precision prevention in school mental health.

Fuqu Liu is a graduate student at the School of Education, Zhengzhou University. Her interests include adolescent psychosocial development, peer relations, and quantitative methods for longitudinal data. She contributes to study design, instrument adaptation, and data management, and supports statistical analyses involving latent‑variable and classification models. Recent work has focused on measuring external supports (teacher–student relationship, peer attachment) and internal resources (gratitude, sense of control, regulatory emotional self‑efficacy) and their links to mental‑health outcomes.

Xinyang Xu is a graduate student at the School of Education, Zhengzhou University. Research interests include school bullying, emotional well‑being, and the network structure of psychosocial variables in adolescence. Responsibilities include multi‑wave data collection, reliability and invariance testing, and visualization of trajectory and network results. Methodological interests include Gaussian graphical models, the Network Comparison Test, and model‑agnostic explainability for ensemble classifiers.

Xuezhen Wang is a PhD student in the School of Education, Renmin University of China, and the corresponding author of this manuscript. His research focuses on adolescent development, school mental health, and trauma‑informed education. He leads and collaborates on projects examining multi‑victimization, risk screening, and intervention design, integrating ecological perspectives with modern analytics to inform practice and policy. His work has been disseminated through peer‑reviewed outlets and collaborations with school and community partners.

References

Avery

J. C.

Morris

Galvin

Misso

Savaglio

Skouteris

(2020). Systematic review of school-wide trauma-informed approaches. Journal of Child and Adolescent Trauma, 14(3), 381–397. https://doi.org/10.1007/s40653-020-00321-1

Boulton

M. J.

Hardy

Down

Farrant

. (2014). Peer victimization during early adolescence: Associations with self-esteem and perceptions of school. British Journal of Educational Psychology, 84(3), 349–361. https://doi.org/10.1111/bjep.12023

Bravo

Á. J.

Gómez

P. L

. (2024). Bullying victimization trajectories in Spanish adolescents: A longitudinal latent class growth approach. Psicothema, 36(1), 85–92. https://doi.org/10.7334/psicothema2023.245

Breiman

(2001). Statistical modeling: The two cultures (with comments and a rejoinder by the author). Statistical Science, 16(3), 199–231. https://doi.org/10.1214/ss/1009213726

Brendgen

Vitaro

Boivin

Dionne

Tremblay

R. E

. (2023). Mental health comorbidities following peer victimization across childhood and adolescence: A 20-year longitudinal study. Psychological Medicine, 53, 1741–1750. https://doi.org/10.1017/S0033291722001729

Bronfenbrenner

(1979). The ecology of human development: Experiments by nature and design. Harvard University Press.

Cheng

Taylor

Khan

(2025a). Bidirectional associations between peer victimization and internalizing symptoms among UK adolescents: A longitudinal study. Journal of Youth and Adolescence, 54(1), 88–105. https://doi.org/10.1007/s10964-024-01832-4

Cheng

Taylor

Khan

(2025b). Longitudinal effects of bullying victimization on mental health: A latent growth modeling approach. Journal of Youth and Adolescence, 54(2), 144–160. https://doi.org/10.1007/s10964-024-01878-4

Chow

Zimmer-Gembeck

M. J.

Hunter

S. C.

Hong

J. S.

Vélez

C. E

. (2022). Joint developmental trajectories of bullying perpetration and victimization from childhood to adolescence: A kml3d longitudinal cluster analysis. Aggressive Behavior, 48(6), 655–669. https://doi.org/10.1002/ab.22004

10.

Cicchetti

Toth

S. L

. (2016). Child maltreatment and developmental psychopathology: A multilevel perspective on prevention and intervention. Development and Psychopathology, 28(4 Pt 2), 1515–1528. https://doi.org/10.1017/S0954579416000994

11.

Connell

(2012). Gender, health and theory: Conceptualizing the issue, in local and world perspective. Social Science & Medicine, 74(11), 1675–1683. https://doi.org/10.1016/j.socscimed.2011.06.006

12.

Debowska

Boduszek

Sherretts

Willmott

Jones

A. D

. (2018). Profiles and behavioral consequences of child abuse among adolescent girls in the child welfare system: Bully only, victim only, bully-victim, and unexposed groups. Child Abuse & Neglect, 85, 13–21. https://doi.org/10.1016/j.chiabu.2018.08.006

13.

Dorado

J. S.

Martinez

McArthur

L. E.

Leibovitz

(2016). Healthy Environments and Response to Trauma in Schools (HEARTS): A whole-school, multi-level, prevention and intervention program for creating trauma-informed, safe and supportive schools. School Mental Health, 8(1), 163–176. https://doi.org/10.1007/s12310-016-9177-0

14.

Elder

G. H.

Jr. (1998). The life course as developmental theory. Child Development, 69(1), 1–13. https://doi.org/10.1111/j.1467-8624.1998.tb06128.x

15.

Epskamp

Fried

E. I.

(2018). A tutorial on regularized partial correlation networks. Psychological Methods, 23(4), 617–634. https://doi.org/10.1037/met0000167

16.

Finkelhor

Ormrod

R. K.

Turner

H. A

. (2009). Lifetime assessment of polyvictimization in a national sample of children and youth. Child Abuse & Neglect, 33(7), 403–411. https://doi.org/10.1016/j.chiabu.2008.09.012

17.

Genolini

Alacoque

Sentenac

Arnaud

(2015). kml and kml3d: R packages to cluster longitudinal data. Journal of Statistical Software, 65(4), 1–34. https://doi.org/10.18637/jss.v065.i04

18.

Gini

Pozzoli

(2013). Bullied children and psychosomatic problems: A meta-analysis. Pediatrics, 132(4), 720–729. https://doi.org/10.1542/peds.2013-0614

19.

Holden

R. R.

Mehta

Cunningham

E. J.

McLeod

L. D

. (2001). Development and preliminary validation of a scale of psychache. Canadian Journal of Behavioural Science, 33(4), 224–232. https://doi.org/10.1037/h0087144

20.

Huang

Lawrence

(2024). Developmental trajectories of bullying victimization and their association with adolescent mental health: Evidence from the UK Millennium Cohort Study. Journal of Youth and Adolescence, 53(1), 45–61. https://doi.org/10.1007/s10964-023-01729-6

21.

Lachman

M. E.

Weaver

S. L

. (1998). The sense of control as a moderator of social-class differences in health and well-being. Journal of Personality and Social Psychology, 74(3), 763–773. https://doi.org/10.1037/0022-3514.74.3.763

22.

Zhang

Wang

(2024). Peer support buffers the longitudinal effect of bullying victimization on depressive symptoms among Chinese early adolescents: A two-year study. Journal of Adolescence, 103(5), 120–131. https://doi.org/10.1016/j.adolescence.2024.03.031

23.

Maynard

B. R.

Farina

Dell

N. A.

Kelly

M. S.

(2019). Effects of trauma-informed approaches in schools: A systematic review. Campbell Systematic Reviews, 15(1–2), Article e1018. https://doi.org/10.1002/cl2.1018

24.

McCullough

M. E.

Emmons

R. A.

Tsang

J. A

. (2002). The grateful disposition: A conceptual and empirical topography. Journal of Personality and Social Psychology, 82(1), 112–127. https://doi.org/10.1037/0022-3514.82.1.112

25.

Moore

S. E.

Norman

R. E.

Suetani

Thomas

H. J.

Sly

P. D.

Scott

J. G.

(2017). Consequences of bullying victimization in childhood and adolescence: A systematic review and meta-analysis. World Journal of Psychiatry, 7(1), 60–76. https://doi.org/10.5498/wjp.v7.i1.60

26.

Olweus

(2013). School bullying: Development and some important challenges. Annual Review of Clinical Psychology, 9, 751–780. https://doi.org/10.1146/annurev-clinpsy-050212-185516

27.

Pekrun

(2006). The control-value theory of achievement emotions: Assumptions, corollaries, and implications for educational research and practice. Educational Psychology Review, 18(4), 315–341. https://doi.org/10.1007/s10648-006-9029-9

28.

Qiu

Wang

Feng

Cui

(2024). Predicting risk of bullying victimization among primary and secondary school students: Based on a machine-learning model. Behavioral Sciences, 14(1), 73. https://doi.org/10.3390/bs14010073

29.

Radford

L. S.

Corral

Bradley

Fisher

H. L

. (2013). The prevalence and impact of child maltreatment and other types of victimization in the UK. Child Abuse & Neglect, 37(10), 801–813. https://doi.org/10.1016/j.chiabu.2013.02.004

30.

Raines

T. C.

Snowden

Franks

. (2022). The role of sense of control in adolescent mental health: Buffering effects on anxiety and helplessness. Journal of Adolescence, 94, 85–94. https://doi.org/10.1016/j.adolescence.2022.10.004

31.

Ren

Liu

Zhang

Wang

(2023). The longitudinal relationship between bullying victimization and depressive symptoms for middle-school students: A cross-lagged panel network analysis. Journal of Affective Disorders, 341, 42–51. https://doi.org/10.1016/j.jad.2023.08.048

32.

Rey

Extremera

Sáez

(2019). The link between gratitude and suicidal ideation in adolescents: The mediating role of school bullying and depression. Child Indicators Research, 12(5), 1745–1758. https://doi.org/10.1007/s12187-018-9606-4

33.

Ryan

R. M.

Deci

E. L

. (2017). Self-determination theory: Basic psychological needs in motivation, development, and wellness. Guilford Press.

34.

Salmon

Chartier

Roos

L. E.

Afifi

T. O

. (2023). Typologies of child maltreatment and peer victimization and the associations with adolescent substance use: A latent class analysis. Child Abuse & Neglect, 140, Article 106177. https://doi.org/10.1016/j.chiabu.2023.106177

35.

Suliman

Mkabile

Fincham

D. S.

Ahmed

Stein

D. J

. (2009). Cumulative effect of multiple trauma on symptoms of post-traumatic stress disorder, anxiety, and depression in adolescents. Comprehensive Psychiatry, 50(2), 121–127. https://doi.org/10.1016/j.comppsych.2008.06.007

36.

Ttofi

M. M.

Farrington

D. P.

Lösel

Loeber

(2011). Do the victims of school bullies tend to become depressed later in life? A systematic review and meta-analysis of longitudinal studies. Journal of Aggression, Conflict and Peace Research, 3(2), 63–73. https://doi.org/10.1108/17596591111132873

37.

van Geel

Goemans

Zwaanswijk

Vedder

. (2022). Does peer victimization predict future suicidal ideation? A meta-analysis on longitudinal studies. Aggression and Violent Behavior, 64, Article 101577. https://doi.org/10.1016/j.avb.2021.101577

38.

van Geel

Vedder

Tanilon

. (2014). Relationship between peer victimization, cyberbullying, and suicide in children and adolescents: A meta-analysis. JAMA Pediatrics, 168(5), 435–442. https://doi.org/10.1001/jamapediatrics.2013.4143

39.

Wang

Jia

Chen

(2018). Emotional regulation self-efficacy and adolescent bullying involvement: The mediating role of coping strategies. Children and Youth Services Review, 91, 324–330. https://doi.org/10.1016/j.childyouth.2018.06.017

40.

Wang

Smith

R. A

. (2024). Trajectories of bullying and sexual harassment among U.S. adolescents: A latent-class growth analysis. Journal of Adolescent Health, 74(3), 321–330. https://doi.org/10.1016/j.jadohealth.2023.11.009

41.

Wang

Ding

Xiao

Wang

Ding

Xie

(2024). Joint developmental trajectories of victimization and nonsuicidal self-injury in adolescents and the impact of childhood emotional maltreatment: “Boys’ crisis” and “Girls’ risk”. Children and Youth Services Review, 163, Article 107015. https://doi.org/10.1016/j.childyouth.2024.107015

42.

Wen

Tang

Wang

Tong

Zhu

Wang

Ding

Wang

(2025). Predictive analysis of bullying victimization trajectory in a Chinese early adolescent cohort based on machine learning. Journal of Affective Disorders, 371, 205–214. https://doi.org/10.1016/j.jad.2024.11.057

43.

Yan

Yuan

Yang

Zhang

Peng

(2023). Detecting the risk of bullying victimization among adolescents: A large-scale machine learning approach. Computers in Human Behavior, 147, Article 107817. https://doi.org/10.1016/j.chb.2023.107817

44.

Yarkoni

Westfall

(2017). Choosing prediction over explanation in psychology: Lessons from machine learning. Perspectives on Psychological Science, 12(6), 1100–1122. https://doi.org/10.1177/1745691617693393

45.

Zhang

Wang

(2023). Meaning in life as a resilience factor between bullying victimization and life satisfaction among adolescents. Children and Youth Services Review, 150, Article 106928.

46.

Zhou

Xie

Chen

Wang

(2022). Developmental trajectories of bullying and victimization among Chinese children: A four-year longitudinal study using latent class growth analysis. Child Abuse & Neglect, 130, Article 105312.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.36 MB