Gender-specific temporal dynamics of depressive symptoms and emotional resilience in adolescents: a cross-lagged network and simulation-based intervention study

Abstract

Background

Adolescent depression involves dynamic interactions between depressive symptoms and emotional resilience, with potential gender differences.

Methods

A total of 818 adolescents completed the CDI-S and AERQ at two time points. Cross-lagged panel networks were estimated separately for males and females, and node centrality was assessed using in- and out-expected influence indices. Simulation-based interventions identified nodes with the greatest impact on overall symptom activation.

Results

In males, sadness showed the largest autoregressive effect, self-deprecation showed the highest out-expected influence, and feeling upset when misunderstood showed the highest in-expected influence. In females, feeling unloved showed the largest autoregressive effect, sadness showed the highest out-expected influence, and feeling upset after criticism the highest in-expected influence. Simulation results indicated that enhancing positive emotion generation alleviated symptoms, whereas sleep- and interpersonal-related dysregulation increased network activation.

Conclusions

Emotional resilience–depression dynamics differed by gender, with males showing cognitive–evaluative propagation and females emotion-driven interpersonal sensitivity. Targeted, gender-sensitive interventions focused on positive emotion generation and interpersonal recovery may help reduce depressive symptoms in adolescents.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13034-026-01138-w.

Introduction

Depressive disorder is one of the most common mental health problems among adolescents and is associated with substantial functional impairment and elevated suicide risk [46]. Adolescence is also a critical period during which gender differences in depression become increasingly evident, with females generally reporting higher levels of depressive symptoms and a higher prevalence of depressive disorder than males [35, 43, 46]. These differences may reflect the combined influence of biological vulnerability, pubertal development, and gendered exposure or sensitivity to interpersonal stressors [14, 22, 50]. Because emotional dysfunction is a core feature of depression, particularly persistent negative mood and anhedonia [3, 23, 27, 55], examining how emotion-related processes interact with depressive symptoms during adolescence is important for clarifying the developmental mechanisms of depression.

Emotion regulation refers to the processes through which individuals monitor, evaluate, and modify emotional responses in accordance with personal goals and environmental demands [6, 9, 10]. Difficulties in regulating emotions after negative life events may increase vulnerability to depressive symptoms, whereas adaptive regulation may reduce the risk of depression [26]. Gender differences in emotion regulation may further contribute to the unequal burden of depression among adolescents. For example, females are more likely to use internalizing strategies such as rumination, which may intensify and prolong negative affect [8, 37]. Prior studies and meta-analytic evidence have shown that females tend to report higher levels of rumination than males, especially in response to low mood [2, 51]. Clarifying how emotion regulation and depressive symptoms influence each other over time, and whether these pathways differ by gender, is therefore essential for understanding adolescent emotional development and depression risk.

Most previous studies have examined associations between emotion regulation and depression using traditional longitudinal regression models. Although such models can estimate temporal associations, they often treat depression and emotion regulation as aggregate constructs, which may obscure the heterogeneous interactions among specific symptoms and emotional components [5, 11, 24, 39]. In contrast, cross-lagged panel network analysis conceptualizes psychological constructs as systems of interacting nodes and estimates directed pathways among specific symptoms or components across time. This approach can identify which emotion-related nodes predict later depressive symptoms, which depressive symptoms predict later emotion regulation difficulties, and which nodes may function as key targets within the longitudinal system [21, 24, 25, 29]. Recent applications of cross-lagged network analysis in adolescent mental health have demonstrated its value for revealing symptom-level temporal pathways that are not captured by total-score approaches [29].

Building on this framework, network-based simulation interventions provide a further method for evaluating how changes in key nodes may influence the development of the broader psychological network over time. By simulating targeted changes in selected nodes at an earlier time point and examining their downstream effects at a later time point, this approach can help identify potential intervention targets and protective pathways within complex psychological systems [7, 18, 32]. Therefore, the present study aimed to elucidate the dynamic interactions between depressive symptoms and emotion regulation among adolescents by integrating cross-lagged panel network analysis with network-based simulation interventions. Specifically, we constructed and compared gender-specific longitudinal emotion–depression networks to identify symptom-level pathways and potential intervention targets among male and female adolescents. This study may contribute to a more precise understanding of gender differences in adolescent depression and provide empirical evidence for the development of tailored emotion-focused intervention strategies.

Methods

Participants and procedures

This study used convenience sampling and included two middle schools in a northern city in China. Data were collected at two time points: October 2023 and November 2024. The study was approved by the Ethics Committee of Xuzhou Medical University (Approval number: XZHMU-2023029). Prior to participation, informed consent was obtained from both the participants and their legal guardians. Paper-based questionnaires were then administered on site. Participants completed the questionnaires during regular school hours and returned them immediately after completion. All original data were securely maintained by the research team and monitored by the ethics committee. As a form of feedback, the study results will be used to inform the development of subsequent intervention programs aimed at promoting adolescent mental health. The analysis report of the overall data is shared only with the psychology teachers at the target middle schools to guide school-based psychological education.

Measures

Demographic information

A self-developed questionnaire was used to collect demographic information in line with the study objectives, including age and gender.

Depressive symptoms

The Children’s Depression Inventory–Short Form (CDI-S) was used to assess depressive symptoms among children and adolescents. It is a shortened version of the original Children’s Depression Inventory (CDI) [52]. The CDI-S consists of 10 self-report items covering core depression-related dimensions, such as depressed mood, loss of interest, sleep disturbance, and appetite change, making it suitable for rapid screening contexts. Each item is rated on a 3-point scale (0, 1, or 2), where 0 indicates that the symptom is “absent” or “rarely occurs,” 1 indicates it “occasionally occurs,” and 2 indicates it “frequently occurs.” Higher scores reflect greater frequency and severity of depressive symptoms. Based on commonly used clinical thresholds, a total score above 12 suggests a potential risk of depression, warranting further professional evaluation. In the present study, the CDI-S demonstrated good reliability and validity, with a Cronbach’s α coefficient of 0.76.

Adolescent emotional resilience

The Adolescent Emotional Resilience Questionnaire (AERQ) is a standardized instrument designed to assess adolescents’ emotional recovery capacity, providing an index of their emotional well-being. The questionnaire comprises two dimensions: (1) the ability to generate positive emotions and (2) the ability to recover from negative emotions. The first dimension evaluates the extent to which adolescents actively elicit and sustain positive emotions such as joy and optimism in daily situations. The second dimension assesses their capacity to cognitively adjust, regulate distress, and restore emotional stability following experiences of frustration, sadness, or anxiety. The AERQ includes 11 items corresponding to these dimensions, ensuring targeted assessment. Responses are rated on a 6-point Likert scale (1 = strongly disagree to 6 = strongly agree), with higher total scores indicating stronger emotional recovery ability. In this study, the AERQ demonstrated good internal consistency, with a Cronbach’s α coefficient of 0.81.

Data analysis

Statistical analyses were conducted using R version 4.4.0. Categorical variables were described using frequencies and percentages, while continuous variables were summarized using means and standard deviations. Independent samples t-tests were performed to compare scores between male and female adolescents in order to identify significant differences across variables.

Network estimation

Cross-lagged panel networks were estimated separately for male and female adolescents using a series of regularized regression models to capture both autoregressive and cross-lagged effects across the two time points [17]. Autoregressive effects refer to the extent to which a symptom at Time 1 predicts itself at Time 2, after accounting for all other variables and covariates. Cross-lagged effects represent the influence of one symptom at Time 1 on a different symptom at Time 2, controlling for the remaining symptoms and covariates. Network estimation was conducted using penalized maximum likelihood with a lasso penalty on regression coefficients [17], which reduces overfitting and eliminates negligible pathways by shrinking small coefficients to zero, thereby enhancing model generalizability [53]. The glmnet package [17] was used for regularized regression analysis, and the resulting networks were visualized using the qgraph package [12, 13]. Given the directed nature of CLPNs derived from longitudinal data, node-level centrality was evaluated using two indices: out-expected influence (out-EI) and in-expected influence (in-EI). Out-EI quantifies the extent to which a node at Time 1 predicts other nodes at Time 2 (i.e., the sum of outgoing edge weights), while in-EI reflects the degree to which a node at Time 2 is predicted by other nodes at Time 1 (i.e., the sum of incoming edge weights). A higher in-EI value indicates greater predictability by other variables, whereas a higher out-EI indicates stronger predictive power over other variables. Together, these indices provide insights into each node’s dynamic role in the temporal network structure [42].

To assess the accuracy and stability of the network, nonparametric bootstrapping procedures were implemented using the bootnet package [12, 13]. Case-drop bootstrapping was applied to evaluate the robustness of centrality indices through the calculation of correlation stability (CS) coefficients. A CS coefficient above 0.25 was deemed acceptable, while a value exceeding 0.50 indicated strong stability [12, 13].

Simulation-based intervention

A simulation-based intervention was conducted using the Ising model combined with the NodeIdentifyR algorithm. The raw data were first preprocessed through median dichotomization to achieve discrete transformation. Specifically, for each variable included in the analysis, observations with values less than or equal to the median were coded as 0, while those above the median were coded as 1. This procedure produced a binarized data matrix suitable for network modeling.

Subsequently, a binary symptom network was constructed by fitting an Ising model using the bootnet package in R. The IsingFit algorithm was employed to estimate both the pairwise association strengths and threshold parameters between nodes. The regularization parameter was automatically selected using the “optimal” criterion to balance model fit and sparsity. The resulting edge-weight matrix and threshold parameters derived from the Ising model provided the foundation for subsequent computer-based simulation interventions implemented via the NodeIdentifyR package.

The NodeIdentifyR algorithm (NIRA) performs simulation interventions by modifying the threshold parameters of nodes, which correspond to the intercept terms in logistic regression. Specifically, for each node in the T1 network, the threshold parameter was systematically decreased or increased to simulate the alleviation or exacerbation of the symptom represented by that node. During the simulation, changes in the overall symptom level of the T2 network were monitored to identify which nodes exerted the greatest influence on the network’s overall state, thereby determining the most impactful key nodes. The strength of the simulated intervention was determined by adjusting the estimated threshold by a fixed multiple of its standard deviation; in this study, the threshold was modified by two standard deviations to define the intervention intensity.

Because the aim of the simulation was to examine whether modifying earlier emotional resilience processes could influence later depressive symptom severity, the simulation intervention was restricted to T1 AERQ nodes rather than the strongest depressive symptom nodes identified in the CLPN. Thus, the selected intervention targets were not intended to reproduce the strongest centrality nodes in the temporal network, but to identify which emotional resilience components, when hypothetically improved or worsened at T1, produced the largest downstream changes in T2 depressive symptoms.

Results

A total of 818 valid participants were included in this study. The mean baseline age of the overall sample was 14.56 years (SD = 1.67). Among the participants, 394 were male, accounting for 48.76% of the overall sample. Overall, differences in depressive symptoms and emotional resilience were observed between male and female adolescents, as shown in Table 1.

Table 1.

Age-band distribution and descriptive statistics of CDI-S and AERQ scores

Age band	Total n	Male n	Female n	T1 CDI M (SD)	T2 CDI M (SD)	T1 AERQ M (SD)	T2 AERQ M (SD)
Total sample	818	394	414	5.38 (2.88)	5.40 (3.21)	38.13 (11.14)	41.75 (9.73)
Male participants	394	394	0	5.23 (2.88)	5.38 (3.28)	40.23 (10.74)	42.88 (10.07)
Female participants	414	0	414	5.52 (2.88)	5.43 (3.15)	36.14 (11.15)	40.67 (9.28)
12 years	108	47	61	4.56 (2.23)	4.83 (2.73)	42.91 (10.96)	41.57 (10.76)
13 years	193	92	101	4.84 (2.70)	4.91 (3.16)	40.75 (11.36)	41.49 (11.85)
14 years	109	55	54	5.03 (2.65)	5.18 (3.37)	40.13 (10.48)	41.45 (11.43)
15 years	102	55	47	5.97 (2.98)	6.52 (3.69)	34.17 (10.63)	41.64 (8.79)
16 years	175	86	89	6.02 (3.09)	5.76 (3.19)	35.05 (9.46)	42.35 (6.67)
17 years	118	57	61	5.72 (3.01)	5.41 (2.98)	35.70 (11.14)	41.56 (7.45)
18 years	3	2	1	7.67 (7.23)	2.67 (0.58)	26.00 (19.80)	50.33 (5.13)

Values are presented as n or mean (standard deviation). CDI-S = Children’s Depression Inventory–Short Form; AERQ = Adolescent Emotional Resilience Questionnaire; T1 = Time 1; T2 = Time 2. Scale means and standard deviations were calculated using available complete scale scores. The raw dataset contained 820 records; two records were excluded from the analytic sample because sex was missing or invalid, yielding 818 valid participants. Age-band rows were calculated among participants with non-missing baseline age (n = 808); ten valid participants had missing baseline age and are therefore included in the overall rows but not in the age-band rows. The 18-year-old age band included only three participants; descriptive statistics for this subgroup should therefore be interpreted cautiously

Network in male

The male adolescents’ network and standardized centrality estimates are presented in Fig. 1. The largest autoregressive effect was observed for “sadness” (CDI 1; β = 0.22). “self-deprecation” (CDI 3; β = 1.37) demonstrated the highest out-expected influence, whereas “feeling upset when misunderstood by others” (AERQ 10; β = 0.55) exhibited the highest in-expected influence. The most robust positive edge was between “Self-deprecation – I cannot regulate my negative emotions quickly” (CDI 3 and AERQ 6, β = 0.33), while the strongest negative edge was between “pessimism – when feeling down, I cannot think of anything pleasant” (CDI 2 and AERQ 1, β = −0.48).

Fig. 1.

Cross-lagged network and centrality of depressive symptoms and emotional resilience among male adolescents. Arrows indicate temporal paths from T1 to T2. Solid and dashed edges indicate positive and negative temporal associations, respectively. CDI1: Sadness; CDI2: Pessimism; CDI3: Self-deprecation; CDI4: Self-hate; CDI5: Crying spells; CDI6: Irritability; CDI7: Negative body image; CDI8: Loneliness; CDI9: Lack of friends; CDI10: Feeling unloved. AERQ1: When in a bad mood, I can think of things that make me happy; AERQ2: Criticism from others makes me feel upset for a long time; AERQ3: I can maintain a good mood no matter what difficulties I encounter; AERQ4: My mood is not easily affected by external events; AERQ5: Unpleasant events will bother me for a long time; AERQ6: I can quickly adjust negative emotions; AERQ7: Unpleasant things during the day make it hard for me to sleep; AERQ8: I can find ways to enjoy myself when facing stress or setbacks; AERQ9: No matter how bad I feel, I can think positively; AERQ10: Misunderstandings from others make me sad; AERQ11: It is difficult for me to calm down after arguing with others.

Network in female

The female adolescents’ network and standardized centrality estimates are presented in Fig. 2. The largest autoregressive effect was observed for “feeling unloved” (CDI 10; β = 0.35). “Sadness” (CDI 1; β = 0.70) showed the highest out-expected influence, whereas “feeling upset for a long time after being criticized by others” (AERQ 2; β = 0.32) remained the most influential in terms of in-expected influence. The strongest positive edge was between “sadness – feeling upset for a long time after being criticized by others” (CDI 1 and AERQ 2, β = 0.31), while the strongest negative edge was between “self-disgust – When feeling down, I cannot think of anything pleasant” (CDI 4 and AERQ 1, β = −0.34).

Fig. 2.

Cross-lagged network and centrality of depressive symptoms and emotional resilience among female adolescents. Arrows indicate temporal paths from T1 to T2. Solid and dashed edges indicate positive and negative temporal associations, respectively. CDI1: Sadness; CDI2: Pessimism; CDI3: Self-deprecation; CDI4: Self-hate; CDI5: Crying spells; CDI6: Irritability; CDI7: Negative body image; CDI8: Loneliness; CDI9: Lack of friends; CD110: Feeling unloved. AERQ1: When in a bad mood, I can think of things that make me happy; AERQ2: Criticism from others makes me feel upset for a long time; AERQ3: I can maintain a good mood no matter what difficulties I encounter; AERQ4: My mood is not easily affected by external events; AERQ5: Unpleasant events will bother me for a long time; AERQ6: I can quickly adjust negative emotions; AERQ7: Unpleasant things during the day make it hard for me to sleep; AERQ8: I can find ways to enjoy myself when facing stress or setbacks; AERQ9: No matter how bad I feel, I can think positively; AERQ10: Misunderstandings from others make me sad; AERQ11: It is difficult for me to calm down after arguing with others

Stability of the networks

The stability of the networks was assessed using CS-coefficients. In the male network, the CS-coefficients for in-EI and out-EI were 0.20 and 0.36, respectively. In the female network, the corresponding coefficients were 0.20 and 0.28. These results indicate limited stability of the in-EI estimates in both networks, whereas the out-EI estimates showed only acceptable but not strong stability. Therefore, the centrality findings should be interpreted cautiously. To further evaluate the robustness of the temporal edge estimates, bootstrap edge inclusion rates and post-selection Bayesian credible intervals were calculated and are reported in Supplementary Table S1.

Simulation intervention on male emotional resilience

The intervention targets were selected from T1 emotional resilience nodes because the simulation was designed to evaluate the downstream effects of modifying emotional resilience on later depressive symptoms.

In the male simulation intervention, the item “I cannot find enjoyment when facing stress or setbacks” (AERQ 8) showed the largest alleviation effect, indicating improvement in this aspect following the simulated intervention. Conversely, the item “unpleasant daytime events make it difficult for me to fall asleep at night” (AERQ 7) exhibited an aggravation effect, suggesting an increase in symptom activation related to sleep disturbance. The results of the male simulation intervention are shown in Fig. 3.

Fig. 3.

Simulation-based intervention effects of T1 emotional resilience nodes on T2 depressive symptom severity among males. The x-axis represents the T1 emotional resilience node altered in the simulation, and the y-axis represents the T2 depressive symptom sum score. AERQ1: When in a bad mood, I can think of things that make me happy; AERQ2: Criticism from others makes me feel upset for a long time; AERQ3: I can maintain a good mood no matter what difficulties I encounter; AERQ4: My mood is not easily affected by external events; AERQ5: Unpleasant events will bother me for a long time; AERQ6: I can quickly adjust negative emotions; AERQ7: Unpleasant things during the day make it hard for me to sleep; AERQ8: I can find ways to enjoy myself when facing stress or setbacks; AERQ9: No matter how bad I feel, I can think positively; AERQ10: Misunderstandings from others make me sad; AERQ11: It is difficult for me to calm down after arguing with others

Simulation intervention on female emotional resilience

In the female simulation intervention, the item “I cannot find enjoyment when facing stress or setbacks” (AERQ 8) demonstrated a significant alleviation effect, indicating improvement in this emotional regulation aspect after the simulated intervention. In contrast, the item “It is difficult for me to calm down after quarreling with others” (AERQ 11) showed an aggravation effect, suggesting heightened activation of interpersonal emotional disturbance. The results of the female simulation intervention are shown in Fig. 4.

Fig. 4.

Simulation-based intervention effects of T1 emotional resilience nodes on T2 depressive symptom severity among females. The x-axis represents the T1 emotional resilience node altered in the simulation, and the y-axis represents the T2 depressive symptom sum score. AERQ1: When in a bad mood, I can think of things that make me happy; AERQ2: Criticism from others makes me feel upset for a long time; AERQ3: I can maintain a good mood no matter what difficulties I encounter; AERQ4: My mood is not easily affected by external events; AERQ5: Unpleasant events will bother me for a long time; AERQ6: I can quickly adjust negative emotions; AERQ7: Unpleasant things during the day make it hard for me to sleep; AERQ8: I can find ways to enjoy myself when facing stress or setbacks; AERQ9: No matter how bad I feel, I can think positively; AERQ10: Misunderstandings from others make me sad; AERQ11: It is difficult for me to calm down after arguing with others

Discussion

This study constructed gender-specific cross-lagged panel networks to examine temporal associations between depressive symptoms and emotional resilience among adolescents. The findings showed that, in both male and female networks, emotion-related responses to negative interpersonal evaluation were the most susceptible nodes, whereas core depressive symptoms showed stronger outward temporal influence. However, the specific pathways differed by gender: self-deprecation showed the highest out-expected influence in males, whereas sadness showed the highest out-expected influence in females. These results suggest that adolescent depression may be maintained through partly distinct gender-specific pathways, with cognitive-evaluative processes being more prominent in males and emotion-driven interpersonal sensitivity being more prominent in females.

The strongest autoregressive symptom in the male depression network was “sadness,” suggesting that once negative emotions are activated, they may mutually reinforce each other within a self-amplifying cycle that intensifies over time, becoming internalized and persistent [45]. In contrast, in the female network, “feeling unloved” emerged as the strongest autoregressive symptom, highlighting the tenacious feedback of perceived interpersonal rejection, consistent with prior findings [30]. Both symptoms are embedded within the core depressive spectrum defined by the DSM-5, rather than the broader emotional domain (such as cross-domain lags involving anxiety or anger) [34].

In both the male and female networks, the nodes with the highest in-EI were emotion-related items, specifically “I feel upset when others misunderstand me” for males and “I feel sad for a long time when criticized by others” for females. This indicates that, regardless of gender, individuals’ emotional responses to negative interpersonal evaluation represent the core nodes most susceptible to influence from other symptoms within the depression network. These findings suggest that the input to adolescent depression is primarily concentrated on social-emotional distress, whereby interpersonal stress further reinforces the depressive system through the activation of emotional responses [40, 41]. In contrast, the nodes with the highest out-EI in both networks corresponded to core depressive symptoms, but with different manifestations: “self-deprecation” for males and “sadness” for females. This pattern indicates that outward propagation within the male depressive network primarily occurs through cognitive pathways, whereby the activation of negative self-evaluation readily triggers widespread symptom connectivity. In females, however, depressive propagation is more reflective of a continuation of the emotional response chain. Overall, emotional responses serve as the common input nodes in depression networks for both genders, while self-deprecation and sadness constitute the output nodes for males and females, respectively, highlighting gender-specific mechanisms in the maintenance of depression.

These results should also be understood in light of the growing salience of peer relationships during adolescence. As adolescents spend more time with peers and become more sensitive to classmates’ and friends’ evaluations, experiences of being misunderstood, criticized, or rejected may carry particular emotional weight [1]. Rather than reflecting isolated negative reactions, these experiences may threaten adolescents’ sense of belonging and social identity, thereby increasing negative affect, reinforcing rumination, and making emotional recovery more difficult [20]. The interpersonal sensitivity nodes observed in the present study may therefore represent one pathway through which peer acceptance and rejection become linked to depressive symptoms [48]. This interpretation is especially relevant for the female network, in which criticism-related emotional distress emerged as a key input node, suggesting that interventions for adolescent depression should also address peer-related stress, perceived rejection, and coping with negative social evaluation [56].

In the male network, the strongest positive association was observed between the items “self-deprecation” and “I cannot regulate my emotions in a short time.” This pattern reveals that individuals with higher levels of self-deprecation are more prone to entering a self-reinforcing cycle of negative emotions when confronted with adverse events. Through persistent self-blame and self-doubt, male adolescents find it difficult to disengage from negative affect, which in turn amplifies the focusing effects of maladaptive strategies such as rumination. Consequently, the pathway for emotional recovery becomes obstructed [37]. This pattern is consistent with evidence linking self-deprecation and self-critical processes to repetitive negative thinking, suggesting a tendency to maintain attention on negative emotional content [19, 36]. The strongest negative association was found between “pessimism” and “When I feel bad, I can think of something that makes me happy.” This indicates that the more pessimistic an individual is, the less likely they are to generate positive thoughts when experiencing low mood. It underscores the pivotal role of pessimism in emotion regulation: individuals with high pessimism may preferentially attend to anticipated failure and threat cues during negative emotional activation, thereby impeding access to positive emotional resources and weakening their ability to recover from distress. This mechanism may reflect the reinforcement of negative future expectations described in Seligman’s learned helplessness model [33].

In the female network, the strongest positive association was observed between “sadness” and “I feel sad for a long time when criticized by others.” This finding suggests that individuals with high sensitivity to sadness are more likely to internalize external criticism as enduring emotional distress during social interactions. Such experiences may trigger sadness that is amplified through social withdrawal and rumination, thereby prolonging the emotional recovery process. This dynamic may reflect the heightened vigilance to interpersonal threats described in the social exclusion model, in which girls—shaped by evolutionary and socialization processes—display greater sensitivity to relational cues [4]. Moreover, women are more likely to engage in rumination following negative events, perceiving criticism as a threat to self-worth [38, 41, 49]. The strongest negative association emerged between self-disgust and “When I feel bad, I can think of something that makes me happy.” This indicates that self-disgust may serve as a core symptom impeding positive emotion recovery. Individuals with higher levels of self-disgust tend to fall into cognitive biases that prioritize negative self-images while suppressing positive emotional activation, thereby obstructing affective restoration. This mechanism resembles an extended form of Seligman’s learned helplessness model [33], and is likely exacerbated in females due to the internalization of gender norms—such as perfectionistic expectations—that intensify self-critical tendencies [44].

During the simulation intervention for the male depression network, alleviation of the inability to find enjoyment in the face of stress or setbacks indicated that the intervention effectively enhanced positive emotion–regulation strategies, thereby helping to ameliorate a core depressive symptom—anhedonia. This outcome is consistent with prior research showing that greater use of cognitive reappraisal is associated with reduced depression severity [54]. Conversely, the exacerbation of daytime unpleasant events interfering with nocturnal sleep suggests that amplified negative rumination may worsen sleep disturbances, creating a bidirectional depression–insomnia loop. Rumination increases pre-sleep hyperarousal, which in turn impairs sleep onset, continuity, and quality; this recurrent cycle of perseverative negative thinking and sleep problems may ultimately contribute to the development or aggravation of depressive symptoms [31].

During the simulation intervention for the female depression network, the alleviation of difficulty in finding enjoyment when facing stress or setbacks indicated that the intervention effectively enhanced positive emotion-regulation strategies—such as cognitive restructuring and positive attentional refocusing—which helped mitigate core depressive symptoms like anhedonia and motivational deficits. This finding aligns with gender difference research suggesting that females tend to exhibit greater proficiency in generating cognitive reappraisal, and that higher use of reappraisal is significantly associated with reduced depressive experiences [28, 54]. However, the exacerbation of “difficulty calming down after an argument” suggests that post-conflict rumination may amplify emotional reactivity and hinder emotional recovery, thereby contributing to a depression–interpersonal cycle. This may occur because girls often place greater emphasis on maintaining close, high-quality relationships, making interpersonal closeness particularly salient to the onset and maintenance of depressive symptoms among females [41].

From a translational perspective, the present findings may inform more targeted assessment and prevention strategies for adolescent depression. The gender-specific temporal patterns suggest that different symptom domains may warrant closer attention across genders. For males, cognitive-evaluative symptoms, particularly persistent sadness, showed relatively strong temporal stability and influence within the network, indicating their potential relevance for early identification and preventive monitoring. For females, interpersonal-affective symptoms, especially feelings of being unloved, may represent a more salient pathway through which depressive symptoms are maintained or amplified over time. Moreover, the simulation results provide preliminary evidence that strengthening positive emotional generation may have downstream effects on depressive symptom dynamics. Nevertheless, these implications should be interpreted cautiously, as network-derived findings do not establish causal effects or directly translate into clinical treatment recommendations [15, 16]. Instead, they offer hypothesis-generating evidence for future personalized and gender-sensitive prevention or intervention studies.

Limitations

This study used longitudinal cross-lagged network and simulation intervention analyses to examine the dynamic influence of emotional resilience on depressive symptoms among male and female adolescents, thereby providing preliminary evidence for the proposed theoretical framework. Several limitations should be acknowledged. First, although the overall sample size was relatively large, the gender-stratified analyses included approximately 394 participants per group. Given the item-level network structure and the large number of potential temporal paths, this subgroup sample size may have limited the precision and stability of some edge estimates. To address this concern, we used LASSO regularization and supplemented the main analyses with bootstrap confidence intervals, edge inclusion rates, and post-selection Bayesian credible intervals. Nevertheless, some edges still showed considerable uncertainty; therefore, the gender-specific findings should be interpreted as exploratory and require replication in larger and more diverse samples. Second, the use of two assessment waves separated by approximately one year limits the extent to which short-term temporal dynamics between emotional resilience and depressive symptoms can be captured. Although this design allows us to examine prospective associations and autoregressive effects over a longer developmental interval, it cannot fully represent the more rapid fluctuations in adolescent depressive symptoms and emotion-related processes that may occur over days, weeks, or even within days, as suggested by experience-sampling research. Therefore, the temporal paths observed in this study should be interpreted as long-term prospective associations rather than fine-grained dynamic processes. Future studies could adopt multi-wave, shorter-interval, or intensive longitudinal designs, including experience-sampling approaches, to further examine the micro-level temporal dynamics of emotional resilience and depressive symptoms [47]. Finally, the generalizability of the findings may be limited by the sample characteristics. Future research should include adolescents from more diverse cultural, regional, and socioeconomic backgrounds to determine whether the observed network patterns are consistent across different populations. Such replication would help clarify the robustness of the identified temporal associations and improve the reliability of network-based intervention implications.

Conclusion

In summary, this study reveals gender-specific temporal dynamics between depressive symptoms and emotional resilience in adolescents. Emotional responses to interpersonal stress served as key entry points, while self-depreciation in males and sadness in females acted as primary output nodes, reflecting distinct maintenance mechanisms. Simulation analyses showed that enhancing positive emotion generation reduced overall symptom activation, whereas sleep-related dysregulation in males and interpersonal recovery difficulties in females increased network vulnerability. These findings provide preliminary, hypothesis-generating evidence for future gender-sensitive prevention and intervention research targeting cognitive-evaluative processes in males and emotion-driven interpersonal processes in females.

Author contributions

Q.L as the first author, X.Y. Liu as the second author, F.Z as the first Corresponding Author, and W.L.W as the second Corresponding Author. Q.L contributed to the conceptualization and design of the study and drafted the manuscript. X.Y.L. was responsible for data collection, data processing, and statistical analysis. W.L.W. and F.Z reviewed and revised the manuscript for critical intellectual content, served as corresponding authors, and supervised the entire project.

Funding

the General Project of Philosophy and Social Science Research in Colleges and Universities in Jiangsu Province, Grant No. (2023SJYB1109); National Natural Science Foundation of China, grant number (72304235); the Scientific Research Foundation of Excellent Talents in Xuzhou Medical University, grant number (D2022032).

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

This study was approved by the Ethics Committee of Xuzhou Medical University (Approval number: XZHMU-2023029).

Informed consent

Informed consents were obtained from all adolescents and legal guardians before the assessment.

Competing interests

The authors declare no competing interests.