The longitudinal relationships between social networks, trust and depression: a cross-lagged panel network analysis

Abstract

Purpose

This study aimed to investigate the longitudinal dynamics among social networks, trust, and depressive symptoms from a network perspective.

Methods

Data was collected from three waves (2018, 2020, and 2022) of the China Family Panel Studies (CFPS), with a sample of 15,570 participants aged 16 and older. Social networks were measured using family and interpersonal network indicators, trust was assessed through six types of individual trust, and depressive symptoms were evaluated using the 8-item Center for Epidemiologic Studies Depression Scale. Cross-sectional and cross-lagged panel network analyses were conducted to examine symptom-level interactions and temporal dynamics.

Results

Cross-sectional network analyses showed that feeling sad (D7) consistently had the highest expected influence across all three waves. Trust in local government officials (T5) also showed relatively high expected influence, whereas trust in neighbors (T2) was comparatively prominent in 2018 and remained moderately central thereafter. Family network (SN1) showed relatively low expected influence across waves. Bridge expected influence was concentrated in the social-network and trust domains, particularly interpersonal network (SN2), as well as trust in neighbors (T2), trust in local government officials (T5), and trust in parents (T1), whereas depressive symptoms showed comparatively weaker bridging roles. Cross-lagged panel network analyses indicated strong autoregressive effects for most nodes. From 2018 to 2020, strong non-autoregressive paths were observed from feeling life is not worth living (D8) to several later depressive symptoms, while feeling sad (D7) showed the highest out-expected influence. From 2020 to 2022, family network (SN1) showed the highest out-expected influence, and social-network and trust-related paths appeared more prominent than in the earlier interval.

Conclusion

Depressive symptoms, social-network indicators, and trust were dynamically interconnected in the longitudinal network models. Feeling sad (D7) remained a central depressive node, whereas social-network and trust variables—especially interpersonal network (SN2) and family network (SN1)—were more prominent in cross-domain and temporal associations. These findings highlight the relevance of social-resource variables in understanding depression-related experiences and may help inform more targeted prevention and intervention strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40359-026-04656-6.

Introduction

Depression is one of the most common mental health problems across all age groups [17], and epidemiological evidence has indicated a significant increase in the prevalence of depressive disorders in recent years [26]. Depression not only causes short-term impairments, such as sleep disturbances, reduced concentration, and social dysfunction, but also leads to long-term consequences, including cognitive decline, increased risk of chronic physical diseases, and substantial reductions in overall quality of life [20, 40]. These broad and enduring impacts of depression underscore an urgent need for research aimed at elucidating the symptom characteristics of depression and the social-relational factors that may shape its development and maintenance.

A useful framework for understanding depression in everyday life is to view it within a broader socio-relational system. In this system, social networks represent the structural and relational resources available to individuals [53], whereas trust reflects a cognitive-evaluative orientation toward others in social interactions [13, 67]. From this perspective, depressive symptoms, social-network resources, and trust are theoretically linked rather than isolated constructs [13, 53, 67]. Examining them together may therefore provide a more integrated understanding of vulnerability and resilience in depression.

Social networks refer to the relational structures formed by individuals through social interactions, typically encompassing two dimensions: family networks and interpersonal networks [56]. Family networks are close, stable connections with family members, while interpersonal networks are supportive interactions with friends, peers, colleagues, and others outside the family. Existing studies have shown that close family networks provide emotional companionship, which in turn enhance psychological resilience and a sense of security when facing adversity [6]. Interpersonal networks influence both the risk of developing depression and the way individuals experience it [33]. In the Chinese context, family ties often remain one of the most stable and salient sources of emotional and practical support, while broader interpersonal connections also shape daily coping, social participation, and help-seeking [38]. Social support theory suggests that close and supportive relationships provide emotional reassurance, tangible assistance, and a sense of belonging, thereby buffering stress and reducing vulnerability to depressive symptoms [21]. Conversely, fragile or conflict-laden social networks may reduce access to support, intensify loneliness and helplessness, and increase risk for depressive experiences [21]. Empirical studies likewise suggest that depression and social networks are linked bidirectionally: weak support may contribute to depressive symptoms [50], whereas depressive states may undermine social engagement [44], support perception [49], and willingness to seek help [54].

Trust is commonly defined as an individual’s positive expectation regarding others’ benevolent intentions or reliability in social interactions [67]. Within social capital theory, trust is not merely an individual attitude but a crucial cognitive component of social relations that supports cooperation, access to resources, and the effective functioning of social networks [25]. In this sense, social networks provide the structural basis for interaction and resource exchange, whereas trust helps determine whether individuals perceive others as reliable sources of support and are willing to rely on them in times of need. This makes trust theoretically relevant to both social-network functioning and depression. In the Chinese social context, trust in specific social targets—such as parents, neighbors, local government officials, and doctors—may be especially meaningful because these actors often play distinct roles in family support, community life, and access to everyday resources and services [19]. Accordingly, target-specific trust may provide more nuanced information than a single global trust score [35]. Higher levels of trust may facilitate the formation and maintenance of stable interpersonal ties, thereby enhancing access to support and coping resources [45]. In contrast, diminished trust may weaken perceived interpersonal safety, reduce willingness to rely on others, and increase feelings of uncertainty, social distance, loneliness, helplessness, and emotional distress. Negative trust, characterized by a tendency to view others as untrustworthy or deceitful, has also been identified as a potential risk factor for depression [1, 18]. At the same time, evidence regarding the association between trust and depression has not always been fully consistent. Some longitudinal studies suggest that low interpersonal trust may predict the emergence and persistence of depressive symptoms [32], and greater trust has been associated with a lower risk of future major depression [23]. However, other findings indicate that this association may weaken after adjustment for confounding factors [37]. These mixed findings suggest that the trust–depression relationship may be heterogeneous and target-specific, underscoring the value of a more fine-grained, node-based analytical approach.

The relationship between social networks and trust also warrants direct theoretical attention. Social networks provide the relational context in which trust is formed, maintained, and revised, whereas trust influences whether individuals are willing to engage with others, disclose vulnerability, and rely on available support. In this sense, the two are mutually reinforcing rather than independent processes. Experiences of interpersonal threat or trauma may disrupt this process: individuals who have encountered harmful social interactions are more likely to develop generalized distrust and to distance themselves from others, thereby reducing opportunities for supportive exchanges and trust repair [5]. By contrast, emotional safety, openness, and reciprocity within social relationships may foster trust, while trust itself can facilitate emotional expression, empathy, and positive interpersonal expectations, further strengthening social ties [66]. This reciprocal association can also be understood through the internal working model of social support. Repeated experiences of support, responsiveness, and emotional safety may be internalized into broader expectations that others are available, reliable, and trustworthy. Conversely, social exclusion or emotionally distant relationships may promote expectations of rejection or unreliability, thereby undermining interpersonal trust. Consistent with this view, individuals who experience social exclusion and emotionally distant peer relationships tend to report lower levels of interpersonal trust than their socially accepted counterparts [65]. Together, these findings suggest that social-network quality and trust are dynamically intertwined, and that both may be relevant to the development and maintenance of depressive symptoms.

Although prior studies have provided valuable insights, most have focused on overall construct-level associations, with relatively limited attention to symptom-level and dimension-level interactions among depressive symptoms, trust, and social-network resources [34, 36, 47]. Network theory offers a useful framework for studying mental disorders as systems of interacting elements [15]. In this view, depressive symptoms may influence one another directly, and their associations may also be shaped by biological, psychological, and social processes [7, 62]. Therefore, network analysis methods have been increasingly applied to examine fine-grained connections not only among mental-health symptoms but also between mental-health symptoms and psychosocial dimensions, such as trust and social networks [8, 16]. Within this framework, each mental-health or psychosocial domain can be represented by a set of specific symptoms or dimensions connected by edges, forming a community structure [14]. Centrality and bridge centrality measures help identify the most important nodes and the nodes that connect different communities, which in turn help explain how comorbidity develops and how symptoms interact [22]. Accordingly, interventions targeting central or bridging nodes may be especially useful for reducing current difficulties and preventing the spread of problems across domains.

Several researchers have employed network analysis to explore the relationships among depressive symptoms, trust, and social-network resources at the node level [11, 70]. Previous studies have largely relied on cross-sectional designs and have not examined how symptoms interact over time. With the development of network methods in psychology, longitudinal network analysis has become increasingly common. By merging network analysis with cross-lagged modeling, researchers developed the cross-lagged panel network (CLPN) model to investigate temporal relationships among nodes across repeated measurements [63]. Building on this foundation, the present study employed three-wave longitudinal data to construct both cross-sectional and cross-lagged networks to examine the dynamic interplay among depressive symptoms, trust, and social-network dimensions over time.

As demand for comprehensive mental health services continues to rise, the precise identification and intervention for depression has become increasingly crucial. Ignoring the complex interactions among depressive symptoms and their evolving relationships with social resources may render existing interventions less effective. To address this gap, the present study combines cross-sectional network analysis and cross-lagged panel network modeling to investigate the dynamic interplay among depressive symptoms, trust, and social networks. Particular attention is paid to identifying central nodes across domains, detecting bridge nodes that link depression with trust and social-network communities, and examining which nodes show stronger predictive influence over time. The findings are expected to provide theoretical insights and practical implications for mental health and public health, and to offer an empirical foundation for more targeted and time-sensitive intervention strategies.

Subjects and methods

Data sources

Data were drawn from the 2018, 2020, and 2022 waves of the China Family Panel Studies(CFPS), a longitudinal survey representing the national population [64]. The survey adopts a longitudinal design, in which all individuals identified at baseline, as well as their subsequently born or adopted close relatives, become genetic members of CFPS. Follow-up surveys are conducted biennially.

As of 2022, CFPS had successfully completed seven nationwide surveys. The survey process was supported by well-trained experts and scholars and employed Computer-Assisted Personal Interviewing (CAPI) technology to ensure face-to-face, in-person interaction. CFPS participants span 25 provinces, municipalities, and autonomous regions in China (excluding Hong Kong, Macao, and Taiwan), covering approximately 94.50% of the national population. This broad coverage supports the generalizability of the findings to the wider Chinese population.

Measures of trust, social networks, and depression were obtained from three waves of the CFPS (2018, 2020, and 2022). These three waves were selected because they were the most recent consecutive CFPS waves that provided comparable measures of depressive symptoms, social-network indicators, and trust indicators required for the present analyses. Relevant variables were extracted from each wave and integrated using individual identifiers. The analytic sample comprised respondents aged 16 years and above. In the primary analyses, missing values were handled using mean imputation to retain a common analytic sample across waves for both the cross-sectional and longitudinal models. The resulting primary analytic sample included 15,570 respondents, consisting of 7,684 males and 7,886 females. Because mean imputation may distort variance and covariance structure in network-based analyses, we additionally reported wave-specific missingness rates and conducted sensitivity analyses using wave-specific complete cases for the cross-sectional networks and interval-wise complete-case samples for the longitudinal CLPN models. Wave-specific missingness rates for the modeled variables are reported in Supplementary Material S5. The effective complete-case sample sizes were 24,442 (2018), 18,273 (2020), and 16,178 (2022) for the cross-sectional analyses, and 14,372 (2018–2020) and 11,297 (2020–2022) for the interval-wise longitudinal analyses.

Measures

Trust

Trust was operationalized using six target-specific trust indicators available in the CFPS. Specifically, respondents rated their trust in parents (T1), neighbors (T2), foreigners (T3), strangers (T4), local government officials (T5), and doctors (T6) on 11-point scales ranging from 0 (no trust at all) to 10 (complete trust), with higher scores indicating greater trust. In the network analyses, these indicators were modeled as six separate single-item nodes rather than combined into a single scale. Therefore, internal consistency was not reported for the trust indicators.

Social networks

This study measures individuals’ social networks from two dimensions: family network and interpersonal network [56]. Family network (SN1) was constructed by averaging relationship quality with father, mother, and children, with higher scores indicating better family relationships. Interpersonal network (SN2) was measured by respondents’ self-rated popularity and was used as an indicator of broader social connectedness. All social-network items ranged from 0 to 10, with higher values indicating more favorable social-network resources. Because SN1 was conceptualized as a composite indicator rather than as a reflective scale, internal consistency was not reported. Because SN2 was a single-item indicator, internal consistency was not applicable.

Depression

Depression was assessed using the 8-item version of the Center for Epidemiologic Studies Depression Scale (CES-D8) [48], which has shown a good psychometric reliability in previous studies [41, 71, 72, 74]. In the present sample, the internal consistency of the CES-D8 was acceptable across waves, with Cronbach’s α = 0.769 in 2018, 0.771 in 2020, and 0.776 in 2022. Participants reported their depressive experiences over the past week, based on eight different statements. Responses were recorded on a 4-point scale reflecting the frequency of these experiences: 1 = “rarely or none of the time” (less than 1 day), 2 = “some or a little of the time” (1–2 days), 3 = “occasionally or a moderate amount of time” (3–4 days), and 4 = “most or all of the time” (5–7 days). Because the present study focused on symptom-level network structure, the eight CES-D items were modeled as separate depressive-symptom nodes rather than combined into a single total score. The eight symptom nodes were: feeling down (D1), everything feels like an effort (D2), restless sleep (D3), feeling unhappy (D4), feeling lonely (D5), not enjoying life (D6), feeling sad (D7), and feeling life is not worth living (D8). Reverse-worded items were recoded so that higher scores consistently reflected greater depressive severity before network estimation. A detailed mapping table for all modeled nodes, including item wording, coding direction, and construction rules for each node or composite, is provided in Supplementary Material S6.

Data analysis

Descriptive statistics were first calculated for the same modeled variables used in the network analyses, namely the eight depressive symptom nodes (D1–D8), two social-network indicators (SN1–SN2), and six trust indicators (T1–T6), across the 2018, 2020, and 2022 waves. This analysis was used to summarize the distributions of the modeled nodes and to describe broad temporal trends across waves.

Cross-sectional networks were estimated separately for each wave using the qgraph and bootnet packages in R [14, 28, 58]. The networks included 16 nodes: eight depressive symptoms, two social-network indicators, and six trust indicators. In the primary cross-sectional analyses, missing values were handled using mean imputation to retain the common analytic sample. Partial-correlation networks were then estimated as Gaussian graphical models (GGMs) with LASSO regularization using the EBICglasso procedure to balance model sparsity and fit. Spearman correlations were used, and the EBICglasso tuning parameter was set to gamma = 0.5. Given the ordinal nature of the CES-D items, we additionally re-estimated the cross-sectional networks using an alternative correlation specification based on cor_auto as a robustness check. The resulting network structure and centrality pattern were broadly similar to those of the primary analyses. Reported network characteristics included the number of non-zero edges, average absolute edge weight, network density, and notable positive and negative edges.

Because the cross-sectional networks contained both positive and negative edges, expected influence (EI) was used as the primary centrality index, and bridge expected influence was additionally computed to identify nodes linking the depression, social-network, and trust communities [9, 42].

Network accuracy and stability were examined using the bootnet package in R. Edge-weight accuracy was evaluated by nonparametric bootstrap confidence intervals based on 1,000 resamples, and centrality stability was assessed using case-dropping bootstrap procedures. Correlation-stability (CS) coefficients were calculated for expected influence and bridge strength.

Longitudinal associations were examined using cross-lagged panel networks (CLPNs) for two adjacent intervals: 2018 → 2020 and 2020 → 2022. Following the CLPN framework proposed by Rhemtulla et al. [63], regularized nodewise regressions were estimated using the glmnet package in R. LASSO regularization (alpha = 1) with tenfold cross-validation was used to select the optimal penalty parameter, and predictors were standardized before estimation. To improve readability, autoregressive paths were suppressed in the CLPN visualizations, allowing non-autoregressive temporal associations to be more clearly displayed [24].

In the primary longitudinal analyses, mean imputation was used to retain the common analytic sample across waves. Because mean imputation may distort variance and covariance structure in network-based analyses, sensitivity analyses were additionally conducted. Specifically, the cross-sectional networks were re-estimated using wave-specific complete cases, and the CLPN models were re-estimated using interval-wise complete-case samples. These sensitivity analyses were used to examine whether the substantive pattern of the findings was robust to an alternative missing-data handling strategy.

Because causal inference cannot be established from observational panel data alone, the CLPN results are interpreted as temporal predictive associations rather than causal effects. All analyses were conducted in R, and the corresponding R code for data processing, network estimation, and bootstrap analyses is provided in the Supplementary Materials S1.

Results

Descriptive statistics of the modeled nodes

Table 1 presents the descriptive statistics of the 16 modeled nodes across the 2018, 2020, and 2022 waves in the primary analytic sample (N = 15,570). In general, the mean levels of the depressive symptom nodes, social-network indicators, and trust indicators were relatively stable across waves, although several variables showed modest variation over time. Among the depressive symptoms, D3 (restless sleep), D4 (feeling unhappy), and D6 (not enjoying life) showed comparatively higher mean scores, whereas D8 (feeling life is not worth living) showed the lowest mean score across all three waves. Most depressive symptom means were slightly higher in 2022 than in 2018. For social-network indicators, SN1 (family network) showed a slight downward trend, whereas SN2 (interpersonal network) showed a gradual increase over time. For trust indicators, T1 (trust in parents) consistently had the highest mean score, while T3 (trust in foreigners) and T4 (trust in strangers) had comparatively low mean scores across all three waves. Trust in local government officials (T5) and trust in doctors (T6) showed small increases over time. Overall, the descriptive statistics indicate broad stability in the modeled nodes, with modest temporal changes in selected depressive, social-network, and trust variables.

Table 1.

Descriptive statistics of the modeled nodes in 2018, 2020, and 2022 (N = 15,570)

Variable	2018 M	2018 SD	2020 M	2020 SD	2022 M	2022 SD
D1	1.76	0.57	1.74	0.58	1.79	0.59
D2	1.70	0.61	1.69	0.62	1.78	0.65
D3	1.76	0.68	1.82	0.68	1.89	0.71
D4	1.95	0.68	1.96	0.67	1.98	0.68
D5	1.44	0.52	1.45	0.53	1.52	0.57
D6	1.81	0.68	1.82	0.66	1.88	0.66
D7	1.48	0.51	1.48	0.51	1.53	0.52
D8	1.22	0.43	1.23	0.43	1.25	0.44
SN1	4.57	0.49	4.53	0.54	4.48	0.56
SN2	6.87	1.28	7.00	1.29	7.12	1.35
T1	9.36	0.96	9.36	0.98	9.37	0.99
T2	6.66	1.50	6.64	1.52	6.70	1.55
T3	1.69	1.69	1.64	1.67	1.65	1.67
T4	2.88	1.72	2.75	1.71	2.62	1.70
T5	5.21	1.88	5.35	1.89	5.50	1.90
T6	6.44	1.69	6.68	1.70	6.86	1.72

D1 Feeling down, D2 Everything feels like an effort, D3 Restless sleep, D4 Feeling unhappy, D5 Feeling lonely, D6 Not enjoying life, D7 Feeling sad, D8 Feeling life is not worth living, SN1 Family network, SN2 Interpersonal network, T1 Trust in parents, T2 Trust in neighbors, T3 Trust in foreigners, T4 Trust in strangers, T5 Trust in local government officials, T6 Trust in doctors

Cross-sectional network analysis

Figure 1 presents the cross-sectional networks of depressive symptoms, social-network indicators, and trust indicators in 2018, 2020, and 2022. The estimated networks showed substantial connectivity across the three waves, with 90, 94, and 84 non-zero edges out of 120 possible, respectively (average absolute edge weights ≈ 0.07; densities = 0.75, 0.78, and 0.70). The retained partial-correlation edges included both positive and negative associations. Positive edges were consistently observed among the depressive symptom nodes and between the two social-network indicators. Within the trust domain, most retained edges were positive, with one notable negative edge between trust in parents (T1) and trust in foreigners (T3). Negative cross-domain edges were also observed between feeling life is not worth living (D8) and trust in parents (T1), and between not enjoying life (D6) and family network (SN1). Full edge-weight tables are provided in the Supplementary Materials S2.

Fig. 1.

Cross-Sectional Network Analysis at 2018, 2020, and 2022. Notes: SN1 = Family network; SN2 = Interpersonal network; T1 = Trust in parents; T2 = Trust in neighbors; T3 = Trust in foreigners; T4 = Trust in strangers; T5 = Trust in local government officials; T6 = Trust in doctors; D1 = Feeling down; D2 = Everything feels like an effort; D3 = Restless sleep; D4 = Feeling unhappy; D5 = Feeling lonely; D6 = Not enjoying life; D7 = Feeling sad; D8 = Feeling life is not worth living. The same applies below. In the figures, edges are colored blue for positive and red for negative partial correlations, while line thickness represents the strength of the relationship

Based on expected influence estimates (Fig. 2), feeling sad (D7) consistently emerged as the most central node across all three waves. Trust in local government officials (T5) also showed relatively high expected influence in each wave, whereas trust in neighbors (T2) showed comparatively high expected influence in 2018 and remained moderately central thereafter. In contrast, family network (SN1) showed comparatively low expected influence across all three waves. Overall, the expected influence results suggested substantial continuity in the central structure of the cross-sectional networks from 2018 to 2022.

Fig. 2.

Expected influence of nodes in the cross-sectional networks at 2018, 2020, and 2022

Bridge expected influence results (Fig. 3) showed that the highest bridge nodes across the three years were primarily drawn from the social-network and trust domains. In particular, interpersonal network (SN2) consistently showed the highest bridge expected influence across all three waves, indicating that it played a prominent role in linking depressive symptoms with social-resource variables. Several trust indicators, especially T2, T5, and T1, also showed relatively high bridge expected influence across waves. By contrast, depressive symptom nodes generally showed lower bridge values than the leading social-network and trust nodes.

Fig. 3.

Bridge expected influence of nodes in the cross-sectional networks at 2018, 2020, and 2022

Cross-lagged panel network analysis

Figure 4 presents the cross-lagged panel networks (CLPNs) for the 2018 → 2020 and 2020 → 2022 intervals. In the visualizations, autoregressive paths were hidden to improve readability. Because causal inference cannot be established from observational panel data alone, the CLPN results are interpreted as temporal predictive associations rather than causal effects.

Fig. 4.

Cross-Lagged Networks Analysis from 2018 → 2020 and 2020 → 2022 (Autoregressive Paths Hidden). Note: Blue arrows indicate positive predictions, and red arrows indicate negative predictions

For the 2018 → 2020 interval, several notable non-autoregressive associations were observed among depressive, social-network, and trust nodes. Strong prospective links were observed from feeling life is not worth living (D8) to later depressive nodes, including feeling sad (D7), everything feels like an effort (D2), feeling down (D1), and not enjoying life (D6). Reciprocal associations between feeling unhappy (D4) and not enjoying life (D6) were also evident. In addition, family network (SN1) showed prospective links to later trust in parents (T1) and trust in foreigners (T3), indicating that family-based social resources were temporally associated with subsequent trust-related nodes.

For the 2020 → 2022 interval, the pattern showed both continuity and change. Reciprocal associations between D4 and D6 remained prominent, and additional paths linked feeling lonely (D5) with later interpersonal network (SN2) and trust in neighbors (T2). Family network (SN1) continued to show a prospective association with later trust in parents (T1). Paths within the trust domain, including T5 → T6 and T6 → T5, were also retained, alongside depressive paths such as D8 → D7 and D7 → D1. Taken together, these results suggest that depressive nodes remained closely involved in temporal predictive associations across both intervals, while the prominence of social-network and trust-related paths appeared greater in the later interval.

Full cross-lagged coefficient tables, including both autoregressive and non-autoregressive paths for the 2018 → 2020 and 2020 → 2022 CLPNs, are provided in the Supplementary Materials S3.

Centrality estimation

Figure 5 presents out-expected influence (out-EI) and in-expected influence (in-EI) for the two CLPNs. Out-EI reflects the extent to which a node prospectively predicts other nodes at the subsequent wave, whereas in-EI reflects the extent to which a node is prospectively predicted by the broader network.

Fig. 5.

Out-EI and In-EI of Symptoms in the Cross-Lagged Networks of Trust, Social Network, and Depression from 2018 → 2020 and 2020 → 2022

From 2018 to 2020, feeling sad (D7) showed the highest out-EI, followed by feeling life is not worth living (D8), not enjoying life (D6), and family network (SN1). Trust in doctors (T6) and trust in local government officials (T5) also showed relatively high out-EI. In contrast, the highest in-EI values were observed for trust in parents (T1), interpersonal network (SN2), feeling sad (D7), feeling lonely (D5), and everything feels like an effort (D2), indicating that these nodes were more strongly predicted by the broader network during this interval.

From 2020 to 2022, family network (SN1) showed the highest out-EI, followed by feeling sad (D7), not enjoying life (D6), trust in neighbors (T2), and feeling life is not worth living (D8). The highest in-EI values were observed for trust in parents (T1), feeling sad (D7), interpersonal network (SN2), everything feels like an effort (D2), and feeling down (D1). Overall, these centrality results indicate substantial continuity in the prominence of depressive nodes across the two intervals, while selected social-network and trust nodes also showed relatively strong predictive influence.

Stability analysis

Two stability analyses were performed using the bootnet package in R [14]. For the cross-sectional networks, edge-weight accuracy was evaluated using nonparametric bootstrap confidence intervals based on 1,000 resamples, and centrality stability was examined using case-dropping bootstrapping. Because the cross-sectional networks contained both positive and negative edges, expected influence was emphasized as the primary centrality index in the stability analyses. In addition, bridge strength was examined in the case-dropping bootstrap analyses to evaluate the stability of bridge centrality. The resulting CS coefficients were 0.75 for expected influence and 0.75 for bridge strength across the 2018, 2020, and 2022 cross-sectional networks, indicating that the main centrality indices were highly stable and suitable for interpretation. For the longitudinal CLPN models, the corresponding CS coefficients were also 0.75, indicating high stability of the out-expected influence and in-expected influence estimates. Detailed bootstrap results, including the case-dropping bootstrap stability plots, are provided in Supplementary Material S4. As an ordinal robustness check, the cross-sectional networks were re-estimated using an alternative correlation specification. The main findings remained broadly unchanged, with the same core depressive and bridge nodes showing the highest prominence.

In addition, sensitivity analyses were conducted to assess the robustness of the findings to missing-data handling. Specifically, the cross-sectional networks were re-estimated using wave-specific complete cases, and the longitudinal CLPN models were re-estimated using interval-wise complete-case samples. The effective complete-case sample sizes were 24,442 (2018), 18,273 (2020), and 16,178 (2022) for the cross-sectional analyses, and 14,372 (2018–2020) and 11,297 (2020–2022) for the interval-wise longitudinal analyses. The substantive pattern of the findings remained largely unchanged, with the main central nodes, bridge nodes, and major cross-lagged associations showing high similarity across the primary and sensitivity analyses. These results increase confidence in the robustness of the reported findings.

Discussion

The cross-sectional network analysis

The cross-sectional network results showed generally positive associations among the depressive symptom nodes, suggesting that depressive experiences tend to co-occur and remain closely connected within the depression domain [31]. This pattern is consistent with the view that depression can be understood as an interconnected symptom system. Symptoms appear to be closely interconnected, which may help explain why depressive experiences tend to co-occur and persist over time. This suggests that interventions for depression may benefit not only from targeting individual symptoms, but also from paying attention to highly connected symptoms within the network. A stable positive association was also observed between family network (SN1) and interpersonal network (SN2) within the social-network domain. This pattern is consistent with the resource accumulation perspective, which suggests that different forms of social support may reinforce rather than replace one another. Individuals with stronger family support are often also better positioned to maintain broader interpersonal resources, and the co-availability of multiple support resources may enhance coping capacity and psychological resilience [46, 68]. Conversely, when one dimension of support is relatively weak, overall social-resource availability may also be reduced. In other words, the synergistic effect of family and interpersonal relationships plays a critical role in maintaining mental health. The trust variables were mostly positively correlated, with the exception of a negative correlation between trust in parents (T1) and trust in foreigners (T3), consistent with previous studies [61]. This pattern may suggest that trust in close in-group ties and trust in more distant out-group targets do not necessarily develop in parallel.

In terms of expected influence, feeling sad (D7) consistently emerged as the most central node across all years, with a slight increase over time. This finding is consistent with previous research conclusions [10, 29, 69]. As a core affective feature of depression, this symptom has been widely identified as highly central in previous depression network studies [10]. The stable prominence of this symptom suggests that affective experiences may occupy an important organizing position within the depression network. Its persistent centrality may indicate close connections with a range of other depressive experiences. Accordingly, changes in affective symptoms may be associated with broader changes within the depression domain. In the trust dimension, trust in neighbors (T2) consistently showed relatively high expected influence, possibly due to its role as a representative indicator of community trust and grassroots social interaction. Compared with institutional or macro-level trust, community and neighborhood trust is closer to daily life. In the present network, neighborhood trust appeared to occupy a relatively prominent position. This may suggest that community-based trust and neighborhood support are relevant to mental-health-related experiences in everyday life. In contrast, the expected influence of family network (SN1) was relatively low. One possible interpretation is that family relationships may be more stable and therefore less sensitive to short-term fluctuations within the broader network. This finding may suggest that, in the present network, community- and neighborhood-based ties were more immediately linked to variations in emotional and trust-related experiences than family-network resources.

The bridge centrality analysis highlighted the pivotal role of cross-system nodes in connecting different psychological and social subsystems. The interpersonal network (SN2) was the most influential bridging node at all time points. This finding suggests that broad social connections (such as friends, colleagues, and other non-family members) may be important in linking depression and trust [4, 55, 73]. Interpersonal network (SN2) can be viewed as a bridge connecting endogenous psychological symptoms with external social functioning. Its high bridging centrality suggests that interpersonal connections may play an important role in linking negative emotions with trust-related perceptions [30]. Meanwhile, trust in parents (T1) and family network (SN1) also demonstrated significant bridging roles, further supporting the notion that family relationships not only provide emotional support but also play a crucial role in cross-domain information integration and the shaping of psychological safety [57]. The bidirectional integration of social support and trust resources across different systems relies on the effective transmission through these bridging nodes. It is worth noting that the overall bridging centrality of depressive symptom variables is relatively low. This is particularly true for symptoms such as restless sleep (D3) and feeling lonely (D5), indicating that these symptoms are largely confined to internal emotional cycles. As a result, their impact on social cognition and behavioral systems is limited. This pattern may suggest that some depressive symptoms are more internally embedded within the depression domain and less strongly connected to trust and social-network variables.

Cross-lagged network analysis

The cross-lagged network results revealed temporal predictive associations among depressive symptoms, trust, and social-network variables across the two intervals. From 2018 to 2020, the temporal predictive associations were more strongly centered on affective depressive nodes. Temporal predictive associations in the 2018–2020 interval were primarily concentrated among depressive-symptom nodes, although selected links to trust-related variables were also observed. This pattern suggests that depressed mood may be linked to a broad range of subsequent emotional and social-cognitive experiences [12, 52]. At the level of predictive centrality, feeling sad (D7) showed the highest out-expected influence from 2018 to 2020, suggesting that this node occupied a broadly influential position in the surrounding longitudinal network. This pattern is consistent with prior work suggesting that negative emotions may be associated with less favorable social-cognitive evaluations, including reduced trust in others [43]. This finding is also consistent with previous research, which suggests that the core of depression lies in persistent low mood—an emotion that is not only internally sustained but also spills over into the domain of social relationships [27, 59]. Meanwhile, cognitive symptoms such as feeling life is not worth living (D8) exhibited relatively strong outward influence, suggesting that hopelessness-related cognitions may be closely connected with a broader range of subsequent emotional experiences.

In contrast, the pattern of centrality shifted during the 2020–2022 period, with family network (SN1) emerging as the node with the highest out-expected influence. This change may reflect the deeper protective role of family support for individual mental health in the context of major social events, such as the COVID-19 pandemic [39]. According to the buffer theory of social support, supportive relationships may not remove external stressors directly, but they can reduce the psychological impact of those stressors by providing emotional reassurance, practical assistance, and a sense of security. From this perspective, family-based support may help weaken the association between stress exposure and subsequent negative emotional experiences. The role of family network may be particularly relevant during periods of post-pandemic social recovery [3]. Meanwhile, feeling sad (D7) remained highly prominent in the later interval, but family network (SN1) became the node with the highest out-expected influence. This transition may suggest that family-related resources became more prominent in the broader longitudinal network. Across both stages, feeling sad (D7) emerged as both a major source and a primary recipient of emotional influence, highlighting its “bidirectional hub” function in the emotional cycle of individuals.

This study identified several recurring temporal association patterns across both time periods, which reflect the structural interrelations among variables within the network. For example, during the 2018–2020 period, family network (SN1) showed prospective associations with later trust in parents (T1) and trust in foreigners (T3), indicating that family-related resources were temporally linked to subsequent trust-related variables. Additionally, several temporal associations were observed among cognitive and affective depressive nodes, suggesting that hopelessness-related and affective experiences may remain closely linked over time. Such temporal clustering among depressive symptoms is broadly consistent with prior work emphasizing the interdependence of depressive experiences over time [51]. Within this pattern, feeling lonely (D5) occupied a linking position, suggesting that loneliness may be closely related to both hopelessness-related cognitions and emotional distress [60].

From 2020 to 2022, social-network and trust-related variables were more visibly involved in the longitudinal structure, including paths linking loneliness with later interpersonal network (SN2) and trust in neighbors (T2), as well as reciprocal associations between trust in local government officials (T5) and trust in doctors (T6). This finding echoes the “affect-as-information” mechanism in social cognition models, whereby negative emotions influence individuals’ evaluations of social others or institutions [2].

The pattern of associations across the three major domains suggests relatively stronger cross-domain linkage in 2020–2022 than in 2018–2020. In the earlier period, the depression domain appeared relatively more closed, with sparse or even negative connections to the social networks and trust systems. Connections with social-network and trust domains appeared relatively sparse. By 2020–2022, the social-network and trust domains appeared more closely connected with the emotional domain through bridge nodes. For example, the later interval showed greater involvement of social-network and trust-related nodes, particularly family network (SN1), trust in parents (T1), trust in neighbors (T2), and trust in local government officials (T5). This pattern is consistent with the view that social-support resources may be relevant to psychological resilience in the face of prolonged stress.

This study has several limitations. First, data was collected within the Chinese cultural context, which enhances the interpretive power of the findings within local social environment, however, cultural differences may limit the cross-cultural generalizability of the conclusions. Second, the present study employed large-scale self-report questionnaires, which offer advantages in terms of sample coverage and representativeness. However, self-assessments may be influenced by social desirability bias, recall bias, or current emotional states, potentially resulting in measurement errors. In addition, this study did not differentiate between specific age groups. Individuals at different life stages may face distinct experiences, social networks, and stressors, which could manifest in different patterns within the psychological network. Taken together, future research should incorporate cross-cultural comparisons, multi-method assessments, and age-specific analyses to provide a more precise understanding of the interplay among depression, social networks, and trust.

Conclusion

This study showed that depressive symptoms, social-network indicators, and trust were dynamically interconnected across the three-wave longitudinal network models. Feeling sad (D7) consistently occupied a central position in the cross-sectional depression networks, whereas social-network and trust variables—particularly interpersonal network (SN2), trust-related nodes, and later family network (SN1)—were more prominent in bridging and temporal predictive associations. These findings underscore the relevance of social-resource variables for understanding depression-related experiences and may help inform more targeted intervention priorities in community and public mental health contexts.

Authors’ contributions

XL led the study conceptualization, designed the research framework, conducted the primary data analyses, and drafted the manuscript. MZ supervised the project, refined the research design, ensured methodological rigor, and made substantial contributions to the theoretical integration and critical revision of the manuscript. MX assisted with methodology and advanced analyses; YS handled data acquisition and preprocessing; HD contributed to the theoretical background and literature review; XZ assisted with data visualization and results revision; XW performed robustness checks and contributed to the discussion. LJ made substantial contributions to the theoretical integration and revision of the manuscript. All authors approved the final manuscript.

Funding

No funding.

Data availability

This study uses data from the China Family Panel Studies (CFPS), publicly available at: https://opendata.pku.edu.cn

Declarations

Ethics approval and consent to participate

Ethical approval was not required for this study because all analyses were conducted using fully anonymized, publicly available secondary data from the China Family Panel Studies. According to the database’s governance policy, the dataset was collected with prior ethical approval and informed consent by the original investigators, and no identifiable personal information was accessible to the authors.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.