Abstract
College students’ mental health problems have become an important challenge facing higher education. As a low-threshold and highly accessible form of social support, the mechanisms of peer support in mental health maintenance remain unclear. This study adopted a three-year longitudinal tracking design, collecting data from 1,842 college students at six time points, and used structural equation modeling and latent growth curve modeling to test the long-term enhancement effects of peer support on mental health through the chain mediation mechanism of self-efficacy and social adaptation. Results showed that peer support had a significant positive predictive effect on mental health (total effect β = 0.33, SE = 0.04, 95% CI [0.25, 0.41], p < .001), with indirect effects totaling 0.21 (SE = 0.03, 95% CI [0.15, 0.27], p < .001), accounting for 63.6% of the total effect. Specifically, the simple mediation effect through self-efficacy was 0.10 (30.3% of total effect), through social adaptation was 0.06 (18.2% of total effect), and the chain mediation effect through “self-efficacy → social adaptation” was 0.05 (15.1% of total effect), with Bootstrap confidence intervals for all indirect effects excluding zero. Longitudinal analysis found that all four core variables showed significant linear growth, with social adaptation having the largest growth slope (0.12). Cross-lagged tests confirmed the causal priority and cumulative enhancement characteristics of peer support. The Random Intercept Cross-Lagged Panel Model (RI-CLPM) analysis further confirmed that within-person effects accounted for 58.2% of the total within-person association, supporting genuine individual change processes. The study also found a significant compensatory growth pattern (r = − .18), with students having lower initial mental health levels showing faster growth rates. Gender and regional moderation effects indicated that female students and students from western regions benefited more from peer support. The progressive “support-efficacy-adaptation-health” mechanism revealed in this study deepens understanding of social support theory and provides empirical evidence for universities to build stratified and classified peer psychological support systems. It suggests that college mental health education should focus on cultivating a supportive campus culture and promoting the coordinated development of students’ self-efficacy and social adaptation abilities through structured peer counseling programs.
Supplementary Information
The online version contains supplementary material available at 10.1186/s40359-025-03939-8.
Keywords: Peer support, Mental health, Self-efficacy, Social adaptation, Chain mediation, Longitudinal tracking
Introduction
College students’ mental health problems have become an important challenge facing global higher education. Particularly against the backdrop of accelerating social transformation and increasing academic pressure, anxiety, depression, and other psychological problems exhibited by college students show an upward trend, affecting not only their academic performance and personal development but also the quality of talent cultivation in higher education and sustainable social development. As a non-professional form of support based on shared experiences and mutual understanding, peer support plays a unique role in maintaining college students’ mental health. Through emotional resonance, experience sharing, and mutual companionship, it provides students experiencing psychological distress with a low-threshold, highly accessible support resource. This form of support not only alleviates the current supply-demand imbalance in university psychological counseling services but also stimulates college students’ agency and spirit of mutual assistance, building a more inclusive and supportive campus cultural atmosphere.
Existing research has extensively explored the relationship between peer support and college students’ mental health. Studies have found that peer support interventions in higher education institutions include three main forms: peer-led support groups, peer tutoring, and peer learning, which show potential value in improving student mental health [1]. Zhang et al.‘s longitudinal study confirmed that college students with higher levels of perceived social support showed lower psychological problem symptoms, and this protective effect had sustained characteristics [2]. Grégoire et al. found through randomized controlled trials that online peer support programs based on Acceptance and Commitment Therapy significantly reduced college students’ psychological inflexibility, stress, anxiety, and depression levels [3]. However, current research still has several key issues requiring further exploration: most studies adopt cross-sectional designs, making it difficult to reveal the dynamic process and long-term effects of peer support on mental health [4]; exploration of peer support mechanisms mainly focuses on single mediating variables, lacking systematic analysis of multiple mediation pathways, particularly chain mediation mechanisms [5]; while studies have explored the effects of peer support on adolescent anxiety and depression, the mechanisms of action in specific contexts still require in-depth research [6].
Based on the current research status, this study adopts a three-year longitudinal tracking design and constructs a chain mediation model containing self-efficacy and social adaptation (as shown in Fig. 1) to deeply explore the long-term enhancement mechanisms of peer support on college students’ mental health. This study’s theoretical framework integrates multiple theoretical perspectives. First, it draws on social support theory [7], which posits that social support can both directly benefit mental health (main effect model) and buffer the negative impact of stressors (stress-buffering model). Second, it incorporates Bandura’s social cognitive theory [8], suggesting that peer modeling and vicarious learning experiences enhance self-efficacy beliefs, which in turn promote adaptive behaviors and positive mental health outcomes. Third, the framework aligns with Arnett’s emerging adulthood theory [9], recognizing that the college years represent a critical developmental period characterized by identity exploration, self-focus, and the gradual acquisition of adult roles and responsibilities. During this developmental stage, peer support serves as a crucial resource that facilitates both psychological growth and social integration.The hypothesized theoretical framework and chain mediation model are presented in Fig. 1.
Fig. 1.
Theoretical framework and chain mediation model. This figure illustrates the hypothesized chain mediation model where peer support (PS) influences mental health (MH) through three pathways: (1) direct effect (c’ = 0.12), (2) indirect effect via self-efficacy (SE) alone (a₁ × b₁ = 0.42 × 0.24 = 0.10), (3) indirect effect via social adaptation (SA) alone (a₂ × b₂ = 0.18 × 0.36 = 0.06), and (4) chain mediation effect via SE and SA sequentially (a₁ × d₂₁ × b₂ = 0.42 × 0.38 × 0.36 = 0.05). All path coefficients shown are standardized estimates from the final structural model. Control variables (gender, age, grade, SES) are included in the model but omitted from the figure for clarity. *p <.05, **p <.01, ***p <.001
The innovations of this research lie in: using multi-time-point longitudinal data to capture dynamic relationships between variables, breaking through the causal inference limitations of cross-sectional studies; constructing and testing the chain mediation pathway of “peer support→self-efficacy→social adaptation→mental health,” revealing the progressive mechanism by which peer support enhances individuals’ internal psychological resources (self-efficacy), thereby promoting external adaptive capacity (social adaptation), and ultimately improving mental health; employing both traditional Cross-Lagged Panel Models (CLPM) and Random Intercept Cross-Lagged Panel Models (RI-CLPM) as complementary analytical approaches to distinguish between-person associations from within-person dynamics; combining the characteristics of Chinese university contexts to provide empirical evidence and theoretical guidance for establishing localized peer psychological support systems.
Research status and literature review
Research on direct effects of peer support on mental health
In recent years, domestic and international scholars have conducted in-depth exploration of the direct relationship between peer support and college students’ mental health, with research results showing complex and diverse characteristics. Pointon-Haas et al.‘s systematic review of peer support interventions in higher education institutions found that peer-led support groups showed mixed results in improving student mental health. Researchers used 20 different measurement tools to assess 14 mental health and well-being outcomes, and this diversity of measurement tools reflects the field’s uncertainty in intervention goal setting [1]. Yeo et al.‘s study confirmed that peer support programs can significantly improve the mental health levels of students experiencing anxiety and depression, particularly showing outstanding performance in improving hope, empowerment, and enhancing self-efficacy [10]. The 2023–2024 Healthy Minds Study report from the United States showed that among 104,000 surveyed college students, the proportion with severe depression symptoms decreased from 23% in 2022 to 19% in 2024, and moderate depression symptoms decreased from 44% to 38%. This improvement trend is partly attributed to the improvement of university peer support systems and strengthening of mutual support networks among students [11].
Research progress on mediation mechanisms of peer support
Regarding the internal mechanisms by which peer support affects mental health, researchers have begun to focus on exploring multiple mediation pathways. Recent studies indicate that self-efficacy plays a key mediating role between peer support and mental health. Dominguez et al.‘s research found that peer support improves social support quality and reduces isolation by enhancing empathy and social self-efficacy, ultimately promoting mental health recovery [12]. Simmons et al.‘s systematic review further confirmed that peer support from those with similar experiences has significant improvement effects on adolescent anxiety and depression, with mechanisms mainly realized through enhancing self-esteem, strengthening effective coping abilities, and reducing loneliness [13]. Abrams et al.‘s study of medical school peer supporters found that students providing peer support showed higher empathy and self-efficacy scores after participating in programs, and this bidirectional benefit mechanism provides a new perspective for understanding the complex effects of peer support [14]. Chinese scholars exploring factors affecting college students’ mental health found that perceived teacher support and peer relationships have positive predictive effects on mental health through the mediation of real and online altruistic behaviors, expanding the boundaries of traditional mediation mechanism research [15].
New findings from longitudinal tracking studies
Longitudinal research designs show unique advantages in revealing the long-term effects of peer support. Worley et al.‘s longitudinal study of 251 American college students found a positive prospective association between peer support and academic ability, while anxiety levels were negatively correlated with future academic ability, indicating that peer support has protective effects on students’ long-term development [16]. Weber researchs’ longitudinal study of college students showed that during the COVID-19 pandemic, loneliness was significantly correlated with depression and anxiety symptoms at time point 1, while adaptive coping strategies and peer support played protective roles in alleviating mental health problems [17]. A longitudinal health study of music major students showed that at the end of the first academic year compared to the beginning, students receiving peer support showed significant improvements in physical and mental health status, health-related attitudes, and coping strategies [18]. Li et al.‘s two-year study of Chinese college students found that students’ mental health status in 2023 showed significant improvement compared to 2022, with psychological capital, particularly hope and self-efficacy enhancement, playing important roles, and perceived peer support being an important predictor of mental health improvement [19].
Localized research in cultural context
Peer support research in the Chinese cultural context shows unique characteristics and development trends. Through focus group interviews, Ning et al. found that main mental health challenges facing Chinese college students in the post-pandemic era include sleep difficulties, anxiety, and stress, with academic pressure and social influences including peer pressure and pursuit of social recognition as main causes, highlighting the special importance of peer support in collectivist cultural contexts [20]. Han et al.‘s large-scale survey of 41,620 Chinese college students showed detection rates of 9.8% for depression, 15.5% for anxiety, and 6.5% for comorbidity, emphasizing the key role of strengthening peer support system construction in improving student mental health [21]. Fan et al.‘s research based on the Psychological Resilience Dynamic System Model found that social support, particularly peer support, together with self-efficacy, constitutes the core protective factor network affecting Chinese college students’ mental health [22]. Gao et al. proposed integrating mental health literacy education into university mental health education systems, emphasizing the unique value of peer education in enhancing students’ mental health literacy, providing new ideas for mental health education reform in the post-pandemic era [23].
Implementation effect evaluation of intervention programs
Research on effect evaluation of peer support intervention programs is increasingly rich and in-depth. Through randomized controlled trials, Grégoire et al. found that online peer support programs based on Acceptance and Commitment Therapy can significantly reduce college students’ stress, anxiety, and depression levels while improving psychological flexibility and well-being [3]. The 2023 report from Mental Health America indicates that peer support, as an evidence-based practice method, effectively reduces overall mental health service costs by reducing rehospitalization rates and hospital days [24]. Richard et al.‘s review of 17 peer support programs for young adults found significant associations between peer support and enhanced self-esteem, effective coping, and reduced depression, loneliness, and anxiety [4]. Chinese scholars’ meta-analysis research shows that positive coping strategies including effective emotion regulation, maintaining adequate sleep and moderate exercise, developing optimistic attitudes, and providing mutual peer support and accessible care have been proven to effectively promote college students’ mental health [25].
New progress in chain mediation model research
Chain mediation models show important value in revealing peer support mechanisms. Zhang et al.‘s research on Chinese college students found chain mediation effects between mindfulness and depression, with meaning in life and psychological resilience playing serial mediation roles with effect sizes of 13% and 22% respectively [26]. Ni et al. explored the impact of academic involution atmosphere on college students’ psychological exhaustion, finding that relative deprivation and academic pressure perception play chain mediation roles, providing a new perspective for understanding the importance of peer support in competitive environments [27]. Lo et al.‘s study of 568 Taiwanese college students showed that social support affects mental health literacy through sequential mediation of mindfulness and hope, with total explained variance reaching 33.9% [28]. Wang et al.‘s cross-cultural research found that COVID-19 symptoms affect mental health outcomes through chain mediation of health information needs and pandemic impact perception, with this model validated in American, Asian, and European samples [29].
Recent methodological advances in longitudinal mediation analysis
Recent years have witnessed significant methodological developments in longitudinal mediation analysis. The Random Intercept Cross-Lagged Panel Model (RI-CLPM), introduced by Sui et al. represents a major advance in separating within-person dynamics from stable between-person differences [30]. Unlike traditional CLPM which conflates these two sources of variation, RI-CLPM includes random intercepts to capture trait-like individual differences while allowing cross-lagged paths to represent genuine within-person fluctuations. This distinction is crucial for mediation analysis, as between-person associations may reflect selection effects rather than causal processes. Additionally, Latent Change Score models (LCS) and Latent Curve Models with Structured Residuals (LCM-SR) offer alternative approaches for modeling dynamic mediation processes [31]. While these advanced methods provide valuable insights, traditional CLPM remains useful for examining population-level temporal associations and has been widely validated in mental health research [32]. The current study employs both traditional CLPM and RI-CLPM as complementary approaches to ensure robust conclusions about the mediating mechanisms of peer support.
Research on social adaptation and peer support
As an important indicator of college student development, social adaptation is closely related to peer support. Sui et al.‘s research found that physical exercise affects college students’ social adaptation ability through chain mediation of self-esteem and peer attachment, with students regularly participating in physical activities showing higher security in peer attachment [30]. Koo et al.‘s year-long tracking study of international students found that acculturation stress significantly decreased at the end of the first academic year compared to the beginning, with peer social support playing an important buffering role in this process, particularly support from local friends being crucial for developing cultural knowledge and abilities [31]. Cobo-Rendón et al.‘s longitudinal study of Chilean college students showed that perceived social support from friends, family, and significant others has statistically significant effects on changes in well-being, with peer support having adaptive value in higher education environments [32]. Osborr et al.‘s research found that even with implementation challenges, training, supervision, and one-on-one support services in peer support programs are proven to be feasible, acceptable, and safe [33].
Research design and methods
Research design framework and theoretical model construction
This study adopts a three-year longitudinal tracking design, constructing a chain mediation model containing self-efficacy and social adaptation to deeply explore the long-term enhancement mechanisms of peer support on college students’ mental health. The research design follows longitudinal panel data collection principles, conducting tracking measurements of the same participants at six time points (T1-T6), with 6-month intervals between each time point to capture dynamic relationships and causal directions between variables. The theoretical model construction is based on an integrated framework of social support theory, social cognitive theory, and ecological systems theory, with peer support as the exogenous variable, self-efficacy and social adaptation as endogenous mediating variables, and mental health as the final outcome variable, forming a progressive chain mediation pathway of “peer support→self-efficacy→social adaptation→mental health” (as shown in Fig. 2).
Fig. 2.
Research Framework with Time-Lagged Structure. This figure illustrates the data collection timeline across six waves from T1 (September 2022, baseline) to T6 (March 2025, + 30 months), with 6-month intervals between adjacent time points. Each time point indicates the measurement wave and corresponding calendar period. Sample retention rates are presented in Table 2. PS Peer Support, SE Self-Efficacy, SA Social Adaptation, MH Mental Health
The chain mediation model was tested using a time-lagged approach to ensure proper causal ordering: peer support at T1 predicting self-efficacy at T2, self-efficacy at T2 predicting social adaptation at T3, and social adaptation at T3 predicting mental health at T4. This temporal separation between predictor and outcome variables strengthens causal inference and reduces the potential for reverse causation. Additionally, we tested alternative temporal sequences (e.g., T1→T3→T5→T6) as sensitivity analyses to verify the robustness of the mediation pathways.The simplified research timeline is presented in Fig. 2.
The core of the research design lies in establishing temporal relationships between variables through multi-time-point measurements, thereby enhancing the validity of causal inference. As illustrated in Fig. 2, the chain mediation model framework clarifies the theoretical relationship paths between variables, where peer support affects mental health through three mediation pathways: simple mediation through self-efficacy alone, simple mediation through social adaptation alone, and chain mediation through self-efficacy and social adaptation. Control variables include demographic characteristics such as gender, age, grade, and socioeconomic status, which are treated as covariates in the model to control for their potential confounding effects.
Research participants and sampling strategy
The study employed stratified cluster sampling, selecting 2–3 comprehensive universities from eastern, central, and western regions of China, with different colleges’ freshmen randomly selected as research participants. Inclusion criteria were: (1) full-time undergraduate students enrolled in participating universities, (2) aged 17–25 years at baseline, (3) ability to provide informed consent (or assent with parental consent for minors), (4) sufficient Chinese language proficiency to complete questionnaires. Exclusion criteria were: (1) diagnosed severe mental illness requiring ongoing psychiatric treatment, (2) current use of psychotropic medications that might affect study variables, (3) planned transfer or dropout from the university within the study period.
Sample size calculation was based on statistical power requirements for structural equation modeling, estimated using the following formula:
 |
1 |
Where 
is the standard normal distribution value corresponding to the significance level (1.96), 
is the standard normal distribution value corresponding to statistical power (0.84), σ is the population standard deviation, d is the effect size, ρ is the intraclass correlation coefficient, k is the number of measurements, and design effect is set at 1.5 to account for cluster sampling effects. The intraclass correlation coefficient (ICC) for university clustering was calculated as 0.03, suggesting minimal clustering effects; nonetheless, we computed cluster-robust standard errors as a sensitivity check to ensure the validity of our findings. Considering attrition issues in longitudinal research, the initial sample size was set at 2,200, with expected final effective sample of no less than 1,500.
Among the final sample of 1,842 participants, 12.3% (n = 226) were under 18 years at baseline (M age = 17.6, SD = 0.3). For minors, we obtained both written parental consent and participant assent following institutional guidelines and the Declaration of Helsinki. All consent documents explicitly described the study purposes, procedures, potential risks and benefits, voluntary nature of participation, and right to withdraw at any time without penalty. Consent records are stored securely in locked filing cabinets for 7 years per institutional data retention policy. Special safeguards for minors included additional check-ins during data collection and immediate referral protocols for any signs of distress.The measurement timeline and sample distribution across regions and gender are presented in Fig. 3.
Fig. 3.
Measurement timeline and sample distribution. This figure presents three panels showing the data collection design and sample characteristics. Panel A displays the data collection timeline across six waves from September 2022 (T1, N = 1,842) to March 2025 (T6, N = 1,481), with retention rates indicated at each time point (T2: 91.3%, T3: 87.6%, T4: 85.2%, T5: 82.7%, T6: 80.4%). Panel B shows the variable measurement matrix indicating that all four core variables—peer support, self-efficacy, social adaptation, and mental health—were measured at all six time points. Panel C illustrates the sample distribution by region (Eastern: n = 778, 42.3%; Central: n = 548, 29.8%; Western: n = 516, 27.9%) and gender (Male: 47.2%; Female: 52.8%)
As shown in Fig. 3, the research timeline design and variable measurement structure cover the key developmental stages from enrollment to junior year, which is an important window period for college students’ socialization and psychological development. The sample distribution strategy fully considered regional differences, with 3 universities and approximately 800 students selected from the eastern region, and 2 universities and approximately 700 students each from central and western regions. This distribution ensures sample representativeness while controlling for potential effects of regional cultural factors.
Measurement instruments and reliability/validity indicators
In this study, the core variables were evaluated with standardized scales that were strictly psychometrically tested. All measurement tools were applied through standardized translation back translation procedures to ensure their applicability and equivalence in the context of Chinese culture. Peer support was measured using the friend support subscale in the perceived social support scale (PSS) developed by Zimet et al. The subscale assessed the perceived support of college students from peer groups through four items. The subjects were asked to rate the statements such as “I can get the emotional help and support I need from my friends” with the 7-point Likert scale. The higher the score, the stronger the perceived peer support level.
The measurement of self-efficacy was completed with the help of the general self-efficacy scale (GSEs), which was compiled by Schwarzer et al. And has been widely used around the world. It contains 10 items and uses a 4-point scoring system. It comprehensively reflects the level of individual confidence in their own ability by allowing the subjects to self evaluate the statements such as “I am confident that I can effectively handle any unexpected thing”, “no matter what happens, I can handle it freely”. The assessment of social adaptability uses the social adaptation subscale of the student adaptation to college questionnaire (sacq), which contains 20 items and adopts a 9-point scoring system, covering the adaptability of college students in many dimensions, such as interpersonal relationship building, campus activity participation, and social network construction, and can comprehensively reflect the degree of students’ integration into the university social environment. The mental health status was measured by the General Health Questionnaire-12 (GHQ-12). The 12 items of the questionnaire focused on recent emotional state, coping ability and daily function. We used the Likert scoring method (0–3 scale), with higher total scores indicating better mental health after reverse-coding negative items. This scoring approach was chosen over the GHQ method (0-0-1-1) to preserve variance and allow for more nuanced assessment of mental health levels.The psychometric properties of all measurement instruments are summarized in Table 1.
Table 1.
Psychometric properties of measurement instruments
| Variable | Scale | Items | Score Range | Cronbach’s α (T1-T6) | Test-Retest r | AVE | CR | HTMT Range |
|---|---|---|---|---|---|---|---|---|
| Peer Support | PSS-Friend | 4 | 1–7 | 0.85–0.89.85.89 | 0.72 | 0.67 | 0.89 | — |
| Self-Efficacy | GSES | 10 | 1–4 | 0.87–0.91.87.91 | 0.83 | 0.71 | 0.92 | 0.52–0.68 |
| Social Adaptation | SACQ-Social | 20 | 1–9 | 0.88–0.92.88.92 | 0.78 | 0.64 | 0.9 | 0.58–0.75 |
| Mental Health | GHQ-12 (Likert) | 12 | 0–3 | 0.82–0.86.82.86 | 0.75 | 0.62 | 0.87 | 0.61–0.78 |
AVE Average Variance Extracted, CR Composite Reliability, HTMT Heterotrait-Monotrait ratio
Test-retest reliability was assessed at two-week intervals
All Cronbach’s α values exceeded the 0.80 threshold for good internal consistency
HTMT values below 0.85 support discriminant validity between constructs
As shown in Table 1, each measurement tool showed good reliability and validity indicators in this study. In terms of internal consistency reliability, Cronbach’s α coefficient of peer support scale maintained between 0.85–0.89 at different measurement time points, self-efficacy scale was expressed between 0.87–0.91, social adaptation scale was 0.88–0.92, and mental health questionnaire was maintained in the range of 0.82–0.86. The internal consistency coefficient of all scales exceeded the good standard of 0.80. The test-retest reliability was obtained by repeated measurements at two-week intervals, and the correlation coefficients were peer support 0.72, self-efficacy 0.83, social adaptation 0.78 and mental health 0.75, respectively, indicating that each measurement tool has good cross time stability. The structural validity was tested by confirmatory factor analysis, and the model fitting indicators should meet the following criteria:
 |
2 |
The actual analysis results show that the four factor measurement model achieves an excellent fitting level (χ²/df = 2.43, RMSEA = 0.042, CFI = 0.946, TLI = 0.938), all indicators meet or exceed the critical value requirements set by formula (2), the standardized load coefficient of each item on the corresponding potential factor exceeds the minimum standard of 0.60, and the aggregate validity indicators such as average variance extraction (AVE) and combined reliability (CR) also meet the recommended level of psychometrics. Additionally, discriminant validity was assessed using the Heterotrait-Monotrait (HTMT) ratio, with all values below 0.85, supporting the distinctiveness of constructs. These evidences jointly support the effectiveness and reliability of the measurement tools in evaluating peer support, self-efficacy, social adaptation and mental health of college students.
Mathematical expression and estimation of chain mediation model
The construction of the chain mediation model is based on the theoretical framework of structural equation model, which depicts the progressive influence relationship between variables through a series of nested regression equations. The data analysis proceeded through four sequential stages: (1) Data Screening and Preparation, including missing data handling via Full Information Maximum Likelihood (FIML), outlier detection, and normality assessment; (2) Latent Growth Curve Modeling to capture developmental trajectories; (3) Cross-Lagged Panel Modeling to test bidirectional temporal relationships; and (4) Mediation Analysis with effect decomposition. The analysis workflow is illustrated in Fig. 4. Model evaluation criteria applied throughout all analyses were: χ²/df < 3, RMSEA < 0.08, CFI > 0.90, TLI > 0.90.
Fig. 4.
Data analysis strategy flowchart. This figure presents the sequential data analysis stages employed in this study. The workflow proceeds from data screening through final mediation analysis. Detailed model specifications and equations are presented in the text below
In the mathematical expression of the chain mediation model, let X be peer support, M₁ be self-efficacy, M₂ be social adaptation, and Y be mental health. The basic equations of the model can be expressed as:
 |
3 |
 |
4 |
 |
5 |
Where i is the individual, t is the time point, the subscripts (t-1, t-2, t-3) indicate time lags to ensure proper temporal ordering, i₁, i₂, i₃ are the intercept terms of each equation, a₁ and a₂ represent the direct effect of peer support on the two mediators, respectively, d₂₁ is the influence coefficient of self-efficacy on social adaptation, b₁ and b₂ are the effects of mediators on mental health, c’ is the direct effect of peer support on mental health after controlling the mediators, and e is the residual term of each equation. This parametric approach can clearly decompose the contributions of different paths, and provides a quantitative basis for understanding the mechanism of peer support.
As shown in Fig. 4, the total effect can be decomposed into the sum of direct effects and multiple indirect effects. The specific effect decomposition follows the following principles:
 |
6 |
The term a₁b₁ represents the simple mediating effect through self-efficacy, reflecting the path of peer support to improve mental health by enhancing individual self-confidence; a₂b₂ is a simple mediating effect through social adaptation, which reflects the mechanism of peer support to improve mental health by promoting social integration; the term a₁d₂₁b₂ represents the chain mediation effect through “self-efficacy → social adaptation”, which reveals that peer support first enhances self-efficacy, then improves social adaptability, and finally improves mental health.
Data analysis strategy
The study employs a multilevel analysis strategy to handle the longitudinal nested data structure. Latent growth curve models are used to describe variable developmental trajectories, capturing individual differences in initial levels and change rates through intercept and slope latent factors. Cross-lagged panel models are used to test bidirectional causal relationships between variables, examining predictive relationships across time points after controlling for autoregressive effects. Additionally, as a robustness check and to address recent methodological recommendations, we estimated Random Intercept Cross-Lagged Panel Models (RI-CLPM) to separate within-person dynamics from stable between-person differences. The RI-CLPM includes random intercepts for each construct to capture trait-like individual differences, while the cross-lagged paths represent within-person fluctuations over time. This distinction is crucial for mediation analysis, as between-person associations may reflect selection effects rather than causal processes. This dual-modeling approach allows us to examine both population-level temporal associations (via traditional CLPM) and individual-level change processes (via RI-CLPM), providing a more comprehensive understanding of the mediation mechanisms.
The basic equation for the latent growth curve model is:
 |
7 |
Where 
is the observation for individual i at time t, 
is the intercept (initial level) for individual i, 
is the slope (rate of change) for individual i, 
is the time coding, and 
is the measurement error. The means and variances of intercept and slope are:
 |
8 |
Where 
and 
are the population means for intercept and slope respectively, and 
and 
are individual deviations.
For the RI-CLPM analysis, the model specification includes:
 |
9 |
 |
10 |
Where 
and 
represent the random intercepts (stable between-person differences), and 
and
represent the within-person deviations at time t. The cross-lagged paths are then estimated between these within-person components, providing a cleaner test of within-person mediation processes.
Measurement model construction and validation
Measurement model construction follows hierarchical structure principles. The measurement structures of the four core constructs are presented in Fig. 5. Peer support uses a single-factor model with four items loading directly on the latent factor. Self-efficacy uses a second-order factor model, with academic self-efficacy and social self-efficacy as first-order factors jointly loading on the second-order general self-efficacy factor. Social adaptation contains four dimensions: interpersonal relationships, campus integration, social activities, and support networks, which together reflect college students’ social adaptation levels. Mental health uses a bifactor model, distinguishing between positive mental health (well-being, coping ability) and negative psychological symptoms (anxiety, depression), which helps provide a more comprehensive assessment of mental health status.
Fig. 5.
Measurement model structure and validation results. This figure displays the measurement model structure for the four core constructs in four panels. Panel A shows the peer support single-factor model with four indicators (PS1-PS4). Panel B presents the self-efficacy hierarchical model with academic and social self-efficacy as first-order factors. Panel C illustrates the social adaptation four-dimensional structure. Panel D shows the mental health bifactor model. Standardized factor loadings and reliability indices are presented in Table 1. Residual variances are included in the model but omitted from the figure for clarity
Measurement invariance testing proceeds from lenient to strict sequences, starting with configural invariance, then testing metric invariance, scalar invariance, strict invariance, and full invariance. Judgment criteria for invariance testing are based on changes in fit indices:
 |
11 |
Mediation effect testing uses the Bootstrap method to estimate confidence intervals for indirect effects, with 5,000 resampling iterations. The significance of indirect effects is judged based on whether the 95% bias-corrected confidence intervals include zero. The specific decomposition of mediation effects is as follows. Simple mediation effect through self-efficacy:
 |
12 |
Simple mediation effect through social adaptation:
 |
13 |
Chain mediation effect:
 |
14 |
Effect size indicators use PM (Percent Mediated) and 
for evaluation:
 |
15 |
Longitudinal research inevitably faces missing data issues. The study employs Full Information Maximum Likelihood (FIML) to handle missing data, which provides unbiased estimates under the Missing At Random (MAR) assumption. Testing of missing mechanisms is conducted through Little’s MCAR test:
 |
16 |
Sensitivity analyses are conducted through multiple methods: comparing consistency of results from different missing data handling methods (FIML, multiple imputation, deletion); testing effects of different time coding schemes on growth curve parameters; evaluating effects of including or excluding control variables on core path coefficients. Model robustness is evaluated through cross-validation, randomly dividing the sample into training (70%) and validation (30%) sets and comparing parameter estimates between subsamples.
The rigor of the research design is reflected in multiple aspects: enhancing validity of causal inference through longitudinal tracking design, capturing dynamic change processes through multi-time-point measurements, revealing complex mechanisms through chain mediation models, and handling nested data structures through multilevel analysis strategies. This comprehensive research design provides a solid methodological foundation for deeply understanding the long-term enhancement mechanisms of peer support on college students’ mental health.
Results
Preliminary analysis
The effective data of 1,842 college students were finally obtained. The retention rates at six time points were 100% (T1), 91.3% (T2), 87.6% (T3), 85.2% (T4), 82.7% (T5) and 80.4% (T6), respectively. The overall loss rate was controlled within 20%, which met the quality requirements of the longitudinal study. To assess potential attrition bias, we compared baseline characteristics between participants who completed all six waves (n = 1,481) and those who dropped out (n = 361). The sample characteristics and descriptive statistics across all measurement points are presented in Table 2.
Table 2.
Sample characteristics, attrition analysis, and descriptive statistics
| Category | Variable | Completers (n = 1,481) | Dropouts (n = 361) | t/χ² | p |
|---|---|---|---|---|---|
| Demographics | Age (years) | 18.34 (0.73) | 18.42 (0.76) | 1.23 | 0.22 |
| Gender (% female) | 52.60% | 53.40% | 0.28 | 0.6 | |
| Eastern region (%) | 42.10% | 42.90% | 0.52 | 0.77 | |
| Central region (%) | 29.70% | 30.20% | 0.18 | 0.84 | |
| Western region (%) | 28.20% | 26.90% | 0.41 | 0.69 | |
| Baseline Variables | Peer Support | 5.13 (1.23) | 5.09 (1.27) | 0.39 | 0.7 |
| Self-Efficacy | 2.85 (0.51) | 2.82 (0.54) | 0.67 | 0.5 | |
| Social Adaptation | 6.25 (1.34) | 6.22 (1.37) | 0.28 | 0.78 | |
| Mental Health | 2.97 (0.68) | 2.94 (0.71) | 0.52 | 0.61 |
| Variable | Time Point | T1 | T2 | T3 | T4 |
|---|---|---|---|---|---|
| Peer Support | Mean (SD) | 5.12 (1.24) | 5.24 (1.22) | 5.38 (1.19) | 5.45 (1.17) |
| Self-Efficacy | Mean (SD) | 2.84 (0.52) | 2.93 (0.50) | 3.02 (0.49) | 3.09 (0.48) |
| Social Adaptation | Mean (SD) | 6.24 (1.35) | 6.33 (1.33) | 6.42 (1.32) | 6.54 (1.30) |
| Mental Health | Mean (SD) | 2.96 (0.70) | 3.04 (0.69) | 3.11 (0.69) | 3.17 (0.68) |
Values are M (SD) unless otherwise indicated. No significant differences were found between completers and dropouts on any baseline variable (all p >.05), supporting the Missing At Random (MAR) assumption
N = 1,842 at T1. All variables showed significant positive growth trends from T1 to T6
As shown in Table 2, no significant differences were found between completers and dropouts on key baseline variables including age (t = 1.23, p =.22), gender (χ² = 0.28, p =.60), baseline mental health (t = 0.52, p =.61), and baseline peer support (t = 0.39, p =.70). These results support the Missing At Random (MAR) assumption, suggesting that attrition was not systematically related to the study variables and that Full Information Maximum Likelihood (FIML) estimation is appropriate for handling missing data.
The distribution of demographic characteristics of the sample shows that 47.2% of the samples are male and 52.8% of the samples are female, with an average age of 18.36 years (SD = 0.74). In terms of regional distribution, the eastern region accounts for 42.3%, the central region accounts for 29.8%, and the western region accounts for 27.9%. This distribution feature is basically consistent with the regional differences of higher education in China. As shown in Table 2, all variables showed significant positive growth trends across the six measurement points. The peer support level gradually increased from 5.12 points (SD = 1.24) in T1 to 5.56 points (SD = 1.15) in T6, reflecting that with the deepening of college life, the support network among students has gradually improved and consolidated. The self-efficacy showed a more obvious growth pattern, from 2.84 points (SD = 0.52) at the beginning of enrollment to 3.18 points (SD = 0.47) at the end of junior year. This continuous growth trend showed that the university education environment played a positive role in the cultivation of students’ self-confidence and ability. The correlation analysis results show that there is a moderate degree of positive correlation between variables. The autocorrelation coefficient of the same variable at different time points is between 0.61 and 0.74, indicating that the individual characteristics have high stability, while the correlation coefficient between different variables is between 0.28 and 0.64, which provides a good basis for the subsequent mediation effect analysis.
Measurement model testing
The results of confirmatory factor analysis of the measurement model support the theoretical hypothesis of four factor structure, and the model fitting index has reached a good level: χ²/df = 2.43, RMSEA = 0.042 (90% CI: 0.038–0.046.038.046), CFI = 0.946, TLI = 0.938, SRMR = 0.045. These indicators meet or exceed the recommended threshold standards. The measurement invariance test results are presented in Table 3, which confirmed the equivalence of the scale across time points and demographic groups.
Table 3.
Measurement model fit indices and invariance testing
| Invariance Type | Model | χ²/df | RMSEA [90% CI] | CFI | TLI | ΔCFI | ΔRMSEA | Decision |
|---|---|---|---|---|---|---|---|---|
| Longitudinal | Configural | 2.43 | 0.042 [0.038, 0.046] | 0.946 | 0.938 | — | — | Accept |
| Metric | 2.46 | 0.045 [0.041, 0.049] | 0.94 | 0.934 | 0.006 | 0.003 | Accept | |
| Scalar | 2.51 | 0.047 [0.043, 0.051] | 0.938 | 0.932 | 0.008 | 0.005 | Accept | |
| Strict | 2.58 | 0.052 [0.048, 0.056] | 0.931 | 0.926 | 0.015 | 0.01 | Marginal | |
| Gender | Configural | 2.45 | 0.043 | 0.945 | 0.937 | — | — | Accept |
| Metric | 2.48 | 0.044 | 0.942 | 0.934 | 0.003 | 0.001 | Accept | |
| Scalar | 2.52 | 0.045 | 0.939 | 0.931 | 0.006 | 0.002 | Accept | |
| Region | Configural | 2.47 | 0.044 | 0.943 | 0.935 | — | — | Accept |
| Metric | 2.51 | 0.045 | 0.94 | 0.932 | 0.003 | 0.001 | Accept | |
| Scalar | 2.55 | 0.046 | 0.936 | 0.928 | 0.007 | 0.002 | Accept | |
| Model Comparison | Four-Factor | 2.43 | 0.042 [0.038, 0.046] | 0.946 | 0.938 | — | — | Accept |
| Three-Factor | 4.28 | 0.078 [0.074, 0.082] | 0.862 | 0.848 | 0.084 | 0.036 | Reject | |
| Single-Factor | 7.41 | 0.126 [0.122, 0.130] | 0.692 | 0.671 | 0.254 | 0.084 | Reject |
Invariance testing criteria: ΔCFI ≤ 0.010, ΔRMSEA ≤ 0.015 indicate acceptable invariance. The four-factor model demonstrated significantly better fit than alternative models, supporting the distinctiveness of the four constructs. All ΔCFI values for gender and regional invariance were below the recommended cutoff of 0.010, allowing for valid path coefficient comparisons in subsequent multi-group analyses
As shown in Table 3, longitudinal measurement invariance was supported including configural invariance, metric invariance (ΔCFI = 0.006, ΔRMSEA = 0.003) and scalar invariance (ΔCFI = 0.008, ΔRMSEA = 0.005), which showed that the research tool maintained stable measurement attributes during the three-year follow-up period, providing a psychometric basis for longitudinal comparison. Additionally, measurement invariance was supported across both gender groups (metric: ΔCFI = 0.003; scalar: ΔCFI = 0.006) and regional groups (metric: ΔCFI = 0.003; scalar: ΔCFI = 0.007), indicating that the measurement instruments function equivalently across these demographic groups.
Factor load analysis showed that the standardized load coefficient of each item on its corresponding potential factor exceeded the minimum standard of 0.60, the factor load of peer support dimension ranged from 0.68 to 0.82, the self-efficacy dimension ranged from 0.71 to 0.86, the social adaptation dimension ranged from 0.65 to 0.79, and the mental health dimension ranged from 0.62 to 0.84. These high factor load coefficients indicated that the observation variables could effectively reflect the characteristics of potential constructs. The combined reliability (CR) test results showed that the CR values of the four core constructs were 0.89, 0.92, 0.90 and 0.87, which exceeded the recommended standard of 0.70, and the average variance extraction (AVE) were 0.67, 0.71, 0.64 and 0.62, respectively. While social adaptation (AVE = 0.64) and mental health (AVE = 0.62) were slightly below the ideal value of 0.65, the Heterotrait-Monotrait (HTMT) ratios were all below 0.85 (ranging from 0.52 to 0.78), supporting discriminant validity between constructs. On the whole, the measurement model showed good reliability and validity characteristics.
Longitudinal change trajectory analysis
The analysis results of the latent growth curve model are presented in Table 4 and visualized in Fig. 6. All variables showed a significant linear growth trend during the three-year tracking period. The initial level of peer support was 5.12 (SE = 0.03, p <.001), and the growth slope was 0.08 (SE = 0.01, p <.001), indicating that students’ perceived peer support increased steadily at a rate of about 0.08 units per semester, which reflected the step-by-step establishment and deepening process of college students’ social networks. Self efficacy showed a steeper growth trajectory, with an initial level of 2.84 (SE = 0.02, p <.001) and a growth slope of 0.11 (SE = 0.01, p <.001). This rapid growth rate may be due to the improvement of students’ sense of ability caused by constantly facing and overcoming various academic challenges during college.
Table 4.
Latent growth curve model parameter estimates
| Parameters | Peer Support | Self-Efficacy | Social Adaptation | Mental Health |
|---|---|---|---|---|
| Fixed Effects | ||||
| Intercept Mean | 5.12*** | 2.84*** | 6.24*** | 2.96*** |
| Intercept SE | −0.03 | −0.02 | −0.04 | −0.02 |
| Slope Mean | 0.08*** | 0.11*** | 0.12*** | 0.10*** |
| Slope SE | −0.01 | −0.01 | −0.02 | −0.01 |
| Random Effects | ||||
| Intercept Variance | 0.68*** | 0.42*** | 0.56*** | 0.48*** |
| Slope Variance | 0.03* | 0.02* | 0.04** | 0.02* |
| Intercept-Slope Corr | −0.15* | −0.18** | −0.12 | −0.14* |
| Model Fit | ||||
| χ²/df | 2.31 | 2.28 | 2.46 | 2.39 |
| RMSEA | 0.039 | 0.038 | 0.041 | 0.04 |
| CFI | 0.958 | 0.961 | 0.952 | 0.956 |
*p <.05,**p <.01,*****p <.001
The negative intercept-slope correlations indicate compensatory growth patterns, where students with lower initial levels showed faster growth rates
Fig. 6.
Longitudinal development trends. This figure displays the longitudinal developmental trajectories of the four core variables across six time points (T1: September 2022 to T6: March 2025) in a four-panel faceted design, with each panel using an appropriate Y-axis scale for its respective variable. Panel A: Peer Support increased from 5.12 (T1) to 5.56 (T6), slope = 0.08, p <.01. Panel B: Self-Efficacy increased from 2.84 (T1) to 3.18 (T6), slope = 0.11, p <.001. Panel C: Social Adaptation increased from 6.24 (T1) to 6.78 (T6), slope = 0.12, p <.001. Panel D: Mental Health increased from 2.96 (T1) to 3.28 (T6), slope = 0.10, p <.001. Social adaptation demonstrated the steepest growth trajectory among the four variables. Shaded bands represent 95% confidence intervals. All slope estimates are unstandardized coefficients from latent growth curve models. **p <.01, ***p <.001
The growth mode of social adaptation shows unique characteristics. Although the initial level is high (6.24, SE = 0.04), its growth slope (0.12, SE = 0.02, p <.001) is the largest among the four variables. As shown in Fig. 6, this mode shows that college students’ social adaptation ability has a certain foundation in the early stage of enrollment, but it is still developing rapidly, especially in the establishment of interpersonal relationships, the integration of campus culture and the improvement of social skills. The development track of mental health is relatively flat, with an initial level of 2.96 (SE = 0.02) and a growth slope of 0.10 (SE = 0.01, p <.001). This stable improvement trend reflects the positive impact of university environment on students’ mental health, and also implies that the improvement of mental health is a gradual process, which requires the accumulation of time and the joint action of various factors.
The analysis of individual differences shown in Table 4 shows that the intercept variance of all variables has reached a significant level (range: 0.42–0.68, p <.001), indicating that there are significant individual differences in the starting level of students at the time of enrollment, while the slope variance is small but still significant (range: 0.02–0.04, p <.05), indicating that there are also individual differences in the growth rate of students. The negative correlation between intercept and slope (r = −.12 to − 0.18) reveals an interesting compensatory growth model, that is, students with lower level of enrollment tend to show a faster growth rate, which is particularly obvious in the dimension of self-efficacy (r = −.18, p <.01), indicating that university education environment has a stronger role in promoting students with lower initial ability.
Chain mediation effect testing
The results of structural equation model analysis support the chain mediation mechanism assumed by the study, and the model fitting indicators reach a good level (χ²/df = 2.52, RMSEA = 0.043, CFI = 0.948, TLI = 0.941, SRMR = 0.046). Path coefficient analysis showed that peer support had a significant positive predictive effect on self-efficacy (β = 0.42, p <.001), which explained the 18% variance of self-efficacy. This result showed that emotional support and practical help from peers could effectively enhance college students’ confidence in self-efficacy. Although the direct effect of peer support on social adaptation is relatively small, it is still significant (β = 0.18, p <.01), while self-efficacy has a stronger predictive effect on social adaptation (β = 0.38, p <.001). Both explain the variance of 31% of social adaptation. This hierarchical and progressive influence mode verifies the role of self-efficacy as a bridge between peer support and social adaptation. The complete structural model with path coefficients and effect decomposition is presented in Fig. 7.
Fig. 7.
Chain mediation model with path coefficients and effect decomposition. This figure presents the chain mediation analysis results in three panels. Panel A displays the structural model with standardized path coefficients: PS → SE: β = 0.42, p <.001; PS → SA: β = 0.18, p <.01; PS → MH: β = 0.12, p <.05; SE → SA: β = 0.38, p <.001; SE → MH: β = 0.24, p <.001; SA → MH: β = 0.36, p <.001. Model fit: χ²/df = 2.52, RMSEA = 0.043, CFI = 0.948, TLI = 0.941. Explained variance: R² = 0.18 for SE, R² = 0.31 for SA. Panel B shows the effect decomposition: total effect = 0.33, direct effect = 0.12 (36.4%), indirect via SE = 0.10 (30.3%), indirect via SA = 0.06 (18.2%), chain mediation via SE → SA = 0.05 (15.1%), total indirect = 0.21 (63.6%). Panel C presents Bootstrap 95% confidence intervals for all indirect effects based on 5,000 resamples, with none including zero. Residuals, covariances among exogenous variables, and control variables (gender, age, grade, SES) are included in the model but omitted from the figure for clarity. PS Peer Support, SE Self-Efficacy, SA Social Adaptation, MH Mental Health. *p <.05, **p <.01, ***p <.001
The decomposition analysis of indirect effects is detailed in Table 5. The total effect of peer support on mental health was 0.33 (SE = 0.04, 95% CI [0.25, 0.41], p <.001). The total indirect effect was 0.21 (SE = 0.03, 95% CI [0.15, 0.27], p <.001), accounting for 63.6% of the total effect. Specifically: (1) the simple mediation effect through self-efficacy was 0.10 (95% CI [0.06, 0.14], p <.001), accounting for 30.3% of the total effect; (2) the simple mediation effect through social adaptation was 0.06 (95% CI [0.03, 0.10], p =.002), accounting for 18.2% of the total effect; and (3) the chain mediation effect through self-efficacy and social adaptation was 0.05 (95% CI [0.03, 0.08], p <.001), accounting for 15.1% of the total effect. The bootstrap confidence interval test (5,000 resampling) showed that the 95% confidence intervals of all indirect effects did not contain zero, which further confirmed the robustness of the mediating effect.
Table 5.
Chain mediation effect decomposition results
| Path | Estimate | SE | 95% CI Lower | 95% CI Upper | p-value | % of Total |
|---|---|---|---|---|---|---|
| Direct Effects | ||||||
| PS → SE | 0.42 | 0.04 | 0.34 | 0.5 | < 0.001 | — |
| PS → SA | 0.18 | 0.06 | 0.06 | 0.3 | 0.003 | — |
| PS → MH | 0.12 | 0.05 | 0.02 | 0.22 | 0.019 | 36.40% |
| SE → SA | 0.38 | 0.05 | 0.28 | 0.48 | < 0.001 | — |
| SE → MH | 0.24 | 0.04 | 0.16 | 0.32 | < 0.001 | — |
| SA → MH | 0.36 | 0.04 | 0.28 | 0.44 | < 0.001 | — |
| Indirect Effects | ||||||
| PS → SE → MH | 0.1 | 0.02 | 0.06 | 0.14 | < 0.001 | 30.30% |
| PS → SA → MH | 0.06 | 0.02 | 0.03 | 0.1 | 0.002 | 18.20% |
| PS → SE → SA → MH | 0.05 | 0.01 | 0.03 | 0.08 | < 0.001 | 15.10% |
| Total Effects | ||||||
| Total Indirect | 0.21 | 0.03 | 0.15 | 0.27 | < 0.001 | 63.60% |
| Total Effect | 0.33 | 0.04 | 0.25 | 0.41 | < 0.001 | 100% |
PS Peer Support, SE Self-Efficacy, SA Social Adaptation, MH Mental Health
All coefficients are standardized estimates. Bootstrap confidence intervals based on 5,000 resamples. The percentage of total effect indicates the relative contribution of each pathway to the overall effect of peer support on mental health
The chain mediation effect showed some heterogeneity in different subgroups. The analysis of gender moderation effect showed that the indirect effect of peer support through self-efficacy (β = 0.13, p <.001) in female group was significantly higher than that in male group (β = 0.08, p <.01), and the moderation effect reached a significant level (Δχ² = 8.74, p =.003). This difference may reflect that women pay more attention to emotional connection and psychological support in interpersonal communication. The analysis of regional differences found that the chain mediation effect of students in the western region was the most obvious (total indirect effect = 0.26), followed by the eastern region (0.20) and the central region (0.18). This regional difference may be related to the allocation of support resources and cultural background of colleges and universities in different regions.
Robustness testing.
The cross-lagged panel model analysis further verified the causal relationship between variables. The results are presented in Fig. 8. The prediction coefficient of peer support T1 on mental health T2 was β = 0.15 (p <.01), while the reverse prediction coefficient of mental health T1 on peer support T2 was only β = 0.08 (p <.05). This asymmetric prediction model supported the dominant effect of peer support on mental health. After controlling the baseline level, the predictive coefficient of peer support T2 on mental health T3 increased to β = 0.18 (p <.001), indicating that the protective effect of peer support showed a cumulative increase over time.
Fig. 8.
Cross-lagged panel model results. This figure presents the cross-lagged panel analysis examining bidirectional relationships between peer support (PS) and mental health (MH) across three time points in the upper panel, and gender moderation effects in the lower panel. Upper panel: Autoregressive paths for PS were β = 0.68 (T1→T2, p <.001) and β = 0.71 (T2→T3, p <.001). Cross-lagged paths from PS to MH (β = 0.15, p <.01 from T1→T2; β = 0.18, p <.001 from T2→T3) were stronger than paths from MH to PS (β = 0.08, p <.05 from T1→T2), supporting the causal priority of peer support. Model fit: χ²/df = 2.18, RMSEA = 0.037, CFI = 0.956. Lower panel: Gender moderation analysis revealed that the indirect effect of peer support on mental health via self-efficacy was significantly stronger for female students (β = 0.13, p <.001) than for male students (β = 0.08, p <.01), with a significant interaction effect (Δχ² = 8.74, p =.003). All coefficients are standardized estimates. Paths without coefficient labels were estimated but values not individually reported. Residuals and control variables (gender, age, grade, SES) are included in the model but omitted from the figure for clarity. PS Peer Support, SE Self-Efficacy, MH Mental Health. *p <.05, **p <.01, ***p <.001
The RI-CLPM analysis provided additional insights into the within-person dynamics of the mediation process. As shown in Table 6, the within-person effects were slightly smaller but remained significant. The within-person effect of peer support on mental health was β = 0.28 (SE = 0.05, 95% CI [0.18, 0.38], p <.001), with indirect effects accounting for 58.2% of the total within-person association. Specifically, the within-person mediation through self-efficacy was 0.08 (28.6%), through social adaptation was 0.05 (17.9%), and the chain mediation was 0.04 (14.3%). These results confirm that the observed relationships reflect genuine within-person change processes rather than merely stable individual differences, strengthening the causal interpretation of our findings.
Table 6.
RI-CLPM Within-Person mediation effects
| Path | Within-Person Effect | SE | 95% CI | p-value | % of Within-Person Total |
|---|---|---|---|---|---|
| Direct Effect | |||||
| PS → MH | 0.12 | 0.04 | [0.04, 0.20] | 0.003 | 41.80% |
| Indirect Effects | |||||
| Via Self-Efficacy | 0.08 | 0.02 | [0.04, 0.12] | < 0.001 | 28.60% |
| Via Social Adaptation | 0.05 | 0.02 | [0.01, 0.09] | 0.011 | 17.90% |
| Chain Mediation | 0.04 | 0.01 | [0.02, 0.06] | < 0.001 | 14.30% |
| Total Within-Person | |||||
| Total Indirect | 0.16 | 0.03 | [0.10, 0.22] | < 0.001 | 58.20% |
| Total Effect | 0.28 | 0.05 | [0.18, 0.38] | < 0.001 | 100% |
RI-CLPM Random Intercept Cross-Lagged Panel Model. Within-person effects represent genuine individual change processes after separating stable between-person differences through random intercepts. The consistency between traditional CLPM and RI-CLPM results supports the robustness of the mediation findings
The sensitivity analysis results are summarized in Table 7 and visualized in Fig. 9. The comparison of different missing data processing methods showed that the total effect obtained by the full information maximum likelihood estimation was 0.33 (SE = 0.04), the multiple imputation method was 0.32 (SE = 0.04), the list deletion method was 0.30 (SE = 0.05), and the mean substitution method was 0.29 (SE = 0.06). The difference of the effect values obtained by the four methods was no more than 0.04, and there was a large amount of overlap in the confidence interval, indicating that the research results were highly robust for the selection of missing data processing methods.
Table 7.
Summary of robustness test results
| Analysis Type | Method/Model | Key Parameter | Estimate (SE) | 95% CI | Δ from Main |
|---|---|---|---|---|---|
| Missing Data Handling | Total Effect | ||||
| FIML (Main) | 0.33 (0.04) | [0.25, 0.41] | — | ||
| Multiple Imputation | 0.32 (0.04) | [0.24, 0.40] | −0.01 | ||
| Listwise Deletion | 0.30 (0.05) | [0.20, 0.40] | −0.03 | ||
| Mean Substitution | 0.29 (0.06) | [0.17, 0.41] | −0.04 | ||
| Control Variables | PS → MH | ||||
| No Controls | 0.35 (0.04) | [0.27, 0.43] | 0.02 | ||
| + Demographics | 0.33 (0.04) | [0.25, 0.41] | 0 | ||
| + SES | 0.32 (0.04) | [0.24, 0.40] | −0.01 | ||
| Full Model | 0.30 (0.04) | [0.22, 0.38] | −0.03 | ||
| Time Coding | Growth Slope | ||||
| Linear | 0.11 (0.01) | [0.09, 0.13] | — | ||
| Quadratic | 0.10 (0.01) | [0.08, 0.12] | −0.01 | ||
| Free Estimation | 0.12 (0.02) | [0.08, 0.16] | 0.01 |
All estimates are standardized coefficients. FIML Full Information Maximum Likelihood. The small differences across methods (Δ ≤ 0.04) and overlapping confidence intervals demonstrate high robustness of the findings
Fig. 9.
Sensitivity analysis results. This figure presents sensitivity analysis results in four panels. Panel A compares total effect estimates across four missing data handling methods: FIML (reference, 0.33, 95% CI [0.25, 0.41]), Multiple Imputation (0.32, 95% CI [0.24, 0.40]), Listwise Deletion (0.30, 95% CI [0.20, 0.40]), and Mean Substitution (0.29, 95% CI [0.17, 0.41]); all estimates fell within an acceptable range with overlapping confidence intervals. Panel B displays model comparison using CFI and RMSEA: chain mediation model (CFI = 0.948, RMSEA = 0.043), parallel mediation model (CFI = 0.926, RMSEA = 0.054), and single mediation model (CFI = 0.912, RMSEA = 0.062); ΔAIC = 124.36 and ΔBIC = 108.52 support the chain mediation model. Panel C shows the stability of the PS → MH path coefficient across model specifications: no controls (0.35), +demographics (0.33), +SES (0.32), and full model (0.30); the effect remained significant across all specifications. Panel D compares growth slope estimates across different time coding schemes: linear (0.11, 95% CI [0.09, 0.13]), quadratic (0.10, 95% CI [0.08, 0.12]), and free estimation (0.12, 95% CI [0.08, 0.16]). Error bars represent 95% confidence intervals. FIML Full Information Maximum Likelihood, PS Peer Support, MH Mental Health
The comparative analysis of alternative models shows that the goodness of fit of the chain mediation model (CFI = 0.948, RMSEA = 0.043) is significantly better than that of the parallel mediation model (CFI = 0.926, RMSEA = 0.054) and the single mediation model (CFI = 0.912, RMSEA = 0.062). The comparison of Akaike information criterion (AIC) and Bayesian information criterion (BIC) also supports the chain mediation model as the optimal model, ΔAIC = 124.36, ΔBIC = 108.52. These results verify the rationality of the research hypothesis from the perspective of model comparison.
Discussion and implications
Discussion
This study systematically tested the long-term enhancement mechanisms of peer support on college students’ mental health through a three-year longitudinal tracking design. Results supported the hypothesized chain mediation model, with peer support promoting mental health improvement by sequentially enhancing self-efficacy and social adaptation levels, achieving a total effect of 0.33 (p <.001), with 63.6% (not 21% as might be misinterpreted) realized through mediating variables. This finding not only validates the integrated framework of social support theory and social cognitive theory but more importantly reveals the specific psychological mechanism pathways through which peer support operates. These findings extend multiple theoretical perspectives in meaningful ways. From the lens of Bandura’s social cognitive theory [8], our results demonstrate that peer support operates through both cognitive (self-efficacy) and behavioral (social adaptation) pathways, with the cognitive pathway showing stronger effects. This aligns with the notion that observational learning and vicarious experiences from peers first shape belief systems before manifesting in behavioral changes. Furthermore, our findings support Arnett’s emerging adulthood framework by showing that the college years represent a critical period where peer relationships serve as primary developmental resources [9], particularly when students are physically separated from family support systems.
The study found that peer support’s direct effect on self-efficacy was most significant (β = 0.42), echoing Masten et al.‘s view that social self-efficacy plays a bridging role between social support and mental health [34]. However, this study further found that self-efficacy is not an endpoint but forms more complex cascade effects by enhancing social adaptation ability (β = 0.38). This progressive influence pattern can be understood through the lens of developmental cascade models, which posit that competence in one domain creates resources that facilitate success in subsequent domains. In our model, enhanced self-efficacy beliefs (cognitive domain) provide the confidence necessary for engaging in social activities and building relationships (behavioral domain), which ultimately contributes to better mental health outcomes. This cascading process is particularly relevant during the college transition, when students must simultaneously navigate academic challenges and establish new social networks.
Longitudinal analysis results revealed the temporal dynamic characteristics of peer support effects. Cross-lagged models showed that peer support’s predictive coefficient for mental health increased from 0.15 during T1-T2 to 0.18 during T2-T3. This cumulative enhancement pattern has important theoretical implications. It suggests that peer support effects are not merely concurrent but build over time through what Cohen and Wills termed “support bank” accumulation - the gradual buildup of social resources that can be drawn upon during times of stress [7]. Our findings extend this concept by showing that the accumulation occurs not just in terms of available support but also in the psychological resources (self-efficacy) and behavioral competencies (social adaptation) that support engenders. The research hypothesis testing results are summarized in Table 8.
Table 8.
Research hypothesis testing results
| Hypothesis | Path | β | 95% CI | Support | Theoretical Implication |
|---|---|---|---|---|---|
| H1: Direct effect | PS → MH | 0.12* | [0.02, 0.22] | Supported | Direct buffering effect exists |
| H2: SE mediation | PS → SE → MH | 0.10*** | [0.06, 0.14] | Supported | Cognitive mediation pathway |
| H3: SA mediation | PS → SA → MH | 0.06** | [0.03, 0.10] | Supported | Behavioral adaptation pathway |
| H4: Chain mediation | PS → SE → SA → MH | 0.05*** | [0.03, 0.08] | Supported | Progressive mechanism confirmed |
| H5: Gender moderation | Δχ² | 8.74** | — | Supported | Female students benefit more |
| H6: Cumulative effect | T1-T2 → T2-T3 | 0.15** → 0.18*** | — | Supported | Enhancement over time |
| H7: Regional differences | F | 4.82** | — | Supported | Western > Eastern > Central |
| H8: Compensatory growth | r (intercept-slope) | −0.18** | — | Supported | Lower baseline, faster growth |
PS Peer Support, SE Self-Efficacy, SA Social Adaptation, MH Mental Health
All coefficients are standardized estimates
Detailed path coefficients and effect decomposition are presented in Fig. 7
*p <.05, **p <.01, *p <.001
The RI-CLPM results provide crucial evidence for within-person processes, addressing a key limitation in previous peer support research that often conflated between-person and within-person effects. The finding that within-person effects (β = 0.28) were slightly smaller than between-person effects (β = 0.33) but remained substantial suggests that peer support genuinely changes individual trajectories rather than simply reflecting stable individual differences in sociability or support-seeking tendencies. This distinction is critical for intervention development, as it confirms that enhancing peer support can produce real changes in mental health outcomes even for students who may not naturally gravitate toward social connections.
The study also found significant compensatory growth patterns, with students having lower initial mental health levels showing steeper growth trajectories (r = −.18, p <.01). This finding aligns with and extends psychological resilience theory in several ways. First, it supports the “steeling effect” hypothesis (Rutter, 2012), which suggests that individuals who face manageable challenges develop stronger coping resources. In the university context, students entering with mental health vulnerabilities may be more motivated to utilize available peer support, leading to accelerated growth. Second, this pattern reflects what we term “support responsiveness” - the differential benefit from social resources based on initial need levels. This has important implications for resource allocation in university mental health services, suggesting that peer support programs may be particularly cost-effective when targeted toward at-risk students.
The observed gender and regional differences warrant careful cultural and contextual interpretation. Female students’ stronger benefit from peer support (chain mediation effect: β = 0.13 vs. 0.08 for males) likely reflects multiple intersecting factors. From a socialization perspective, Chinese cultural norms encourage emotional expression and interpersonal connection more strongly in females, potentially making them more skilled at both providing and utilizing peer support. Additionally, research on gender and coping suggests that women more frequently employ emotion-focused and social coping strategies, which align well with the mechanisms through which peer support operates. The regional variations (Western: 0.26 > Eastern: 0.20 > Central: 0.18) may reflect differential availability of professional mental health resources across China’s regions. Western provinces, with fewer mental health professionals per capita, may see stronger peer support effects as these informal networks partially compensate for limited formal services. This interpretation is supported by the social compensation hypothesis, which suggests that informal support becomes more critical when formal support is scarce.
Implications
Research results have important guiding significance for university mental health education practice. The validation of the chain mediation mechanism indicates that effective peer support interventions must go beyond simply increasing social contact frequency. Instead, interventions should be designed to systematically build the complete “support-efficacy-adaptation-health” pathway. This can be operationalized through structured programs that include: (1) peer mentoring components that explicitly focus on building self-efficacy through guided mastery experiences and positive feedback; (2) social skills training that translates enhanced confidence into effective interpersonal behaviors; and (3) regular assessment and reinforcement of progress along each step of the chain. For example, peer counseling programs could incorporate structured activities where students first share success experiences (building efficacy), then practice social skills in supportive environments (enhancing adaptation), with facilitators explicitly linking these experiences to mental health improvements.
When implementing peer support programs, universities should adopt a differentiated approach based on student characteristics. For female students, programs can emphasize emotional processing and mutual support aspects, incorporating activities like support circles and emotion-focused discussions. For male students, programs might benefit from more task-oriented and problem-solving elements, such as study groups that naturally incorporate supportive interactions. This gender-sensitive approach does not imply rigid stereotypes but rather acknowledges different socialization experiences and comfort levels with various support modalities.
Regional considerations suggest that universities in western China, given their limited professional mental health resources, should prioritize peer support system development as a primary intervention strategy. This could include: (1) training larger cohorts of peer counselors to ensure adequate coverage; (2) establishing formal partnerships between peer support programs and the limited professional services available; (3) developing culturally adapted materials that resonate with local values and communication styles. Eastern region universities, with better professional resources, might position peer support as a complementary first-line intervention that identifies and refers more complex cases to professional services.
The discovery of compensatory growth patterns suggests that universities need to adapt to local conditions and individual differences when building peer support systems. For students identified as at-risk during entrance mental health screening, universities should implement “enhanced peer support protocols” that include: (1) immediate assignment to peer mentors; (2) inclusion in small support groups with trained facilitators; (3) regular check-ins to monitor progress along the mediation pathway; and (4) clear escalation procedures for cases requiring professional intervention. This targeted approach maximizes the benefit-to-cost ratio of peer support programs while ensuring vulnerable students receive timely support.
The longitudinal findings showing cumulative enhancement effects have implications for program duration and structure. Rather than one-time orientations or brief interventions, peer support should be conceptualized as an ongoing developmental process spanning the entire college experience. Universities might implement “developmental peer support” models where the focus evolves across academic years: freshman year emphasizing basic social connection and belonging, sophomore year targeting self-efficacy building through academic and social challenges, and junior year focusing on advanced social adaptation skills needed for career preparation and adult transitions.
Conclusion
Through a three-year longitudinal tracking design, this study collected data from 1,842 college students at six time points, systematically validating the theoretical model of peer support producing long-term enhancement effects on mental health through the chain mediation mechanism of self-efficacy and social adaptation. The study found that peer support’s total effect on mental health reached 0.33 (SE = 0.04, 95% CI [0.25, 0.41], p <.001), with indirect effects totaling 0.21 (SE = 0.03, 95% CI [0.15, 0.27], p <.001), accounting for 63.6% of the total effect. Specifically: (1) 30.3% of the total effect was mediated through self-efficacy alone (indirect effect = 0.10), (2) 18.2% through social adaptation alone (indirect effect = 0.06), and (3) 15.1% through the chain mediation of “self-efficacy→social adaptation” (indirect effect = 0.05). All indirect effect confidence intervals excluded zero, confirming the robustness of these mediation pathways. This multipath influence mechanism reveals the complexity and hierarchical nature of peer support effects.
Latent growth curve model analysis showed that all four core variables presented significant linear growth trends, with social adaptation having the largest growth slope (0.12), followed by self-efficacy (0.11) and mental health (0.10). Although peer support growth was relatively slow (0.08), it provided a stable foundation for the development of other variables. Cross-lagged panel analysis further confirmed causal directions between variables, with peer support’s predictive coefficient for mental health increasing from 0.15 to 0.18, showing cumulative enhancement characteristics. The Random Intercept Cross-Lagged Panel Model (RI-CLPM) analysis provided additional support for within-person mediation processes, with within-person effects accounting for 58.2% of the total within-person association between peer support and mental health. This confirms that the observed relationships reflect genuine individual change processes rather than merely stable between-person differences.
The discovery of compensatory growth patterns (r = −.18) indicates that the university environment has stronger promoting effects for students with lower initial levels. These findings not only deepen theoretical understanding of social support mechanisms but also provide scientific basis for universities to build differentiated and precise peer support systems.
The study achieved important progress in theoretical innovation and practical application. The chain mediation model constructed by integrating social support theory, social cognitive theory, and ecological systems theory breaks through previous research’s limitations of focusing on single mediating variables. It reveals the progressive mechanism by which peer support affects mental health through dual pathways of cognition (self-efficacy) and behavior (social adaptation), providing a new analytical framework for understanding the complex influence processes of social support. The use of both traditional CLPM and RI-CLPM approaches represents a methodological advance, allowing for more nuanced understanding of how peer support operates at both population and individual levels.
The discovered gender differences (female chain mediation effect 0.13 vs. male 0.08) and regional differences (western 0.26 > eastern 0.20 > central 0.18) provide empirical support for designing intervention programs adapted to local conditions and individual differences. These findings suggest that peer support interventions cannot adopt a “one-size-fits-all” approach but must be tailored to the specific needs and cultural contexts of different student populations.
Several limitations should be acknowledged. First, despite our use of time-lagged analysis and advanced statistical models, the observational nature of the study precludes definitive causal conclusions. Experimental studies manipulating peer support levels would provide stronger causal evidence. Second, all measures relied on self-report, which may introduce social desirability bias, particularly for mental health outcomes. Future research could incorporate behavioral observations, peer ratings, or objective indicators of adjustment. Third, while our sample was large and geographically diverse within China, the findings may not generalize to other cultural contexts where peer relationships and support systems operate differently. Fourth, we assessed peer support quantity but not quality dimensions such as reciprocity, intimacy, or support adequacy, which may moderate the observed relationships. Fifth, the study did not examine potential negative aspects of peer relationships, such as co-rumination or social comparison, which could attenuate support benefits.
Future research could consider using ecological momentary assessment methods to capture dynamic processes of peer support and explore unique mechanisms of online peer support in the digital age. Additionally, investigating the optimal balance between peer and professional support, examining cultural variations in peer support mechanisms, and developing standardized peer support quality measures would advance the field. Research should also explore potential boundary conditions and individual differences that may influence peer support effectiveness, such as attachment styles, cultural values, and prior mental health history.
The practical implications of this research extend beyond individual universities to inform national mental health policy in higher education. The demonstrated effectiveness and cost-efficiency of peer support suggest it should be recognized as an essential component of comprehensive campus mental health services. Policy recommendations include: (1) establishing national standards for peer support program training and implementation; (2) allocating dedicated funding for peer support initiatives, particularly in resource-limited regions; (3) integrating peer support training into general education curricula to build campus-wide supportive capacity; and (4) developing evaluation frameworks that capture the full chain mediation pathway rather than focusing solely on mental health outcomes.
This research provides a robust evidence base for the effectiveness of peer support in promoting college students’ mental health through clearly identified mechanisms. The findings offer actionable insights for developing more effective, efficient, and culturally responsive mental health interventions in higher education settings, ultimately contributing to the wellbeing and success of college students navigating this critical developmental period.
Supplementary Information
Acknowledgments
Figure Declarations
All figures presented in this manuscript (Figs. 1, 2, 3, 4, 5, 6, 7, 8 and 9) are entirely original and were created specifically for this study. Figures 1 and 2 were created using Adobe Illustrator CC 2023. Figures 3, 4, 5, 6, 7, 8 and 9 were generated using R version 4.2.1 with ggplot2 package and Mplus version 8.6 diagram output, with subsequent editing in Adobe Illustrator for visual clarity. No figures have been reproduced or adapted from previous publications. All path diagrams represent the actual models estimated in this study.
Author’s contributions
Qing Zhang: Conceptualization, Methodology, Formal analysis, Supervision, Writing—original draft, Writing—review and editing.
Funding
This work was supported by Wuhan College University-level Research and Innovation Platform Project (Project Number: KYP202201).
Data availability
De-identified data and analysis code supporting the findings of this study are available from the corresponding author upon reasonable request. Due to ethical restrictions and participant privacy protection requirements, raw data containing potentially identifying information cannot be made publicly available. Researchers interested in accessing the data should contact the corresponding author (email: 281339508@163.com) with a methodologically sound research proposal. The proposal should include: (1) specific research questions and hypotheses; (2) analysis plan; (3) data security measures; and (4) institutional ethics approval. A data sharing agreement will be required before data access is granted.
Declarations
Ethics approval and consent to participate
This study was approved by the Research Ethics Committee of Wuhan College (Approval No.: WHC-2022-078, dated March 15, 2022). All procedures were conducted in accordance with the ethical standards of the Declaration of Helsinki (2013 revision) and relevant Chinese regulations for research involving human participants. Written informed consent was obtained from all participants aged 18 and above. For participants under 18 years (n = 226, 12.3% of sample), both written parental consent and participant assent were obtained. All consent documents explicitly described the study purposes, procedures, potential risks and benefits, voluntary nature of participation, confidentiality measures, and the right to withdraw at any time without penalty. Participants were assured that their decision to participate or withdraw would not affect their academic standing or access to university services. Consent records are securely stored in locked filing cabinets and will be retained for 7 years per institutional data retention policies. The study protocol included provisions for referring participants showing signs of significant distress to university counseling services, with established referral pathways and follow-up procedures.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.Pointon-Haas J, Waqar L, Upsher R, Foster J, Byrom N, Oates J. A systematic review of peer support interventions for student mental health and well-being in higher education. BJPsych Open. 2023;10(1):1–16. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2.Min Z, Jiu Z, Xiao Z, et al. Mental health changes among college students and its relation with perceived social support. Chin J School Health. 2018;39(2):232–5. [Google Scholar]
- 3.Grégoire S, Beaulieu F, Lachance L, Bouffard T, Vezeau C, Perreault M. An online peer support program to improve mental health among university students: a randomized controlled trial. J Am Coll Health. 2022;72(7):1–13. [DOI] [PubMed] [Google Scholar]
- 4.Richard J, Rebinsky R, Suresh R, et al. Scoping review to evaluate the effects of peer support on the mental health of young adults. BMJ Open. 2022;12(8):e061336. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Ding Y, Wang X, Zhang F, et al. The relationship between perceived social support, coping style, and the quality of life and psychological state of lung cancer patients[J]. BMC Psychol. 2024;12:439. 10.1186/s40359-024-01927-y [DOI] [PMC free article] [PubMed]
- 6.Simmons MB, Cartner S, Macdonald R, et al. The effectiveness of peer support from a person with lived experience of mental health challenges for young people with anxiety and depression: a systematic review[J]. BMC Psychiatry. 2023;23(1):194. 10.1186/s12888-023-04578-2. [DOI] [PMC free article] [PubMed]
- 7.Cohen S, Wills TA. Stress, social support, and the buffering hypothesis. Psychol Bull. 1985;98(2):310–57. [PubMed] [Google Scholar]
- 8.Bandura A. Social foundations of thought and action: A social cognitive theory. National Inst of Mental Health; 1986. [Google Scholar]
- 9.Arnett JJ. Emerging adulthood: a theory of development from the late teens through the twenties. Am Psychol. 2000;55(5):469–80. [PubMed] [Google Scholar]
- 10.Yeo G, Gladys L, Oon M, Pang R, Ho D. A randomized controlled trial for a digital peer support platform to translate online peer support for emerging adult mental well-being (Preprint). JMIR Ment Health. 2023;10:e43956. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Heinze J, Lipson SK, Eisenberg D, et al. College students’ mental health improving, more finding support: 2023–2024 Healthy Minds Study Data Report. University of Michigan; 2024. [Google Scholar]
- 12.Dominguez MG, Brown LD. Exploring pathways to recovery and psychological well-being: examining the role of empathic and social self-efficacy, social support and social isolation. Front Psychol. 2025;16:1552827. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Simmons MB, Cartner S, MacDonald R, Whitson S, Bailey A, Brown E. The effectiveness of peer support from a person with lived experience of mental health challenges for young people with anxiety and depression: a systematic review. BMC Psychiatry. 2023;23(1):194. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Abrams MP, Salzman J, Espina Rey A, Daly K. Impact of providing peer support on medical students’ empathy, self-efficacy, and mental health stigma. Int J Environ Res Public Health. 2022;19(9):5135. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Feng L, Zhang L. Perceived teacher support, peer relationship, and university students’ mental health: the mediation of reality and internet altruistic behaviors. Front Psychol. 2022;13:1000069. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Worley JT, Meter DJ, Hall AR, Nishina A, Medina MA. Prospective associations between peer support, academic competence, and anxiety in college students. Soc Psychol Educ. 2023;26:1017–35. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Weber M, Schulze L, Bolzenkötter T, Niemeyer H, Renneberg B. Mental health and loneliness in university students during the COVID-19 pandemic in Germany: a longitudinal study. Front Psychiatry. 2022. 10.3389/fpsyt.2022.848645. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Rosset M, Baumann E, Altenmüller E. A longitudinal study of physical and mental health and health-related attitudes among music students: potentials and challenges for university health promotion programs. Front Psychol. 2022;13:885739. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Li X, Wu R, Wu MY, Zhu G. Changes and predictors of mental health of Chinese university students after the COVID-19 pandemic: a two-year study. J Affect Disord. 2024;354:216–23. [DOI] [PubMed] [Google Scholar]
- 20.Ning X, Luo X, Guo S. Researching into Chinese university students’ mental health in the post-pandemic era – problems and causes. Front Psychol. 2024;15:1393603. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Han SS, Zhang YS, Zhu W, et al. Status and epidemiological characteristics of depression and anxiety among Chinese university students in 2023. BMC Public Health. 2025;25(1):442. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Fan J, Huang Y, Yang F, Cheng Y, Yu J. Psychological health status of Chinese university students: based on psychological resilience dynamic system model. Front Public Health. 2024;12:1382217. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Gao T, Gao B, Wang L, Qing Z. Integrating mental health literacy into Chinese college student mental health education in the post-COVID-19 era. Front Public Health. 2024;12:1456579. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Mental Health America. Peer support: Research and reports. Mental Health America. 2023. [Google Scholar]
- 25.Shan Y, Ji M, Xie W, et al. Interventions in Chinese undergraduate students’ mental health: systematic review. Interact J Med Res. 2022;11(1):e38249. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Zhang J, Hu Z, Zheng S, Liu Y. A chain mediation model reveals the association between mindfulness and depression of college students. Sci Rep. 2023;13(1):16984. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Ni J, Liu A, Shi Y, Guo J. Impact of academic involution atmosphere on college students’ mental exhaustion: a chain mediation model. Int J Educ Res. 2024;128:102475. [Google Scholar]
- 28.Lo CC, Chen I-C, Ho WS, Cheng YC. A sequential mediation model of perceived social support, mindfulness, perceived hope, and mental health literacy: an empirical study on Taiwanese university students. Acta Psychol. 2023;240:104016. [DOI] [PubMed] [Google Scholar]
- 29.Wang C, Chudzicka-Czupała A, Tee M, et al. A chain mediation model on COVID-19 symptoms and mental health outcomes in Americans, Asians and Europeans. Sci Rep. 2021;11:6481. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Sui X, Zhao B, Na D, Liu J, Zhang Q. The relationship between physical exercise and sense of social fairness among college students: the chain-mediated role of perceived social support and life satisfaction. Front Psychol. 2024;15:1453675. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Koo K, Baker I, Yoon J. The first year acculturation: a longitudinal study on acculturative stress and adjustment among the first year international college students. J Int Students. 2021;11(2):278–98. [Google Scholar]
- 32.Cobo-Rendón R, López-Angulo Y, Pérez-Villalobos MV, Díaz-Mujica A. Perceived social support and its effects on changes in the affective and eudaimonic well-being of Chilean university students. Front Psychol. 2020;11:590513. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Osborn TG, Town R, Ellis R, Buckman JEJ, Saunders R, Fonagy P. Implementing peer support in higher education: a feasibility study. SSM. 2022;2:100175. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Masten AS, Cicchetti D. Developmental cascades [Editorial]. Dev Psychopathol. 2010;22(3):491–5. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
De-identified data and analysis code supporting the findings of this study are available from the corresponding author upon reasonable request. Due to ethical restrictions and participant privacy protection requirements, raw data containing potentially identifying information cannot be made publicly available. Researchers interested in accessing the data should contact the corresponding author (email: 281339508@163.com) with a methodologically sound research proposal. The proposal should include: (1) specific research questions and hypotheses; (2) analysis plan; (3) data security measures; and (4) institutional ethics approval. A data sharing agreement will be required before data access is granted.