Network based analysis of student self governance networks and predictive role in civic participation outcomes

Abstract

This study examines the relationship between student Self-Governance Networks (SGN) and Civic Participation Outcomes (CPO) using a network-based analytical framework. Data were collected from 237 student governance participants at a regional university in eastern China and analyzed using Social Network Analysis (SNA), Multivariate Regression (MR), and Machine Learning (ML) classification. The SGN displayed a modular network comprising 6 functional communities, including central administration, academic councils, disciplinary bodies, and interest-based organizations. Structural analysis revealed a hierarchical hub-and-spoke configuration, with the central student government serving as the primary bridge across communities, while peripheral groups remained comparatively isolated. Centrality measures varied systematically across roles, with formal leadership positions occupying structurally advantaged positions. Regression analyses controlling for demographics confirmed that network position significantly predicted civic engagement, with eigenvector centrality and the Community E–I Index emerging as consistent predictors across results. ML models, particularly XGBoost, achieved strong predictive accuracy (accuracy = 0.781, AUC-ROC = 0.842), indicating that network features reliably differentiate engaged from non-engaged students. These findings indicate that the civic benefits of governance participation are not uniformly distributed but depend on an individual’s structural position, with ties to central actors and cross-community linkages enhancing involvement. The study outlines theoretical contributions and practical implications for governance design and civic education.

Introduction

The connection between participation in student Self-Governance Networks (SGN) and Civic Participation Outcomes (CPO) remains a vital area of inquiry in Higher Education (HE), bearing essential implications for preparing students for democratic citizenship, enhancing institutional governance, and supporting civic renewal^1,2. Involvement in campus governance proposals is a formative experience that enables students to engage in collective decision-making, develop deliberative competencies, and cultivate participatory habits with potential relevance beyond the academic setting^3,4. Although prior studies have consistently shown that student governance participation is positively associated with CPO^5,6, comparatively limited research has examined how different structural roles within SGN impact the nature and magnitude of these benefits.

Student SGNs are typically structured as intricate organizational networks, with participants distributed across diverse structural positions that reflect varying degrees of connectivity, impact, and boundary-crossing capacity^7,8. These networks serve as conduits for data exchange, decision-making coordination, and collaborative policy development, shaping institutional functioning through dynamic processes^9,10. From a network-theoretical perspective, an individual’s location within these structures directly shapes their access to institutional data, social capital, and opportunities for meaningful participation^11,12. As a result, the civic gains derived from governance involvement will likely differ systematically depending on one’s structural positioning, rather than stemming solely from general participation.

Recent applications of Social Network Analysis (SNA) within educational research have highlighted its value in uncovering how relational dynamics impact student development^13,14. Despite its promise, this analytical lens has been infrequently applied to formal student SGN, especially regarding CPO. Existing studies exploring the structural dimensions of student SGN have tended to focus on official roles and hierarchical titles rather than on network-based indicators that capture the relational significance and data flows within SGN¹⁵. This methodological gap has limited insight into how governance networks, beyond formal roles, contribute to civic learning and student engagement.

The civic significance of governance involvement has become increasingly pronounced amid increasing problems with student disengagement and political apathy in HE¹⁶. While many institutions now promote participation in governance as a high-impact practice for civic formation¹⁷, such initiatives frequently lack empirical evidence on which network features best support CPO. Investigating the relationship between SGN features and CPO could provide actionable insights, enabling the more deliberate design of SGN that fosters meaningful CPO.

Recent methodological developments in predictive analytics propose new opportunities to explore how network positions relate to CPO using advanced analytical approaches. Machine Learning (ML) methods enable the detection of complex patterns and interaction effects that frequently elude conventional statistical models, thereby revealing subtler associations between SGN features and CPO^18,19. While such methods have been effectively employed in educational backgrounds to predict academic and behavioral results, their application to SGN and CPO remains relatively limited.

This study addresses these gaps by employing a network-oriented analytical model to investigate how structural positions within student SGN relate to CPO.

Specifically, this research seeks to answer three primary questions:

RQ1: What Network Structural Features (NSF) define student SGN, and how do centrality measures, community affiliations, and connectivity patterns vary across governance positions?

RQ2: To what extent do NSF predict CPO beyond Demographic Features (DF), and which specific network metrics most strongly associate with different dimensions of civic engagement?

RQ3: How effectively can an ML model classify CPO using network-derived features, and what does this reveal about the complexity of network-civic relationships?

By combining SNA and ML classification, this research investigates the predictive value of centrality measures, community affiliations, and other network-derived metrics across multiple facets of CPO. These research questions guide a systematic exploration of structural predictors of CPO while maintaining a proper scope to detect unexpected patterns and relationships.

This study advances beyond prior research on student governance and civic participation using 3 novel contributions.

1. First, while previous studies have examined student governance by organizational models focusing on formal roles^48,49 or e-participation mechanisms⁴⁷, this study is the first to apply social network analysis to quantify how structural positions within face-to-face governance networks predict civic results. This reveals positional advantages invisible to conventional role-based analysis, demonstrating that students with identical formal titles can occupy vastly different structural positions with correspondingly different civic trajectories.

2. Second, prior research has treated civic participation as unidimensional or examined single outcomes in isolation. This study validates that distinct network positions predict different civic outcomes: eigenvector centrality uniquely predicts civic attitudes (β = 0.271, p = .001) and leadership initiative (OR = 1.57), while betweenness centrality specifically predicts policy advocacy (OR = 1.47), and clustering negatively affects outward engagement (OR = 0.71). This differentiated pattern has not been established in prior governance research.

3. Third, while computational approaches have predicted civic engagement from DF⁵⁰, this study represents the first application of machine learning to network structural positions in student governance, achieving robust classification performance (AUC = 0.842) that proves network topology provides predictive power comparable to traditional models. The integration of SNA, multivariate regression, and ML operationalizes theoretical constructs of positional advantage into measurable, predictive tools that have not been verified previously.

Together, these contributions transform the understanding of governance participation from a binary phenomenon into a structurally differentiated process in which position determines civic benefits, advancing theoretical understanding in Chinese HE contexts and providing empirical foundations for evidence-based governance design.

The paper is organized as follows. First, relevant literature on student SGN, CPO, and SNA applications in educational contexts is reviewed in section “Literature review”. Next, the methodology, including research design, data collection procedures, network construction, and analytical techniques, is presented in section “Methodology”. Results are then presented in section “Results”, Discussion in section “Discussion”, and conclusions in section “Conclusion and future works”.

Literature review

The role of student SGN and adult-led initiatives in designing CPO has become an increasingly prominent focus of academic discourse. Studies conducted across diverse educational and cultural settings have examined how civic types emerge within institutional environments, frequently emphasizing the interplay among participatory skills, organizational support, and broader sociocultural factors.

A recent scoping review by ²⁰ investigated the impact of the COVID-19 pandemic on CPO among adults. Drawing on studies from multiple regions, the review identified shifting patterns in civic behavior, including a rise in digital activism, evolving models of volunteerism, and the growing significance of virtual communities. Notably, the findings pointed to a paradox: while conventional avenues for CPO declined, new forms emerged, particularly among adults who benefited from structured institutional support. This transformation in civic practice, driven by external disruptions, underscores the importance of adaptable institutional practices that integrate in-person and digital modes of participation.

From a regional perspective, Schoon et al. ²¹ conducted a comprehensive quantitative investigation of CPO skills among university students in Qatar. Their analysis uncovered statistically significant gender-based differences in modes of participation: female students were more frequently involved in community service and public awareness initiatives, while male students were more frequently involved in electoral activities and financial support. These patterns were interpreted in light of prevailing cultural norms and differences in access, leading the authors to advocate for CPO that are sensitive to cultural context and designed to promote inclusivity and equity.

The institutional dimensions of CPO were further examined by Magill and Davis ²², who explored how place-based curricula and community partnerships can foster transformative civic learning. Drawing on qualitative case studies, they argued that student involvement in addressing real-world community challenges—when embedded in well-structured pedagogical models—supports the development of civic identity, critical awareness, and long-term participatory habits. Their perspective is echoed in ²³, who investigated participatory teaching strategies in Zambian secondary schools. Her findings highlighted that learner-centered, collaborative methods enhanced CPO and empowered students to enact democratic values within their communities.

Student autonomy as a catalyst for CPO was central to ²⁴, who synthesized findings from diverse tertiary education contexts to examine student-led initiatives. The study identified 3 key factors shaping their effectiveness: institutional trust, leadership size, and social application. Notably, the analysis also described attention to structural blocks—such as bureaucratic oversight and uneven resource distribution—that frequently hinder the civic potential of student organizations, even when motivation and commitment are strong.

In the Southeast Asian context, Ahmad et al. ²⁵ conducted a survey-based investigation among university students in Malaysia, focusing on factors that predict CPO. Their results highlighted the primary impact of larger social factors, including cultural traditions, family expectations, and perceptions of governmental awareness, on shaping students’ CPO. This regional insight aligns with international comparative findings, notably those from the ICCS 2022 report by Owen and Irion-Groth ²⁶, which documented cross-national differences in CPO, institutional trust, and predicted political participation. The ICCS data further verified that students’ confidence in their civic agency and perceptions of institutional openness strongly predict their future involvement paths.

Program-level evaluations have also provided essential insights into the design and impact of effective CPO²⁷. evaluated the “Project Citizen” curriculum, which integrates classroom instruction with the development of public policy proposals. Their results verified notable improvements in students’ CPO, procedural comprehension, and confidence in engaging with participatory governance processes. In a related context, Schulz et al. ²⁸ examined the civic role of academic libraries in South Africa, emphasizing their function as community anchors that support civic literacy, facilitate lifelong learning, and foster outreach efforts, thereby broadening the institutional landscape of CPO beyond formal classroom settings.

Turning to psychosocial dimensions, Fong and To ²⁹ investigated the associations between CPO, social support, and perceived life purpose among adolescents in Hong Kong. Their study found a strong positive relationship between CPO and a sense of meaningfulness, with social support networks as a significant mediating factor. These findings support the view that CPO operates not only as an expression of behavioral participation but also as a developmental resource contributing to psychological well-being and identity formation.

Student governance networks and civic development

While the broader civic engagement literature has established that institutional participation fosters civic development, research specifically examining student SGN remains comparatively limited. Recent scholarship has begun to address this gap by investigating how student leadership structures and governance participation shape civic competencies. Patrick ⁴⁸ provided a comprehensive examination of student government and leadership in higher education, documenting how formal governance networks enable students to develop democratic skills by experiential participation in institutional decision-making. Building on this foundation, Rejubi et al. ⁴⁹ found that student participation spans multiple governance networks, including academic councils, policy committees, and co-curricular organizations, each providing distinct opportunities for civic skill development.

The Chinese HE context presents unique considerations for understanding student governance and civic participation. Li and Zhao ⁴⁷ examined democratic involvement by e-participation mechanisms in Chinese HE universities, revealing how digital platforms enable student engagement with governance methods within institutional and regulatory constraints. Their work highlighted the tension between hierarchical institutional structures rooted in Confucian educational traditions and emerging participatory governance models. However, while these studies establish that student governance participation matters for civic development, they have not employed network-based analytical models to examine how structural positions within governance networks shape civic outcomes differentially.

Network theory, social capital, and civic participation

Theoretical models from social capital and network science provide essential basics for understanding how relational structures shape civic development. Putnam’s seminal work ³⁵ established that social capital—defined as networks of trust, reciprocity, and civic norms—serves as a vital resource that enables collective action and democratic participation. This model distinguishes between bonding capital (strong ties within homogeneous groups), bridging capital (weak ties across diverse groups), and linking capital (vertical connections across hierarchical levels), each contributing differently to civic capacity. Recent applications extend this model to higher education contexts, with Dutta and Bhatia ⁴⁶ signifying that social capital embedded in institutional networks significantly shapes student belonging and developmental outcomes.

Network theory provides analytical tools for operationalizing these conceptual models through structural measurement. Borgatti and Halgin ³⁷ articulated how network position determines access to information, resources, and influence within organizational systems, thereby establishing the theoretical basics for relating structure to outcomes. Lin et al. ¹² further developed structural hole theory, signifying that actors positioned to bridge otherwise disconnected network segments gain strategic advantages through brokerage opportunities and exposure to diverse data flows. In governance contexts, Bodin ¹⁰ demonstrated that network structure shapes collaborative capacity and collective decision-making, with implications that extend beyond environmental governance to institutional settings more broadly.

Despite these theoretical advances, empirical application of network analysis to student governance networks remains underdeveloped. Existing research has typically examined governance participation as a binary variable—students either participate or do not—without investigating how different structural positions within governance networks create systematically different civic learning opportunities. This analytical gap is particularly pronounced in the Chinese HE context, where formal governance networks interact with cultural norms regarding hierarchy, collectivism, and authority in ways that may basically shape how network position translates into civic development.

Methodological gaps in measuring CPO

The civic participation literature has also been constrained by outcome measurement approaches that may obscure meaningful variation in how governance participation shapes development. Many studies employ unidimensional measures of civic engagement, conflating attitudes, behaviors, and specific forms of participation into single composite scores. This aggregation may mask important distinctions between developing positive civic orientations, assuming leadership roles, engaging in policy advocacy, and participating in democratic discourse—outcomes that may respond differently to governance experiences.

Furthermore, while regression-based analyses have established general associations between governance participation and civic outcomes, these methods may inadequately capture the complex, nonlinear relationships between network position and civic development. Recent computational advances have proved ML’s capacity to model intricate patterns in social phenomena.

Tsoni et al. ⁵⁰ applied ML to predict civic engagement from behavioral and DF data, achieving meaningful classification performance that revealed complex interaction effects undetectable by traditional statistical methods. However, NSF has not been leveraged as a predictive variable in civic engagement classification, thereby representing an unexploited opportunity to advance theoretical understanding and the practical application of governance network impacts.

Synthesis of research gaps and study rationale

This literature review reveals four interconnected gaps that motivate the current investigation.

1. First, while student governance has been examined through organizational and developmental lenses ^48,49, the application of social network analysis to quantify how structural positions within governance systems predict civic outcomes remains unexplored.

2. Second, although network theory and social capital networks develop that relational structure matters for collective action and civic capacity ^12,35,37, empirical validation within student governance contexts—particularly in Chinese HE—has not been demonstrated.

3. Third, existing research has not differentiated how specific network positions (e.g., central hubs, boundary spanners, clustered actors) may predict distinct civic outcomes (attitudes vs. behaviors; leadership vs. advocacy), limiting understanding of the mechanisms by which governance participation shapes development.

4. Fourth, predictive modelling approaches have not been applied to determine whether NSF enables practical classification of civic engagement, thereby validating theoretical propositions about positional advantage and informing institutional design.

The present study addresses these gaps by integrating social network analysis, multivariate regression, and ML classification to examine how structural positions within student SGN predict multiple dimensions of CPO. By conducting this investigation within a Chinese regional university context, the research contributes empirical evidence on governance-civic relationships in a setting where hierarchical institutional networks, collectivist cultural norms, and emerging participatory governance models intersect in theoretically and practically significant ways. This multi-method analytical method enables not only the identification of network-civic associations but also the assessment of their predictive strength and practical utility for institutional governance networks.

Methodology

Research design

This study adopted a quantitative, network-based research design to investigate the NSF of student SGN and their predictive relationship with CPO. Grounded in the methodological principles of SNA, the design was implemented using empirical data drawn from the full spectrum of student SGN at a university. A cross-sectional method was used to represent governance dynamics as a graph-based model, with nodes corresponding to student representatives, committees, and organizational units, and edges capturing relational ties defined by collaboration, shared membership, or hierarchical interaction.

The analytical model was developed to encompass the SGN’s structural topology and CPO’s behavioral proportions. Key network features, including degree centrality, betweenness centrality, eigenvector centrality, clustering coefficients, and modularity, were systematically computed to illustrate positional impact and interconnectivity within the SGN. These structural indicators served as explanatory variables in predictive models assessing their association with specific CPOs, such as the frequency of event participation, assumption of leadership roles, involvement in community service activities, and contributions to the CPO.

The research design incorporated a demographic layer to account for probable confounding variables, including academic major, year of study, and prior CPO. The study found that NSF had significant explanatory power for predicting civic behavior when network metrics were integrated into regression and classification models. The methodological context prioritizes replicability and scalability, providing a systematic method for examining how the organization of SGN is linked to CPO within institutional settings.

Figure 1 presents the overarching research design, outlining the sequential process from data collection to SGN and SNA, including demographic controls, measurement of result variables, and execution of the analytical model, which leads to key research findings. This structured method enabled a rigorous examination of the associations between individuals’ network positions within student SGN and their subsequent CPO, while accounting for relevant background variables.

Figure 1.

Quantitative network-based research design.

Study context and participants

The study was conducted at a regional university in eastern China, renowned for its sizable and demographically diverse undergraduate student body and its well-established model for student SGN. This institution was selected as the research site due to its high cost and formalized SGN, which includes a wide array of registered student organizations, a centralized student SGN, and multiple college-level councils. The university’s institutional model provided a strong basis for investigating SGN, as it features a structured system of student representation spanning all academic departments and administrative units.

The study population consisted of all students (N = 842) who held formal positions within recognized student SGN bodies during the 2022–2023 academic year. This included elected representatives to the university-wide student SGN, appointed members of advisory committees, executive officers of college councils, and student representatives on university committees. A stratified random sampling method was employed to ensure proportional representation across governance levels, with oversampling of specific underrepresented demographic categories to enhance analytical validity.

The final participant sample consisted of 237 students (response rate: 28.1%), exceeding the minimum required sample size of 128, as determined by G*Power analysis for detecting medium effect sizes (f² = 0.15) with 80% power at α = 0.05. The sample included representation from all major governance bodies as detailed in Table 1, which summarizes participants’ DF. Statistical power analysis confirmed that this sample was sufficient for regression models incorporating up to 8 predictor variables while maintaining an acceptable level of statistical power.

Table 1.

DF of student SGN participants.

Features	Type	Percentage (%)	Number (n)
Governance body	Central Student Government	28.7	68
	College Councils	30.4	72
	Residential Governance	18.6	44
	Special Interest/Advocacy	22.3	53
Academic discipline	Engineering	22.4	53
	Liberal Arts	19.8	47
	Business	17.7	42
	Natural Sciences	16.0	38
	Education	13.5	32
	Other	10.6	25
Year of study	First-year	14.8	35
	Sophomore	22.4	53
	Junior	27.8	66
	Senior	27.0	64
	Graduate	8.0	19
Gender	Female	54.0	128
	Male	46.0	109

*Total sample N = 237. Response rate = 28.1% of eligible governance position holders.

As shown in Table 1, the DF of the sample reflected diversity across academic disciplines, with representation from Engineering (22.4%), Liberal Arts (19.8%), Business (17.7%), Natural Sciences (16.0%), Education (13.5%), and other disciplines (10.6%). The class-year distribution included first-year students (14.8%), sophomores (22.4%), juniors (27.8%), seniors (27.0%), and graduate students (8.0%). Gender distribution was 54.0% Female and 46.0% Male. This composition approximated the university’s overall demographics, with a slight overrepresentation of upper-division students, which was expected given the greater prevalence of governance positions among experienced students.

Inclusion criteria required participants to: (1) hold a formal position within a recognized student SGN for a minimum of one semester, (2) be enrolled as degree-seeking students, and (3) attend at least 70% of scheduled meetings for their respective governance bodies. These criteria ensured participants maintained sufficient involvement in governance tasks to generate meaningful network connections. Exclusion criteria eliminated temporary appointments and positions with limited governance responsibility.

The study adhered to network limit specification protocols by using a positional approach to define network membership based on formal roles within SGN rather than self-reported connections. This method minimized boundary specification errors commonly encountered in network studies of organizational networks.

Data collection

The data collection process occurred during the fall and spring semesters of the 2022–2023 academic year and consisted of three primary components: SGN data, demographic data, and CPO.

• (i)Network data collection: SGN data were collected by HE data analysis and structured reviews. Following HE approval, official university records documenting committee memberships, organizational hierarchies, and formal reporting structures were attained from the Office of Student Affairs. The network data collection instrument incorporated sociometric survey techniques to capture relational data. Participants identified their governance networks using a roster-based method, indicating their relationships with other governance entities from a list of all committees, councils, and leadership positions within the HE system.

Additional relational data were collected using attendance records for joint meetings, co-sponsorship records of campus initiatives, and documentation of collaborative developments between governance entities.

• (ii)Demographic data: Demographic data were collected by a standardized questionnaire administered alongside the network review. The tool captured data on academic discipline, year of study, gender, length of organizational involvement, prior leadership skills, previous CPO, academic performance, residential status, degree program, and international/domestic student status.
• (iii)CPO metrics: CPO data were collected from multiple sources. The primary tool was the CPO Scale, a validated 14-item measure assessing attitudes and behaviors related to CPO. This standardized tool was enhanced with data from HE students documenting their involvement in campus elections, civic HE events, community service initiatives, leadership roles, CPO, policy advocacy, governance proposal submissions, and the democratic method.

All participation metrics were verified by HE data where available. The CPO data collection occurred after the spring semester.

Data collection methods

Data collection adhered to procedural protocols. The survey tools were distributed electronically via the HE’s secure research platform, which assigned unique identifiers to participants to help with accurate network mapping while maintaining confidentiality. Non-respondents received up to three follow-up communications at one-week intervals. All data collection measures received approval from the HE’s Institutional Review Board. Informed consent was obtained from all participants, with explicit permission to access their institutional records related to organizational membership and CPO.

Table 2 presents a comprehensive list of variables collected in this study, along with their measurement types and data sources.

Table 2.

Variables collected in the study.

Type	Variable name	Variable type	Measurement method	Data source
Network variables	Position title	Categorical	Direct encoding	Institutional records
	Governance level	Ordinal	Direct encoding	Institutional records
	Connection type	Categorical	Self-reported	Network survey
	Interaction frequency	Ordinal	Self-reported	Network survey
	Tie strength	Continuous	5-point Likert scale	Network survey
	Communication channels	Categorical	Self-reported	Network survey
	Data flow direction	Binary	Self-reported	Network survey
Demographic variables	Academic discipline	Categorical	Self-reported	Demographic survey
	Year of study	Ordinal	Self-reported	Demographic survey
	Gender	Categorical	Self-reported	Demographic survey
	Organizational tenure	Continuous	Self-reported (months)	Demographic survey
	Prior leadership experience	Count	Self-reported	Demographic survey
	Previous CPO	Categorical	Self-reported	Demographic survey
	Academic performance	Categorical	Self-reported	Demographic survey
	Residential status	Binary	Self-reported	Demographic survey
Civic participation variables	Civic attitudes	Continuous	CIS mean score	CIS instrument
	Civic behaviors	Continuous	CIS mean score	CS instrument
	Electoral participation	Count	Verified participation	Institutional records
	Civic event attendance	Count	Verified attendance	Institutional records
	Community service hours	Continuous	Verified service	Institutional records
	Leadership initiative	Binary	Verified leadership role	Institutional records
	Civic discourse involvement	Count	Verified participation	Institutional records
	Policy advocacy	Binary	Verified involvement	Institutional records
	Governance proposals	Count	Verified submissions	Institutional records
	Democratic bnowledge	Continuous	Assessment score	Knowledge instrument

*CIS = Civic Involvement Scale³³.

Network construction

Governance relationship definition and operationalization

Student SGN encompasses multiple forms of HE relationships that collectively enable civic learning and democratic participation ³⁵.

To capture the comprehensive nature of governance participation, governance relationships were operationalized into three theoretically coherent types:

1. Co-membership relationships were established when individuals served simultaneously on formal committees, councils, or registered student SGN during the 2022–2023 academic year. This included voting and advisory positions verified by HE data from the Office of Student Affairs.

2. Collaborative relationships were defined as active joint participation in governance activities, including co-sponsorship of campus initiatives, joint authorship of policy proposals, or shared leadership of governance plans. These relationships were documented through meeting minutes, event records, and HE databases tracking collaborative activities.

3. Hierarchical relationships were captured by formal reporting structures and advisory connections within governance bodies, such as committee chairs overseeing members or student representatives reporting to council executives. These vertical connections were mapped using official organizational charts and position descriptions.

The integration of these relationship types is theoretically grounded in governance participation theory, which posits that civic learning occurs through multiple pathways of institutional engagement ³⁶. Formal memberships create network opportunities for impact, while collaborative tasks enable skill development through practice, and hierarchical relationships support mentorship and the transfer of HE data. Rather than representing distinct phenomena, these relationship types constitute complementary dimensions of governance participation that jointly contribute to CPO.

Network representation and mathematical construction

The student SGN was represented as an undirected graph G = (V, E), where V denotes the set of nodes corresponding to individual participants, and E ⊆ V×V denotes the set of edges indicating a governance-related tie between two individuals. A binary edge was developed between the nodes and node suppose the corresponding individuals , (where ) shared at least one formal governance relationship, including co-membership in a committee, joint participation in a council, or co-leadership of a registered student SGN during the academic year. This structure is illustrated in Figure 2(a), which depicts the student SGN, color-coded by governance type and distinguished by tie strength (solid vs. dashed lines).

Figure 2.

Network representation and metrics in student SGN, (a) Sample network visualization by governance type; (b) Adjacency matrix representation; (c) Degree and betweenness centrality with graphical interpretation; (d) Advanced metrics: eigenvector centrality, clustering, and modularity; (e) Community structure based on governance affiliation and cross-cluster connectivity.

The adjacency matrix was constructed such that Eq. (1)

1

where

− the total number of student SGN participants included in the network.

This matrix served as the basis for computing network metrics and as input for extracting NSF. A simplified representation of is shown in Figure 2 (b), emphasizing the binary encoding of connections and the exclusion of diagonal self-loops. This binary encoding approach follows developed precedent in organizational network analysis, where multiple relationship types are integrated to capture HE connectivity patterns³⁷.

Weighted network construction

To account for varying levels of relational intensity, a weighted version of the network was also generated using a tie strength matrix , where captured the interaction frequency between individuals i and j, derived from Likert-scaled survey responses.

The weighted adjacency matrix was defined as Eq. (2)

2

where.

is the reported frequency of interaction, with higher values indicating stronger relational ties.

Network metrics computation

For each node , the following centrality measures were computed as illustrated in Fig. 2 (c), (d):

• Degree centrality: The number of direct connections a node has. In SGN, this represents how many direct governance relationships an individual maintains, indicating their immediate impact or reach, Eq. (3)

3

representing the number of direct SGN connections for individual .

• Betweenness centrality: Measures how frequently a node lies on the shortest path between other nodes. Individuals with high betweenness act as bridges or brokers in SGN, controlling data flow between different parts of the network, Eq. (4)

4

where,

− the sum of shortest paths between nodes and , and is the sum of those paths that pass through .

• Eigenvector centrality: Measures a node’s impact based on the impact of its networks. This recursive metric indicates how connected an individual is to other well-connected individuals, revealing power networks where importance depends on the quantity and quality of connections.

It is defined recursively as Eq. (5)

5

where,

− the largest eigenvalue of the adjacency matrix . This measure captured the extent to which a node was connected to other well-connected nodes.

• Clustering coefficient: Measures how tightly connected a node’s neighbors are to each other. In SGN, this indicates the cohesiveness of an individual’s local network, showing whether their connections form closed groups or cliques.

6

where

−the sum of the triangles using the node , and is the degree of the node . This metric reflected the tendency for local cohesion or clique formation around each actor.

• Modularity (Q): Evaluates how well a network can be divided into distinct communities or clusters³⁸. Higher values indicate a stronger community network, where connections are dense within communities but sparse between them, revealing organizational subgroups in SGN. It was computed for detected community networks using Eq. (7)

7

where

m−the sum of edges, and are the degrees of nodes i and is the community assignment for node i, and is the Kronecker delta function.

Modularity was used to assess the extent to which the SGN was organized into modular subgroups, such as departmental clusters or cross-functional alliances. As illustrated in Fig. 2(e), modularity-based community detection revealed structurally distinct clusters corresponding to governance types, including central student SGNs, college councils, and special interest/residential groups. Notably, cross-community ties (dashed lines) emerged at the intersection of central governance and specialized student SGN, indicating bridge actors with high strategic significance ³⁹.

Analytical techniques

The analysis proceeded through a multi-stage model to quantify the relationship between SGN properties and CPO. This model included three major components: descriptive SNA, inferential statistical modelling, and predictive classification.

Descriptive network analysis

Initial analyses focused on illustrating the global and local properties of the made SGN. Descriptive metrics, including network density, average degree, mean clustering coefficient, and modularity, were computed to assess overall connectivity, subgroup cohesion, and fragmentation. These global indicators provided a structural overview of the SGN and informed the selection of explanatory variables for subsequent modelling.

Variable operationalization

Network-based independent variables were derived from the centrality and community metrics described in section “Network construction”. Each participant was associated with a structural feature vector, Eq. (8)

8

where

− degree centrality, is betweenness centrality. − eigenvector centrality. − the local clustering coefficient. − the modular community membership.

In addition, the tie strength values were aggregated to compute mean interaction intensity for each node, Eqs. (9) and (10)

9

10

This weighted measure was used to differentiate impact arising from frequent versus sparse governance interactions⁴⁰.

Dependent variables were drawn from the CPO dataset (Table 2), including continuous outcomes (e.g., CIS scores, community service hours) and binary indicators (e.g., leadership initiative, policy advocacy involvement).

Before modelling, all continuous variables were standardized using Z-score normalization:

11

where

− the experimental value of variable for participant i, are the mean and standard deviation of variable across the sample.

Regression modelling

MR models were used to evaluate NSF’s predictive contribution to CPO.

For continuous results (e.g., CIS civic attitude scores), the general linear model took the form, Eq. (12)

12

where

− the CPO for the participant. − standardized network metrics. − demographic control variables (e.g., year of study, prior CPO). , − projected coefficients. − the error term.

Variance Inflation Factors (VIFs) were examined to assess multicollinearity, and robust standard errors were used to correct heteroscedasticity.

For binary results (e.g., leadership initiative), Logistic Regression (LR) was employed, Eq. (13)

13

Model fit was assessed using pseudo- ; Akaike Information Criterion (AIC) and probability ratio tests. Marginal impacts were computed to interpret the practical impact of each predictor.

Classification modelling

The classification model integrated two advanced ML models: Gradient Boosting Machines (GBM) and Support Vector Machines (SVM). These models were selected for their ability to handle structured data, nonlinear feature interactions, and high-dimensional inputs derived from network topology and participant-level attributes⁴¹.

• (i)Gradient boosting classifier: Gradient boosting constructs an ensemble of weak learners—typically decision trees—by iteratively minimizing a differentiable loss function via gradient descent in the function space.

Given training data , where is the feature vector and is the binary class label, the boosting model estimates the function, Eq. (14)

14

where

i− the base learner at iteration. m, − the learning rate-adjusted weight.

The model is trained to minimize the following regularized objective:

15

where

− the binary logistic loss function:

16

and is a regularization term penalizing model complexity (e.g., tree depth, number of leaves).

The gradient boosting implementation used in this study was XGBoost, which supports second-order gradient optimization and column subsampling to enhance generalization and reduce overfitting ⁴².

Key hyperparameters included:

• Number of estimators .

• Learning rate .

• Maximum tree depth .

• Regularization coefficients .

• (ii)Support vector machines.

SVMs were applied to construct maximum-margin classifiers in transformed feature spaces. For binary classification, SVM seeks a decision boundary defined by Eq. (17)

17

where

− a nonlinear feature mapping. − the weight vector. − the bias term.

The model is trained by solving the following convex optimization problem, Eq. (18)

18

subject to the constraints, Eq. (19)

19

where

− slack variables allowing soft-margin classification. C− a regularization constant controlling the trade-off between margin maximization and misclassification tolerance.

A Radial Basis Function (RBF) kernel was employed to model nonlinear relationships among features⁴³, Eq. (20)

20

Hyperparameters for the SVM were optimized using grid search:

• Regularization constant .

• RBF kernel parameter .

Model performance was validated using stratified 5-fold cross-validation, and classification outputs were evaluated based on AUC-ROC, F1-score, and precision-recall curves. Comparative performance metrics between GBM and SVM were analyzed to find the dominant predictors of CPO within the SGN ⁴⁴.

Ethical considerations

This study was conducted in full compliance with institutional and national ethical standards for research involving human participants. Before data collection, ethical approval was obtained from the Institutional Review Board (IRB) at Universitas Pendidikan Ganesha, which reviewed the study protocol, data-handling measures, and participant recruitment methods to ensure adherence to the principles of autonomy, confidentiality, and non-maleficence. Informed consent was obtained from all student participants by a standardized electronic process embedded within the survey tool. The consent form explicitly outlined the study’s purpose, the voluntary nature of participation, the types of data to be collected, and participants’ rights to withdraw at any time without penalty. Participants were also informed of their right to decline access to HE data; only data from individuals who provided explicit authorization for HE use were included in the final dataset.

All datasets were anonymized before analysis to protect participant privacy and minimize the risk of re-identification. Unique alphanumeric identifiers were used to link survey responses, institutional records, and network positions. Personally Identifiable Information (PII), including names, email addresses, and student identification numbers, was removed immediately after data linkage and stored separately on an encrypted server accessible only to the principal investigator. Sensitive data elements, such as political views or behavioral indicators of CPO, were handled with enhanced caution. Access to raw data was restricted to the core research team under a data-use agreement. Outputs from the analysis were reported in aggregate form, and no individual responses were presented in a manner that could reveal identity or sensitive attributes.

The network boundary was defined using a positional method based solely on formal governance roles, thereby reducing the risk of including informal social ties without the participants’ consent. Sociometric data were collected using a bounded roster to avoid free recall nominations, ensuring that only recognized governance networks were tested in the analytic model.

Results

Hardware and software configuration

The data processing and analysis computational set-up consisted of a high-performance computing environment featuring an Intel Xeon E5-2698 v4 processor (20 cores, 2.2 GHz), 128 GB of RAM, and an NVIDIA Tesla V100 GPU with 16 GB of memory. SGN and SNA were implemented in Python 3.8 using specialized libraries, including NetworkX 2.6.3 and igraph 0.9.9 for graph operations, Sci-Kit-learn 1.0.2 for ML implementations, and XGBoost 1.5.1 for gradient boosting models. Statistical analysis was performed using R 4.1.2, with key packages including statnet 2019.6 for network modelling, lme4 1.1–27.1 for multilevel regression, and pROC 1.18.0 for classification performance evaluation. Data visualization utilized Gephi 0.9.2 for network representations and ggplot2 3.3.5 for statistical graphics. This computational environment ensured efficient processing of the complex SGN and helped the integration of descriptive SNA with advanced statistical modelling techniques.

Results and analysis

• (i)Network statistics

Addressing RQ1, this analysis examines the overall NSF of the student SGN, including connectivity patterns, centralization tendencies, and community organization. Understanding these global features provides essential context for interpreting how individual positions within the network impact CPO.

The SGN analyzed in this study (N = 237, E = 1,738), as shown in Table 3, exhibits several notable NSF that propose insight into student SGN within HE contexts (Figure 3). The overall network density (0.062) indicates a sparse yet functional connectivity pattern typical of HE systems, in which not all actors are directly linked, yet meaningful channels of interaction exist. The average degree (14.67) indicates that individual actors maintain a substantial number of direct ties, signifying that most student leaders are embedded in multiple organizational domains. This combination of moderate density and significant average degree indicates a balance between manageability and reach: governance participants remain connected enough to circulate data efficiently without being overburdened by excessive relational obligations.

Table 3.

Global structural features of the SGN.

Network property	Value
Nodes	V
Edges	E
Network density	0.062
Average degree	14.67
Degree range	1–47
Network diameter	7
Average path length	2.83
Global clustering coefficient	0.418
Average local clustering coefficient	0.527
Modularity (Q)	0.371
Number of communities	6
Largest community size	63
Smallest community size	17
Degree centralization index	0.139
Betweenness centralization index	0.092
Network transitivity	0.342
Assortative coefficient	0.216
Average edge weight	3.27
Graph connectedness	Connected
Number of structural holes	174

Fig. 3.

SGN flow diagram.

The degree distribution (Range: 1–47) shows predictable heterogeneity in positional importance, with the network maintaining a core-periphery network that distributes impact across multiple governance domains rather than concentrating authority exclusively in central positions. The modest centralization index (0.139) proposes an organizational design that promotes broader representation while maintaining coordination efficiency, challenging traditional hierarchical student SGN models. This distributed impact pattern proposes that effective governance arises from connectivity across multiple domains, rather than relying solely on formal authority.

The network’s compact structure, as evidenced by a diameter of 7 and an average path length of 2.83, enables rapid data flow, with most governance participants reachable from any other actor within 3 connection steps. This network efficiency proves crucial for policy dissemination and collective decision-making, as data and impact can flow seamlessly throughout the entire governance system without extensive intermediation. The clustering coefficients (Global: 0.418; Average Local: 0.527) reveal substantial triadic closure within the SGN, indicating formation of cohesive subgroups that support focused governance functions while maintaining system-wide integration.

The modularity value (Q = 0.371) confirms meaningful community structuration with 6 distinct governance clusters, where the largest community (63 nodes) organizes around central administrative functions while smaller communities (minimum size: 17) represent specialized advocacy or residential governance domains. The transitivity measure (0.342) indicates moderate collaborative triangulation, promoting consensus-building and data validation through multiple pathways, which implies that governance decisions benefit from diverse input sources rather than isolated deliberation.

Particularly significant are the 174 structural holes within the network, representing untapped chances for strategic bridging between otherwise disconnected governance segments. Students positioned to span these gaps could substantially enhance both their personal impact and the system’s coordination capacity. The positive assortative coefficient (0.216) indicates that highly connected actors preferentially connect with other well-connected peers, hypothetically generating impact hierarchies that concentrate CPO among already-advantaged participants.

The moderate betweenness centralization index (0.092) indicates that intermediary control is distributed across multiple positions rather than concentrated among a few gatekeepers, supporting governance resilience by ensuring that data flow does not critically depend on a small number of central actors. The average edge weight (3.27 on a 1–5 scale) indicates that governance ties represent substantive collaborative relationships rather than perfunctory associations, signifying that network connections help meaningful civic engagement rather than nominal participation.

The distribution of network centrality metrics provides crucial insights into how impact and opportunity are graded across the SGN (Figure 4; Table 4).

Figure 4.

Violin plots for Centrality Measure metrics.

Table 4.

Descriptive statistics for node-level network metrics.

Centrality measure	Mean	SD	Median	Min	Max	Skewness	1st quartile	3rd quartile
Degree centrality	14.67	8.93	13.00	1	47	0.89	8.00	19.00
Normalized degree	0.062	0.038	0.055	0.004	0.199	0.89	0.034	0.080
Betweenness centrality	263.41	467.32	107.95	0	3,926.16	4.12	38.78	293.14
Normalized betweenness	0.009	0.017	0.004	0	0.141	4.12	0.001	0.011
Eigenvector centrality	0.195	0.173	0.146	0.003	0.821	1.42	0.059	0.292
Clustering coefficient	0.527	0.219	0.533	0	1.000	-0.23	0.365	0.689
Constraint	0.296	0.164	0.257	0.078	0.972	1.27	0.186	0.371
Effective size	8.84	5.24	7.65	1	32.16	1.19	5.28	11.47
Coreness (k-core)	9.32	4.21	9.00	1	18	0.06	6.00	13.00
Average tie strength	3.27	0.72	3.24	1.57	5.00	0.11	2.75	3.78

• (A)Degree centrality (Mean = 14.67, SD = 8.93) exhibits positive skewness (0.89), indicating that while most participants maintain a moderate number of governance ties (Median = 13.00), a small subset of actors are exceptionally well-connected (Max = 47). These highly connected students likely function as organizational hubs, enjoying uneven access to data and opportunities for civic leadership.
• (B)Betweenness centrality shows the most extreme asymmetry (Skewness = 4.12), with a small number of individuals dominating intermediary roles. The maximum attained value (3,926.16) is far higher than the median (107.95), underscoring that boundary-spanning positions are scarce but authoritative. These actors are positioned to act as agents between communities, shape deliberation across governance clusters, and control the circulation of civic initiatives.
• (C)Eigenvector centrality (Mean = 0.195, SD = 0.173) proves stratification in access to essential peers. While most participants occupy peripheral positions (1st Quartile = 0.059), a minority are embedded in dense, high-impact clusters (Max = 0.821). Such embeddedness amplifies civic learning by exposing actors to governance norms and decision-making practices transmitted through central peers.
• (D)Clustering coefficients (Global Mean = 0.527, Skewness = − 0.23) indicate that most students operate in tightly knit neighborhoods, where local trust and coordination are strong. However, excessive clustering also implies limited exposure to novel perspectives, potentially constraining innovation in policy advocacy or cross-community collaboration.
• (E)Constraints and effective size illustrate the trade-off between redundancy and opportunity. Most actors face moderate structural constraints (Mean = 0.296), but those with lower constraints and larger effective sizes (up to 32.16 non-redundant ties) occupy positions ideal for brokerage. These students can access diverse data streams and leverage structural holes to impact broader governance discourse.
• (F)Coreness values (Mean = 9.32, SD = 4.21) display near symmetry, showing that students are distributed relatively evenly across peripheral and core shells. The highest k-core value (18) identifies a nucleus of densely interconnected actors, representing a governance elite with consistent impact across multiple domains.
• (G)The average tie strength (Mean = 3.27 on a 5-point scale) is symmetrical and moderately high, indicating that governance relationships involve recurring collaboration rather than superficial contacts. This reinforces the substantive nature of SGN ties, which are grounded in shared responsibilities and cooperative tasks.

The centrality analysis highlights how formal positional roles map onto structural impact within the SGN, illustrating which actors function as hubs, brokers, or embedded networks (Figure 5(a-c)).

Figure 5.

Top 10 governance positions for: (a) Degree, (b) Betweenness, and (c) Eigenvector.

1. The Student Union President dominates across all three dimensions (Table 5)—degree, betweenness, and eigenvector centrality—confirming this role as a direct connector (47 ties) and strategic bridge (betweenness = 3,926.16). Its eigenvector score (0.821) indicates that the President is also closely associated with the most important peers. This structural dominance validates the President’s position as the primary integrator of student SGN and underscores its disproportionate impact on civic agenda-setting.

2. Vice-presidential positions act as secondary hubs of impact. The VP Academic Affairs combines broad connectivity (38 ties) with strong bridging capacity (betweenness = 2,301.47), reflecting its role in linking academic councils with central governance. By contrast, the VP Student Affairs achieves its prominence through eigenvector centrality (0.779), indicating dense integration with other vital actors rather than spanning structural holes. These contrasting patterns illustrate that impact can be exercised either by boundary-crossing or elite embeddedness.

3. Council Chairs of Engineering and Business hold structurally significant roles within their respective academic domains. The Engineering Chair ranks higher in degree and betweenness, indicating outward bridging, while the Business Chair achieves higher eigenvector centrality, signalling strong integration into the governance core. This divergence reveals how structurally similar titles can yield different forms of impact: one oriented toward breadth of connectivity, the other toward embedded authority.

4. The University Committee Representative exemplifies a specialized bridging role. Its relatively modest degree (28 ties) contrasts with its very high betweenness (2,784.53), demonstrating that even a limited set of ties can provide critical access points across communities. Such actors are structurally positioned to mediate between central administration and distributed councils, facilitating cross-domain policy dialogue.

5. The Diversity Committee Chair appears only in betweenness rankings (4th ), signalling an exclusive bridging role. Despite lacking a high degree of eigenvector prominence, this position connects otherwise isolated advocacy subgroups to mainstream governance. Its structural footprint reflects the theoretical role of diversity-focused positions as boundary spanners that infuse alternative perspectives into dominant governance channels.

6. Balanced actors, such as the Student Media Director and Residential Council President, show moderate but consistent centrality across all measures. Their profiles recommend a dual function: maintaining relevance within their own domains while retaining meaningful connectivity to central leadership. Such actors may play a stabilizing role, ensuring that specialized governance units (media, residential life) remain aligned with institutional priorities.

Table 5.

Top 10 governance positions by centrality measures.

Rank	Position title	Degree	Position title	Betweenness	Position title	Eigenvector
1	Student Union President	47	Student Union President	3,926.16	Student Union President	0.821
2	VP Academic Affairs	38	University Committee Rep	2,784.53	VP Student Affairs	0.779
3	VP Student Affairs	36	VP Academic Affairs	2,301.47	VP Academic Affairs	0.756
4	Council Chair, Engineering	29	Diversity Committee Chair	1,987.28	Council Chair, Business	0.683
5	University Committee Rep	28	Academic Senate Student Rep	1,705.31	Council Chair, Engineering	0.618
6	Council Chair, Business	27	Council Chair, Engineering	1,523.89	University Committee Rep	0.571
7	Student Media Director	26	Residential Council President	1,215.64	Finance Committee Chair	0.558
8	Academic Senate Student Rep	25	International Student Rep	1,142.96	Student Media Director	0.512
9	Finance Committee Chair	24	Student Media Director	1,076.38	Academic Senate Student Rep	0.489
10	Residential Council President	23	Finance Committee Chair	947.11	Residential Council President	0.478

Tables 3 and 4 reveal a governance network comprising 237 nodes and 1,738 edges, exhibiting a low density (0.062) but a substantial average degree (14.67). Table 4 shows extreme centrality stratification—betweenness skewness (4.12) indicates concentrated bridging power, while eigenvector centrality ranges from 0.003 to 0.821. These metrics answer RQ1 by indicating a distributed yet hierarchical system in which 174 structural holes provide strategic opportunities, but the impact concentrates among select actors with exceptional connectivity and access to governance elites.

• (ii)Community detection and role analysis

Continuing RQ1, we investigate how the governance network divides into functional communities and examine the cross-community connectivity patterns that enable institutional coordination. This analysis reveals which positions serve as critical bridges and how community membership shapes access to diverse governance perspectives.

The community detection analysis reveals a modular SGN comprising 6 distinct functional clusters, each representing specialized governance networks while maintaining essential interconnections that sustain HE coherence (Figure 6 (a-d)). This modular structure shows how governance responsibilities are differentiated across functional areas, yet integrated through bridging actors that help coordination and data exchange.

Figure 6.

(a) Community Size, (b) internal vs. external density, (c) conductance, and (d) E-I index.

• (A)Central Administration and Student Union (C1)

Community C1, the largest cluster (63 members), encompasses the Central Student Union and University Committees. Its moderate internal density (0.234) and the highest external connectivity (0.038) reflect its dual role as an internally cohesive body and the primary hub linking specialized networks. The negative E–I index (–0.572) highlights its inward orientation, yet its conductance value (0.274) proves substantive bridging across boundaries. The Student Union President emerges as the key connector, confirming that central executive roles anchor the SGN and channel cross-community governance activity.

• (B) Academic Governance (C2)

The second-largest cluster (47 members) includes the Academic Senate and faculty liaisons. Its lower internal density (0.193) and higher conductance (0.321) indicate that academic governance is a network designed as an integrative domain rather than an insular one. The Academic Senate Student Representative serves as the primary boundary spanner, ensuring the transfer of data between faculty governance and the broader student network. This highlights how academic governance networks promote civic learning by exposing students to multiple institutional perspectives.

• (C)College Councils (C3 and C4)

The College Councils are divided into two clusters, reflecting disciplinary specialization.

• C3 (Science and Engineering, 42 members) exhibits high internal density (0.285) and low conductance (0.187), signifying strong within-discipline cohesion but limited external engagement. Its highly negative E–I index (–0.693) recommends an inward-looking structure in which disciplinary norms may constrain civic learning opportunities.

• C4 (Arts, Humanities, and Business, 38 members), by contrast, exhibits more balanced internal-external connectivity (conductance = 0.265; E–I = − 0.517), indicating that students in these councils engage in more cross-boundary collaboration. The Council Chairs of Engineering and Business emerge as primary bridging actors, with Engineering emphasizing structural integration within STEM domains and Business achieving stronger integration into the SGN core.

• (D) Residential Governance (C5)

Community C5 (30 members) represents residential governance and student life. With the highest internal density (0.314) and the most negative E–I index (–0.724), this community is characterized by strong cohesion but relative isolation. Its low conductance (0.153) indicates limited boundary-crossing, which reflects how residential governance frequently focuses on localized problems. The Residential Council President serves as the primary liaison to central governance, ensuring that residential concerns are not excluded from broader institutional decision-making.

• (E)Special Interest and Cultural Organizations (C6)

The smallest cluster (17 members) includes special interest and cultural organizations. Despite its size, it maintains a distinct profile, characterized by moderate internal density (0.248), the least negative E–I index (–0.403), and relatively high conductance (0.286). These metrics reveal that advocacy and cultural organizations engage with external actors more than other clusters, positioning them as bridges for diverse perspectives. The Diversity Committee Chair functions as the primary connector, channeling the advocacy groups’ concerns into the wider governance network.

The analysis of cross-community connections reveals a hierarchical pattern of inter-cluster relationships that primarily shapes data flow and coordination within the SGN (Figure 7). The overall network resembles a hub-and-spoke model, in which central governance (C1) governs boundary-spanning activity while peripheral domains remain comparatively isolated.

Figure 7.

Cross-community connections and role distribution.

• (F) Central governance as the dominant hub

Community C1 accounts for 89.3% of all inter-community ties, signifying its role as the dominant connectivity hub. The strongest connection pathway (C1↔C2: 76 edges, 26.4%) links the Student Union and Academic Senate, establishing a robust channel between executive leadership and academic representation. This relationship, primarily facilitated by the Student Union President and Academic Senate Student Representative, demonstrates how central actors function as structural brokers, ensuring that policy priorities and academic problems circulate across the SGN.

• (G)Secondary connectivity pathways

Substantial secondary pathways connect C1 with disciplinary and residential governance:

• C1↔C3 (20.5%) connects central governance to science and engineering councils, primarily through the VP Academic Affairs.

• C1↔C4 (18.1%) links to arts, humanities, and business councils, with coordination led by the VP Student Affairs.

• C1↔C5 (14.2%) integrates residential governance, again helped by vice-presidential roles.

These pathways indicate functional specialization among vice-presidential offices, each of which maintains responsibility for specific governance domains. This specialization provides efficiency but also underscores the dependency of domain-specific councils on central leaders for cross-boundary visibility.

• (H) Advocacy and diversity pathways

The pathway connecting central governance to special interest and cultural organizations (C1 ↔ C6: 10.1%) is minor but symbolically significant. It is helped primarily by diversity leadership roles, confirming that advocacy groups maintain direct access to central governance rather than relying on mediation by academic or residential councils. This structural pattern enhances the institutional visibility of advocacy-based initiatives and provides marginalized groups with a clear route to impact executive decision-making.

• (I) Peripheral isolation

A critical finding is the minimal connectivity among non-central communities themselves, which collectively account for only 10.7% of cross-community ties. The strongest of these (C2↔C3: 8 edges, 2.8%) links academic governance with STEM councils via specialized academic policy representatives. Most other peripheral connections remain negligible (≤ 7 edges, ≤ 2.4%), and some potential pathways (C4 ↔ C6, C5 ↔ C6) are absent. This peripheral isolation proposes that without mediation by central governance, most specialized networks have limited chances to share data directly or collaborate on cross-cutting civic initiatives.

Table 6 identifies six communities with varying sizes (17–63 members) and connectivity patterns, where C1 (central governance) exhibits the highest external density (0.038), while C5 (residential) displays extreme insularity (E-I = -0.724). Table 5 reveals that the Student Union President dominates all centrality measures (degree = 47, betweenness = 3,926.16, eigenvector = 0.821). Figure 7 shows that 89.3% of cross-community ties involve central governance. These findings answer RQ1 by revealing systematic access inequalities, in which boundary-spanning opportunities are concentrated in executive positions.

Table 6.

Detected governance communities and their structural properties.

Community ID	Size	Dominant governance types	Internal density	External density	Conductance	E-I Index	Top bridging position
C1	63	Central Student Union, University Committees	0.234	0.038	0.274	-0.572	Student Union President
C2	47	Academic Senate, Faculty Liaisons	0.193	0.029	0.321	-0.438	Academic Senate Student Rep
C3	42	College Councils (Science, Engineering)	0.285	0.027	0.187	-0.693	Council Chair, Engineering
C4	38	College Councils (Arts, Humanities, Business)	0.226	0.031	0.265	-0.517	Council Chair, Business
C5	30	Residential Governance, Student Life	0.314	0.024	0.153	-0.724	Residential Council President
C6	17	Special Interest, Cultural Organizations	0.248	0.019	0.286	-0.403	Diversity Committee Chair

• (iii)Civic participation descriptives

Before examining network-civic relationships (RQ2), this work develops baseline patterns of civic engagement among governance participants. This descriptive analysis reveals the range and distribution of CPO, highlighting areas where governance involvement appears to create differential benefits. The CPO reveals significant variation in how governance involvement translates into broader civic engagement, with distinct patterns signifying differential benefits across participation domains (Figure 8).

Figure 8.

Statistics for CPO.

The Civic Attitudes vs. Behaviors Gap: Civic attitude scores (Mean = 4.37, SD = 0.89) demonstrate a strongly positive orientation among governance participants, with most students clustering toward the upper end of the scale (Median = 4.52). However, civic behavior scores follow a notably different pattern (Mean = 3.82, SD = 1.03), showing greater variability and a lower central tendency. This attitude-behavior gap indicates that while governance participation consistently fosters positive civic orientations, translating these attitudes into sustained civic actions remains more challenging. The broader behavioral variance proposes that structural factors—such as network position or community membership—may determine which students successfully convert civic attitudes into active participation.

• (A) Community service stratification: Community service hours exhibit extreme variation (SD = 19.86) with pronounced positive skewness (1.34), revealing stark inequality in service commitment among governance participants. While most students contribute moderate service hours (Median = 22.00), a small subset proves special commitment, reaching 106.50 h. This pattern recommends that governance participation develop differential pathways to community engagement, with some positions or network locations supporting extensive service involvement, while others offer limited opportunities for community connection.
• (B) Electoral engagement as a governance norm: Electoral participation shows high consistency (Mean = 2.38, Median = 3.00 out of 4 possible elections), with 75% of governance participants voting in at least two campus elections. This pattern indicates that governance involvement effectively socializes students into democratic participation, establishing voting as a normative expectation within governance networks. The relatively narrow variation suggests that electoral engagement is a vital civic competency developed through governance participation, regardless of one’s specific position or network location.
• (C) Civic event participation differentiation. Civic event attendance (Mean = 5.92, SD = 4.31) proves substantial variation, ranging from 0 to 21 events, indicating that governance positions provide dramatically different exposure to civic programming. This wide range recommends that some governance roles offer rich opportunities for civic learning through diverse programming, while others remain isolated from broader civic education initiatives. The positive skewness indicates that high civic engagement focuses among a subset of governance participants, hypothetically, those in central network positions.

The Elite Governance proposal submission reveals an extreme concentration (Mean = 1.43, SD = 1.82, Median = 1.00), with most students (first quartile = 0.00) submitting no formal proposals, while a small subset proves high policy activity (maximum = 9.00). This pattern proposes that chances for policy innovation are focused among governance elites, particularly those in high-centrality positions with extensive cross-community connections. The report recommends that most governance participants gain civic skills through implementation rather than policy creation.

• (D)Democratic knowledge scores: The Democratic knowledge scores (Mean = 7.63, SD = 2.14) indicate generally high governance understanding, with moderate variation. The high median (8.00) indicates that governance participation effectively transmits institutional knowledge; however, the range (2.00–10.00) suggests that knowledge acquisition varies significantly across positions and network locations.

These participation patterns reveal that governance involvement creates a stratified civic learning environment in which central, well-connected participants gain comprehensive opportunities for civic development, while peripheral participants experience more limited civic skill acquisition. Figure 8 reveals substantial civic result stratification: attitudes (Mean = 4.37) exceed behaviors (Mean = 3.82), community service exhibits extreme variance (SD = 19.86, range 0–106.5 h), and policy proposals are concentrated among elites (75% submit ≤ 1 proposal, with a maximum of 9). These baseline patterns establish the CPO landscape for RQ2, signifying that governance participation produces highly differentiated civic development rather than uniform benefits across participants.

• (iv)Correlation between network metrics and CPO

Addressing RQ2, this analysis quantifies the relationships between NSF and CPO. The correlation patterns reveal which features of network position most strongly predict civic development, highlighting potential mechanisms through which governance networks impact CPO. As shown in Figure 9, network metrics exhibit consistent and significant relations with CPO, confirming that basic position within the SGN meaningfully predicts civic orientations and behaviors.

Figure 9.

Correlation matrix between network centrality measures and CPO.

1. Eigenvector centrality emerges as the most robust predictor of civic attitudes (r = .428, p < .001), outperforming degree (r = .383, p < .001) and betweenness centrality (r = .306, p < .001). This recommends that being connected to vital peers is more important than merely having numerous ties or serving as a bridge. Students embedded in influence-rich neighbourhoods are likely to internalize governance norms and values through repeated exposure to central actors, thereby reinforcing attitudinal development.

2. The same pattern extends to behavioral outcomes. Eigenvector centrality maintains the strongest associations with CES behavior (r = .395, p < .001), electoral participation (r = .332, p < .001), and leadership initiative (r = .356, p < .001). These findings indicate that structural prestige within the network not only fosters positive attitudes but also translates into tangible behaviors. The association with leadership initiative highlights that students linked to central peers are more likely to seek additional leadership roles, amplifying civic engagement beyond their initial positions.

3. Betweenness centrality plays a distinctive role in policy advocacy (r = .314, p < .001), surpassing degree centrality (r = .287, p < .001) and eigenvector centrality (r = .275, p < .001). This pattern implies that boundary-spanning actors gain unique exposure to diverse perspectives and policy requirements, equipping them with the motivation and capacity to engage in advocacy. In practice, this highlights the civic importance of positions that connect otherwise fragmented governance domains.

4. Clustering coefficient exhibits consistently negative correlations with CPO, including CES attitudes (r = − .142, p < .05), electoral participation (r = − .128, p < .05), leadership initiative (r = − .194, p < .01), and policy advocacy (r = − .213, p < .01). These results propose that students embedded in highly clustered neighbourhoods may experience informational redundancy, limiting exposure to diverse viewpoints and reducing innovation in civic action. The most substantial negative effect on policy advocacy highlights that tight-knit subgroups may hinder outward-facing civic initiatives.

5. Substantial intercorrelations exist among network metrics. Degree and eigenvector centrality are strongly related (r = .803, p < .001), reflecting the structural reality that students with many ties tend to connect with other well-connected peers. Betweenness centrality and clustering coefficient are negatively associated (r = − .563, p < .001), illustrating that brokers typically operate in less clustered neighbourhoods. Despite these overlaps, the distinctive correlations with CPO prove that different structural roles—hubs, brokers, or clustered actors—each contribute uniquely to CPO.

6. Community service hours show weaker correlations with network position, with the strongest being eigenvector centrality (r = .308, p < .001). This proposes that service involvement is partly shaped by individual dispositions or institutional requirements outside the network structure, unlike electoral participation or leadership initiatives, which are more structurally embedded.

7. Finally, robust associations among CPO measures themselves reinforce theoretical expectations. The strong correlation between civic attitudes and behaviors (r = .683, p < .001) indicates that orientations and actions develop in tandem within governance contexts, and structural position moderates the strength of this relationship.

Figure 9 shows that eigenvector centrality is the strongest predictor across CPO (r = .428 for attitudes, 0.395 for behaviors, and 0.356 for leadership), whereas betweenness centrality uniquely predicts policy advocacy (r = .314). The clustering coefficient consistently inhibits civic engagement (r = -.213 for policy advocacy). These systematic correlations directly answer RQ2 by signifying that connections to significant peers drive most CPO, while boundary-spanning positions specifically enable policy innovation.

• (v)Result analysis of MR

Continuing with RQ2, we conduct a controlled analysis to examine whether network impacts continue when accounting for DF. The multivariate models isolate the unique contribution of network position while identifying which DF interact with network opportunities to shape CPO. The MR and LR analyses reveal nuanced relationships between SGN position and CPO, indicating that the network model provides predictive power beyond DF. As illustrated in Figure 10(a-b), the structural position affects attitudinal development and specific behavioral outcomes.

Figure 10.

(a) MR analysis for predicting CES civic attitude scores; (b) LR analysis for predicting binary CPO.

• (A) Predicting civic attitudes

The hierarchical regression model for CES civic attitude scores (Figure 10(a)) confirms that network features make a substantial contribution to the model’s explanatory power beyond DF. DF alone accounted for 21.8% of the variance, while network metrics accounted for 27.3%. The integrated model increased the total variance explained to 34.2%, with NSF adding a significant 12.4% (ΔR² = 0.124, p < .001).

Within the whole model, eigenvector centrality emerged as the strongest network predictor (β = 0.271, p = .001), indicating that exposure to well-connected peers fosters more positive civic orientations. The Community E–I Index (β = 0.129, p = .030) was also significant, indicating that students who balance internal community ties with external linkages report stronger civic attitudes. In contrast, degree and betweenness centrality became non-significant once eigenvector centrality was included, signifying that the quality of connections matters more than the quantity or bridging alone.

Among demographics, prior CPO (β = 0.221, p < .001) retained a strong effect, highlighting the role of existing civic dispositions. Year of study (β = 0.137, p = .033) and female gender (β = 0.149, p = .011) also positively predicted civic attitudes, signifying that more prolonged institutional exposure and gendered socialization may reinforce civic orientation. The previously significant negative effect of a STEM major became non-significant when network metrics were controlled, implying that disciplinary effects may be mediated by governance position.

• (B) Predicting civic behaviors

The logistic regression models (Figure 10 (b)) reveal differentiated structural predictors for behavioral outcomes:

• Leadership initiative was most strongly predicted by eigenvector centrality (OR = 1.57, p = .015). A one-standard-deviation increase corresponded to a 10% point rise in the probability of taking leadership roles. This proposes that students residing in influence-rich neighborhoods are more likely to assume new leadership positions, reinforcing the compounding impact of centrality. Prior CPO (OR = 1.68, p = .001) and year of study (OR = 1.41, p = .036) also contributed, presenting a complementary impact of experience and network integration.

• Policy advocacy was predicted by betweenness centrality (OR = 1.47, p = .034), Community E–I Index (OR = 1.43, p = .030), and a negative impact of clustering coefficient (OR = 0.71, p = .046). These results highlight that boundary-spanning roles, cross-community positioning, and reduced local redundancy enable advocacy behaviors. Students in tightly clustered neighbourhoods were less likely to engage in advocacy, consistent with the notion that innovation emerges from exposure to diverse perspectives rather than homogenous groups.

• Civic discourse participation was associated with eigenvector centrality (OR = 1.46, p = .036) and the Community E–I Index (OR = 1.42, p = .024), indicating that students connected to important peers and spanning multiple communities are more engaged in deliberation and debate. The prior CPO had a substantial marginal effect (0.142), reinforcing the idea that discourse participation reflects pre-existing dispositions and network opportunity structures.

The civic attitude regression achieved an adjusted R² of 0.313, explaining nearly one-third of the variance. Logistic models generated Nagelkerke R² values ranging from 0.224 to 0.256, along with classification performance characterized by AUC values between 0.733 and 0.755, indicating moderate but meaningful predictive accuracy.

Figure 10 validates that network features contribute 12.4% unique variance beyond DF (ΔR² = 0.124, p < .001), with eigenvector centrality (β = 0.271, p = .001) and E-I Index (β = 0.129, p = .030) significantly predicting civic attitudes. Binary outcomes reveal differentiated network effects: leadership initiative depends on eigenvector centrality (OR = 1.57), while policy advocacy requires betweenness centrality (OR = 1.47) and low clustering (OR = 0.71). These controlled analyses answer RQ2 by confirming that structural position independently shapes CPO beyond individual features.

• (vi) Performance of classification model (CM)

Addressing RQ3, this work evaluates ML models for predicting civic engagement using network and DF features. The classification analysis tests whether the identified network-civic relationships have sufficient strength and consistency for practical application in finding students likely to benefit from targeted civic development interventions. The classification model results provide strong empirical validation that network position and DF reliably predict civic engagement patterns, signifying the practical utility of network analysis for finding students likely to develop strong civic skills (Figure 11).

Figure 11.

Classification performance for predicting CPO.

• (A) Ensemble methods capture governance complexity XGBoost achieves superior performance across all metrics,accuracy (0.781), precision (0.736), recall (0.723), F1-score (0.730), AUC-ROC (0.842), specificity (0.821), and balanced accuracy (0.772),indicating that Gradient Boosting (GB) effectively captures the complex, nonlinear relationships between network position and CPO. This comprehensive performance recommends that civic development through governance participation involves intricate interactive effects among structural position, DF, and community membership, which ensemble methods can successfully model.

Random Forest (RF) consistently performs well, albeit with slightly lower accuracy (0.768) and AUC-ROC (0.834), confirming that tree-based ensemble methods excel at modelling governance network impacts. The strong performance of ensemble models recommends that civic engagement prediction requires capturing complex feature interactions and nonlinear relationships, which these algorithms efficiently identify by recursive partitioning and aggregation.

• (B)Moderate nonlinearity in network effects: SVM with an RBF kernel achieves intermediate performance (accuracy = 0.756, AUC-ROC = 0.823), signifying that kernel-based methods partially capture nonlinear network-civic relationships, but less effectively than ensemble methods. This pattern proposes that while civic engagement exhibits moderate nonlinearity, the relationships are sufficiently complex to benefit from ensemble methods’ capacity to model intricate feature interactions.

Logistic regression maintains respectable performance (accuracy = 0.738, AUC-ROC = 0.797) despite linear modelling constraints, indicating that substantial predictive power exists in direct linear relationships between network metrics and CPO. The relatively modest performance gap between logistic regression and complex nonlinear models proposes that while interaction effects enhance prediction, linear network effects capture meaningful variance in civic development.

• (C) Network interdependencies require sophisticated modelling. The substantially lower performance of simpler algorithms, Decision Tree (accuracy = 0.717) and Naive Bayes (accuracy = 0.697), highlights the critical limitations of methods that cannot capture complex feature interactions or violate independence assumptions. Naive Bayes particularly struggles due to substantial intercorrelations among network metrics, signifying that civic engagement prediction requires accepting the interconnected nature of governance network positions.
• (D) Strong discriminative capability: The high AUC-ROC scores across top-performing models (XGBoost = 0.842, RF = 0.834, SVM = 0.823) indicate a robust ability to decide between students who will and will not engage in civic behaviors. XGBoost’s superior AUC-ROC proves exceptional ranking performance, correctly ordering students by probability of civic engagement across multiple classification thresholds. This predictive power provides a practical basis for classifying students who would benefit from targeted civic development interventions.
• (E) Balanced performance across outcome classes: XGBoost maintains high specificity (0.821) alongside strong recall (0.723), indicating balanced performance in correctly classifying civically engaged and non-engaged students, rather than simply predicting the majority classes. This balanced capability proposes that the model effectively captures determinants of civic involvement and non-participation, providing a complete understanding of governance network effects.
• (F) Practical implementation implications The consistent algorithmic ranking across evaluation dimensions—XGBoost leading, followed by RF, SVM, Logistic Regression, Decision Tree, and Naive Bayes—provides a robust basis for practical civic engagement prediction. These results recommend that institutions can reliably identify students likely to benefit from enhanced governance chances or targeted civic development programming based on their network position and DF.

Figure 11 shows that XGBoost achieves superior performance (accuracy = 0.781, AUC-ROC = 0.842) compared to RF (0.768, 0.834), SVM (0.756, 0.823), and Logistic Regression (0.738, 0.797). The dominance of ensemble methods over linear methods indicates the need for sophisticated modelling of nonlinear network-civic relationships. These results answer RQ3 by signifying that network features reliably predict civic engagement with sufficient accuracy for practical implementation in classifying students for targeted civic development interventions.

Discussion

Restating objectives and major findings

This study aimed to investigate how structural position within SGNs predicts CPO, while controlling for DF. Three objectives guided the analysis: (i) to characterize the model of SGN, (ii) to assess the associations between network features and civic attitudes and behaviors, and (iii) to evaluate predictive modelling approaches for classifying CPO. The results consistently prove that network position is a significant determinant of civic participation, with eigenvector centrality strongly predicting civic attitudes and leadership initiative, betweenness centrality predicting policy advocacy, and clustering negatively associated with outward civic engagement. Importantly, regression models proved that these impacts persist even when DF predictors are taken into account, while ML classification confirmed the viability of predicting CPO with high accuracy.

Integration with prior literature

The findings reinforce earlier work showing that institutional networks shape CPO. The central coordinating role of SGN leaders aligns with research indicating that institutional structures adapt to maintain civic engagement during disruptions, as proved during the COVID-19 pandemic, when new forms of participation emerged with structured support ²⁰. This adaptability ensures continuity of civic engagement even when conventional pathways are constrained.

The DF observed here also aligns with those reported in prior studies. Gender-based differences in participation align with findings from Qatar ²¹, while the predictive role of prior civic experience supports broader evidence that individual background continues to influence civic development. Notably, this study also highlights that network features account for unique variance beyond demographics, thereby extending regional and international insights from Malaysia ²⁵ and the ICCS 2022 report ²⁶.

Cross-community positioning (E–I Index) was associated with stronger attitudes and discourse, resonating with research on community partnerships and participatory curricula that emphasize outward-facing engagement ^22,23. Similarly, the predictive strength of eigenvector centrality aligns with findings on the importance of autonomy and peer impact ²⁴, while the negative effects of clustering highlight the structural barriers to innovation noted in that work.

Finally, the developmental benefits of civic engagement observed in Hong Kong adolescents ²⁹ are consistent with this study’s finding that attitudes and behaviors develop in tandem. Together, these connections recommend that SGN functions not only as a governance mechanism but also as a structured environment for civic learning and identity formation.

• (A)Advancing beyond prior findings.

While the preceding discussion aligns with existing scholarship, this study extends beyond prior research in several critical areas. Patrick ⁴⁸ and Rejubi et al. ⁴⁹ established that student governance participation generally benefits civic development through formal organizational structures, but their analyses did not differentiate how structural position within networks systematically produces different outcomes. Our network-based approach reveals that students occupying identical formal roles (e.g., council members) can experience vastly different civic benefits depending on their centrality and cross-community connectivity,a structural stratification invisible through conventional organizational analysis.

Li and Zhao ⁴⁷ proved that Chinese students engage with governance through digital platforms, but their work examined participation mechanisms rather than the differentiation of outcomes. This work’s empirical results advance this by demonstrating that, in face-to-face governance networks, eigenvector centrality predicts civic attitudes and leadership (r = .428, OR = 1.57), while betweenness centrality uniquely predicts policy advocacy (r = .314, OR = 1.47), thereby establishing differentiated relationships not previously established. This reveals that governance networks operate by multiple structural pathways: influence-based mechanisms (eigenvector) foster attitudinal development and leadership aspiration, while brokerage positions (betweenness) specifically enable advocacy by exposing students to diverse policy perspectives.

Prior research on social capital in HE ⁴⁶ has shown that network ties broadly shape student outcomes, but has not operationalized these concepts using quantitative centrality measures or examined civic-specific outcomes. Our integration of network metrics with civic participation scales provides empirical precision lacking in previous work, indicating that a one-standard-deviation increase in eigenvector centrality corresponds to a 10-percentage-point increase in leadership initiative probability, a level of quantitative specification that enables practical application in governance design.

Finally, while Tsoni et al. ⁵⁰ verified ML’s capacity to predict civic engagement from behavioral features, our study is the first to achieve robust classification (AUC = 0.842) using network structural positions in student governance contexts. This validates that relational topology contains sufficient predictive information to rival or exceed demographic-behavioral models, operationalizing network theory into actionable institutional tools. The superior performance of ensemble models (XGBoost > RF > SVM > Logistic Regression) further indicates that network-civic relationships involve complex nonlinear interactions that are not adequately captured by traditional regression methods used in prior governance research.

Theoretical contributions

This study advances civic network theory using 3 interconnected contributions that address gaps in existing scholarship:

• (A) Structural differentiation of civic outcomes: Prior research has shown that governance participation generally promotes civic development ^48,49, but has not proved how specific network positions predict distinct result types. This study reveals that different forms of civic participation, attitudes, leadership initiative, policy advocacy, and DF, correspond to specific structural roles: core actors embedded in influence-rich neighborhoods (high eigenvector centrality), boundary spanners bridging disconnected communities (high betweenness centrality), and cross-community connectors maintaining diverse ties (positive E-I Index). This differentiation shows that civic development is not uniformly distributed but systematically stratified by network position, challenging the assumption that participation impacts are homogeneous across governance roles.
• (B) Independent predictive power of network structure: While social capital theory ^35,46 has long proposed that relational networks matter for civic capacity, empirical validation demonstrating that the network model explains variance in civic outcomes beyond individual DF in student governance contexts had not been established. This study provides validation, showing that network features account for 12.4% of unique variance (ΔR² = 0.124, p < .001) after controlling for gender, academic discipline, year of study, and prior civic experience. This empirically confirms that civic learning is embedded in institutional network design rather than residing solely in individual predisposition, with practical implications: governance structures can be intentionally designed to democratize access to high-centrality positions, thereby distributing civic benefits more equitably.
• (C) Complex interaction-based determinants: Traditional governance research has relied on linear models that assume additive effects of participation variables. The superior classification performance of ensemble ML models (XGBoost accuracy = 0.781, AUC = 0.842) over linear approaches (Logistic Regression accuracy = 0.738, AUC = 0.797) validates that civic outcomes emerge from complex, nonlinear interactions among NSF that cannot be adequately captured by conventional statistical methods ⁵⁰. This advances theoretical understanding by revealing that network position, community membership, and tie strength interact synergistically rather than operating independently, signifying that governance interventions targeting multiple structural dimensions simultaneously may produce multiplicative rather than additive civic benefits.

Practical implications

The results underscore several practical implications for HE governance:

• Support for boundary-spanners: Since policy advocacy is facilitated by betweenness and cross-community ties, institutions should recognize and resource positions that connect disparate governance domains.

• Balancing cohesion with openness: While clustering fosters trust, it risks insularity; universities should encourage inter-community projects that promote diversity of perspectives.

• Democratizing access to impact-rich cores: Civic benefits accrue disproportionately to students embedded in central clusters. Mentorship and leadership pathways should be designed to extend these opportunities to peripheral actors.

• Predictive monitoring for intervention: The strong performance of ensemble classifiers proves the feasibility of predictive tools in identifying students at risk of civic disengagement and directing support accordingly.

While these interventions require institutional investment, the predictive modelling demonstrates feature returns in terms of enhanced CPO across broader student populations.

Limitations and future directions

This study is limited by its focus on a single institutional setting, which restricts its generalizability. Future work should test these models in diverse contexts, replicating across cultural regions as highlighted in Malaysia ²⁵ and the ICCS 2022 findings ²⁶. The reliance on cross-sectional data limits causal claims; longitudinal tracking would better capture how governance trajectories evolve. Finally, while network and DF were central to this analysis, unobserved factors such as motivation, socioeconomic background, and psychological resources (as emphasized in Hong Kong research ²⁹⁾ likely contribute to CPO. They should be integrated into future mixed-methods research.

Conclusion and future works

This study proves that structural position within student SGN significantly predicts CPO beyond DF, providing the first empirical validation of network-theoretic predictions in student governance contexts by integrated social network analysis, multivariate regression, and ML classification. While prior research examined student governance using organizational models or participation mechanisms, this investigation reveals that civic benefits are systematically stratified by network position rather than uniformly distributed—a structural differentiation not previously developed.

Social network analysis revealed a modular network with hierarchical connectivity, where central leadership roles maintained extensive cross-community links while specialized governance domains operated with greater insularity. NSF accounted for 12.4% unique variance (ΔR²=0.124, p < .001) in civic attitudes after controlling for DF, validating that relational structure shapes civic capacity independent of individual features.

Critically, this study proves differentiated prediction patterns not shown in prior governance research: eigenvector centrality uniquely predicted civic attitudes (β = 0.271) and leadership initiative (OR = 1.57), while betweenness centrality specifically predicted policy advocacy (OR = 1.47). This reveals that governance networks operate through multiple structural pathways—influence-based mechanisms foster attitudinal development, while brokerage positions enable advocacy by exposing individuals to diverse perspectives.

ML classification achieved robust performance (XGBoost: accuracy = 0.781, AUC = 0.842), representing the first application of ML to network structural positions in student governance. This proves that network topology provides predictive power comparable to traditional demographic models, operationalizing theoretical constructs into actionable institutional tools. The superior performance of ensemble methods reveals that civic outcomes emerge from complex, nonlinear interactions among NSF—complexity that single-method studies cannot capture.

These findings transform the understanding of governance participation from a binary phenomenon to a structurally differentiated process in which position confers developmental benefits. Institutions should foster distributed governance networks that develop multiple pathways for cross-community connections, rather than relying solely on central leadership. Expanding opportunities for peripheral actors to connect across organizational boundaries may democratize civic learning and improve overall outcomes.

Future research should examine how SGNs evolve over time and whether intentional interventions—such as structured cross-committee collaborations or mentorship pipelines—can strengthen civic participation across all structural positions. Longitudinal studies could assess whether network position changes correspond to civic development trajectories, while cross-institutional comparisons would illuminate whether these relationships generalize beyond Chinese regional university contexts.

Author contributions

Conceptualization, Methodology, Software, Validation, Formal analysis, Investigation, Resources, Data CurationWriting - Original Draft, Writing - Review & Editing, Visualization : Jing Liu, Putu Kerti Nitiasih, Made Hery Santosa, Putu Nanci Riastini.

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.