Innovation Networks in the New Energy Vehicle Industry: A Dual Perspective of Collaboration Between Supply Chain and Executive Networks

Abstract

Driven by the global energy transition and the pursuit of dual carbon goals (carbon peaking and carbon neutrality), the innovation network of the new energy vehicle (NEV) industry, composed of enterprises, universities, and research institutes, has become a key driver of sustainable industrial development. The evolution of this network is jointly shaped by both supply chain networks (SCNs) and executive networks (ENs), representing formal and informal relational structures, respectively. To systematically explore these dynamics, this study analyzes panel data from Chinese A-share-listed NEV firms covering the period 2003–2024. Employing social network analysis (SNA) and Quadratic Assignment Procedure (QAP) regression, we investigate how SCNs and ENs influence the formation and structural evolution of innovation networks. The results reveal that although all three networks exhibit sparse connectivity, they differ substantially in their structural characteristics. Moreover, both SCNs and ENs have statistically significant positive effects on innovation network development. Building on these findings, we propose an integrative policy framework to strategically enhance the innovation ecosystem of China’s NEV industry. This study not only provides practical guidance for fostering collaborative innovation but also offers theoretical insights by integrating formal and informal network perspectives, thereby advancing the understanding of multi-network interactions in complex industrial systems.

1. Introduction

Driven by the global energy transition and the pursuit of dual carbon goals (carbon peaking and carbon neutrality), the new energy vehicle (NEV) industry has emerged as a pivotal arena for technological convergence and strategic competition. Its sustained development depends heavily on cross-organizational collaborative innovation [1]. Industrial innovation networks, which consist of enterprises, universities, and research institutes, integrate diverse technological resources and social capital. Over time, these networks have gradually built systemic structural advantages, becoming a core driver of industrial growth [2]. However, this rapid market expansion has not been accompanied by a proportional improvement in network efficiency or inclusiveness. By the end of 2024, cumulative global NEV sales had exceeded 17 million units, with China alone contributing more than 11 million units, firmly maintaining its position as the world leader [3]. Despite this dominant market share, China’s NEV innovation network exhibits marked structural imbalances. Specifically, leading firms have formed closed “innovation islands” that restrict knowledge diffusion and resource sharing [4]; meanwhile, the participation rate of small and medium-sized enterprises (SMEs) in collaborative activities remains below 20% [5]. Furthermore, the integration level of cross-domain technological networks is less than one-third of that seen in traditional industries [6]. These issues have become key structural constraints limiting the transition of China’s NEV industry toward high-quality development. Overcoming these barriers necessitates an in-depth and systematic analysis of the multifaceted factors that shape the structure and dynamics of the innovation network.

In the literature on the formation mechanisms of innovation networks, prior studies have predominantly focused on single-dimensional factors, such as industrial technological complementarity [7] or policy incentives [8]. Few have examined the phenomenon from a holistic perspective of cross-network embeddedness, leading to fragmented and unsystematic explanations of the structural characteristics of innovation networks. Given that innovation networks are inherently network-based in nature, it is crucial to construct and empirically measure the influence of other related networks on their formation. This necessity forms the core research motivation of the present study. Organizational institutionalism suggests that both formal and informal institutions shape organizational innovation behaviors [9]. Formal institutions govern innovation through codified rules and contractual arrangements, while informal institutions exert influence through social norms, trust, and interpersonal interactions. Likewise, network embeddedness theory [10] posits that external networks affect innovation networks via two distinct mechanisms: structural embeddedness, which captures the formalized positions and connections within a network, and relational embeddedness, which emphasizes the quality of interpersonal ties and trust-based exchanges. Integrating these perspectives within the new energy vehicle (NEV) industry context, we conceptualize a dual-network framework. Specifically, the supply chain network (SCN) embodies structural embeddedness, as it arises from formal contractual relationships among suppliers and manufacturers. In contrast, the executive network (EN) represents relational embeddedness, being shaped by executives’ social interactions and trust capital. We therefore argue that the formation and evolution of innovation networks in the NEV industry are jointly influenced by the interdependent dynamics of these two networks.

To empirically test this proposition, we employ the Quadratic Assignment Procedure (QAP) regression method. QAP controls for network autocorrelation by using random permutation of relational matrices, effectively distinguishing true cross-layer effects from spurious structural noise. This makes it particularly well-suited for analyzing interdependent networks [11]. Compared to conventional methods that rely solely on node-level attributes or a single network structure, QAP enables the simultaneous analysis of multiple relational matrices, offering a rigorous empirical pathway for testing the “institution-relationship” dual interaction central to embeddedness theory.

This study makes four key contributions to theory and practice:

First, it reveals how China’s NEV innovation network emerges from the synergistic interaction between formal supply chain contracts and informal executive ties. This finding bridges the long-standing divide between technology-driven and institution-driven perspectives, thereby extending the application boundaries of Granovetter’s embeddedness theory.

Second, we develop an innovative dual-network QAP regression approach specifically designed to mitigate cross-network autocorrelation biases. This methodological advancement offers a novel analytical tool for accurately decoding the dynamics of interdependent innovation ecosystems.

Third, our findings demonstrate that such dual-network interdependence is not unique to the NEV sector. Similar patterns are likely to exist in other technology-intensive industries, where innovation processes must navigate the dual challenges of formal institutional governance and informal relational governance.

Finally, by identifying the mechanisms underlying structural imbalances within innovation networks, we provide actionable governance implications for optimizing network structures, improving collaboration efficiency, and promoting inclusive and sustainable innovation.

2. Literature Review

To systematically review research on innovation networks in the new energy vehicle industry, this study adopted a systematic literature retrieval approach. The search was conducted primarily through the following databases: Web of Science, Scopus, CNKI (China National Knowledge Infrastructure), and CSMAR, supplemented by industry reports from organizations such as the International Energy Agency (IEA) and the Ministry of Industry and Information Technology. The literature search covered the period from 2000 to 2024 to encompass major research outcomes in the NEV industry and innovation network-related fields over the past two decades. Key search terms included: “New Energy Vehicle (NEV)”, “Innovation Network”, “Supply Chain Network (SCN)”, “Executive Network (EN)”, “Collaborative Innovation”, “Social Network Analysis (SNA)”, and “QAP regression”. The included literature types encompassed academic journal articles, conference papers, policy research reports, and industry studies, ensuring both academic rigor and practical relevance.

2.1. Theoretical Foundations of Innovation Networks

Innovation networks are collaborative systems that transcend organizational boundaries. They are formed by diverse actors—including firms, research institutions, government agencies, and intermediaries—through the flow of knowledge and technological cooperation [12]. Their defining characteristics lie in the synergistic integration of knowledge sharing, resource complementarity, and collaborative innovation. Knowledge sharing relies on trust mechanisms and organizational learning capacity [12]; resource complementarity emerges from bridging knowledge gaps across domains [13]; and collaborative innovation stems from the deep integration of heterogeneous capabilities [14]. Empirical studies demonstrate that these networks enhance innovation efficiency by consolidating dispersed resources (e.g., shared technology patents) and reducing cooperation risks (e.g., establishing joint R&D standards) [15].

The theoretical foundation of innovation network research is rooted in Granovetter’s embeddedness theory, which challenges the dichotomy between “undersocialized” and “oversocialized” perspectives by emphasizing that economic actions are deeply embedded within social relationships [10]. Granovetter distinguishes between two dimensions of embeddedness: relational embeddedness, which focuses on the trust and depth of bilateral ties [16], and structural embeddedness, which examines how an actor’s position within the broader network shapes opportunities and constraints [17]. This framework provides critical insights into innovation collaboration, highlighting that inter-firm knowledge exchange and technological cooperation are not solely governed by formal contracts but are profoundly shaped by informal social relationships.

While embeddedness theory has significantly advanced our understanding of inter-firm collaboration, its single-network perspective is insufficient to capture the complexity of modern innovation systems, where organizations are simultaneously embedded in multiple, overlapping networks. The emergence of multilayer network theory provides a new analytical lens to address this limitation [18]. This approach emphasizes that different types of relationships—such as supply chain ties, executive social connections, and innovation collaborations—interact to form a coupled, interdependent system [19]. Under this view, the formation of innovation networks is not merely the outcome of a single network structure but rather the product of cross-layer interactions.

Despite these advances, notable theoretical gaps remain. First, much of the existing research overemphasizes individual firm attributes (e.g., size) or local network characteristics (e.g., node centrality) [20], while neglecting the systemic effects of supply chain networks (SCNs) and executive social ties on broader patterns of collaboration. For example, heavy supply chain dependency may shape the direction of innovation cooperation, potentially overshadowing the role of technological complementarity [21]. Second, prior studies tend to focus on explicit innovation outputs (e.g., patent counts) without addressing relational imbalances within innovation networks. Collaboration between core firms and peripheral participants often leads to unidirectional knowledge flows, reinforcing a “Matthew Effect” in resource allocation [22,23]. Such imbalances not only undermine network inclusiveness but also intensify resource misallocation, as critical technological expertise held by small and medium-sized enterprises (SMEs) remains underutilized due to structural deficiencies in collaboration governance [24]. Finally, limited attention has been given to how network relationships adapt dynamically to technological transitions or shifts in policy environments. This lack of theoretical explanation constrains our understanding of how innovation networks evolve in response to external changes [25].

2.2. Impacts of Supply Chain Networks on Innovation Networks

SCNs are multi-actor collaborative systems established by core firms. They are characterized by the bidirectional integration of resource flows (e.g., materials and capital) and knowledge flows, forming a dynamically evolving innovation ecosystem [26]. Their core functions include resource provisioning, information transmission, and risk sharing. In the NEV industry, these networks exhibit a distinct “glocalization” feature: core firms (e.g., battery manufacturers) rely on modular supply chain designs to balance global resource integration (e.g., global sourcing of lithium) with localized innovation demands (e.g., rapid responsiveness to regional markets) [27].

As the foundational architecture of industrial innovation, the impact of supply chain networks can first be understood through the lens of structural embeddedness as defined by Granovetter. Through legally binding formal contracts, these networks establish a rigid framework for inter-firm resource dependence and knowledge flow, providing a stable institutional environment for innovation activities [28]. Specifically, supply chain networks influence innovation networks through two primary mechanisms. Firstly, the central network position of core firms grants them dominant authority in technological coordination and resource integration [29]. For instance, Tesla’s vertical integration strategy in restructuring the battery supply chain has not only accelerated technological iteration but also reshaped innovation collaboration patterns across upstream and downstream segments [30]. Secondly, complementary resources among supply chain members—such as material suppliers’ patents and vehicle manufacturers’ data—are activated through formal collaborations like joint R&D, ultimately translating into tangible outputs within innovation networks [31]. Nonetheless, it is important to note that such structural embeddedness may also entail risks of innovation path dependence, where excessively strong ties within supply chains could potentially constrain firms’ external knowledge search scope [32].

It should be emphasized that the knowledge transfer effects of supply chain networks are not automatically realized—their value realization largely depends on firms’ absorptive capacity [33]. This perspective bridges classical structural embeddedness with the dynamic capabilities view, offering critical insights into why supply chains generate differentiated innovation outcomes across firms. However, existing studies have predominantly examined supply chains in isolation, failing to systematically analyze their synergistic effects with informal relationships such as executive social networks. Moreover, traditional econometric methods often struggle to adequately address structural endogeneity issues inherent in network data, which constitutes a key motivation for the methodological innovation pursued in this study.

2.3. Impacts of Executive Networks on Innovation Networks

As a prominent form of relational embeddedness, executive networks are built upon informal social ties among senior managers, including board interlocks, alumni affiliations, and industry-related social interactions [34]. Its core functions encompass three dimensions: trust building, established through repeated interactions and relational commitment [35]; information bridging, by leveraging structural holes to access heterogeneous knowledge [36]; and resource mobilization, harnessing social capital to acquire external innovation assets [37].

Granovetter’s theory of relational embeddedness posits that social relationships constitute a fundamental basis for economic action. Burt’s structural holes theory further reveals that executives occupying intermediary positions in networks can access non-redundant information and gain control advantages. Recent research has deepened this understanding by demonstrating that the value of executive networks stems not only from their structural characteristics but also critically depends on executives’ knowledge brokerage behaviors—specifically, their capacity to proactively identify, filter, and integrate heterogeneous knowledge distributed throughout the network [38]. Empirical studies demonstrate that ENs enhance collaborative efficiency within innovation networks primarily through informal channels that facilitate tacit knowledge transfer (e.g., technical know-how and market insights) and social sanction mechanisms that mitigate partnership risks [39].

Despite considerable scholarly attention, significant theoretical gaps remain in the literature. While numerous studies have investigated the direct effects of executive networks (ENs) on firm innovation outcomes, such as how board interlocks promote patent collaborations [40], much less is known about their bridging role and function as knowledge brokers within cross-network interactions [41]. These dimensions have not yet been systematically theorized or empirically validated. One critical issue concerns the structural misalignment between networks. Within the supply chain network (SCN), formal and rigid contractual relationships often operate in isolation from the fluid and dynamic knowledge flows characteristic of innovation networks. This disconnection raises an important question: can executives, through cross-appointments and board interlocks, activate weakly connected regions and bridge these two networks [42,43], Addressing this unresolved issue is essential for understanding how formal and informal networks jointly shape innovation collaboration.

2.4. Research Framework

Drawing upon the theoretical foundations discussed above, this study develops an integrative analytical framework to systematically examine how supply chain networks (SCNs) and executive networks (ENs) jointly shape the formation and evolution of innovation networks (INs), as illustrated in Figure 1. In this framework, SCNs are rooted in formal contractual relationships, while ENs are characterized by informal social ties. These two networks represent the dual dimensions of Granovetter’s embeddedness theory: institutional structures and relational. Together, they form a “dual-engine system” that drives firms’ innovation activities. From this perspective, the innovation network can be viewed as an emergent structure shaped by the dynamic interaction between these two foundational networks. This conceptualization reflects the temporal and strategic primacy of SCNs and ENs. A firm’s supply chain configuration and executive composition typically serve as structurally stable antecedent conditions, providing a foundation upon which firms can actively search for and construct innovation collaborations [44,45]. In other words, the innovation network evolves within pre-existing formal and informal relational structures that jointly influence how innovation partnerships are initiated, maintained, and transformed over time.

Guided by this framework, the empirical analysis focuses on examining the independent effects of SCNs and ENs on innovation networks, as well as their combined explanatory power. To accomplish this, the study employs the Quadratic Assignment Procedure (QAP), a network-based statistical method that provides rigorous empirical evidence for the hypothesized relationships while addressing potential autocorrelation issues. Importantly, this study emphasizes the incremental contributions of SCNs and ENs rather than their statistical interaction effects. This approach offers a conceptually coherent perspective on the relative importance of different embeddedness mechanisms. Furthermore, it highlights future research opportunities: moving beyond static correlation analyses toward dynamic co-evolution models that capture the interplay of multilayer networks over time. Such a shift would deepen theoretical insights and provide a stronger foundation for understanding collaborative innovation in complex industrial ecosystems.

3. Methods and Data

3.1. Methodological Framework

3.1.1. Social Network Analysis

Social Network Analysis (SNA) models social structures using nodes and edges, integrating topological visualization with quantitative metrics. Following Yuan et al. [46], a dual-dimensional framework is adopted to assess macro and micro characteristics. At the macro level, network density reflects collaborative intensity via the ratio of actual to potential connections, while average path length indicates information diffusion efficiency. At the micro level, node-specific metrics include degree centrality, (identifying resource hubs), betweenness centrality (highlighting gatekeepers), and closeness centrality (measuring communication efficiency). This approach captures both ecosystem-wide properties and individual actor roles, enabling multiscale insight into network dynamics. Detailed metric definitions are provided in

Table 1. Network metrics and their formulas.

Metric	Formula	Symbolic Meaning
Network density	$D ={\displaystyle \frac{2L}{N(N-1)}}$	The closeness of the connections between enterprises in an enterprise network. L refers to the number of actual edges in the network; N represents the number of nodes.
Clustering coefficient	$C={\displaystyle \frac{1}{n}}\sum _{i=1}^n{\displaystyle \frac{2E_i}{n(n-1)}}$	The degree of connection and clustering among enterprises in the enterprise network. Ei indicates the number of neighboring enterprises actually associated with the enterprise; n is the number of nodes.
Average path length	$L={\displaystyle \frac{\sum _{i,j}^n{q}_{ij}}{n(n-1)}}$	The average number of steps along the shortest paths for all possible pairs of enterprises in the network. n is the number of nodes; qij is the shortest path length between enterprises i and j.
Degree centrality	$C_i={\displaystyle \frac{\sum _{j=1,j\ne i}^n{w}_{ij}}{n-1}}$	The position and importance of an enterprise within the enterprise network. n is the number of enterprise nodes; wij represents the strength of the connection between enterprises i and j.
Closeness centrality	$CC={\displaystyle \frac{N-1}{\sum _{j=1}^n{d}_{ij}}}$	The degree to which an actor is not controlled by other actors. N is the number of enterprise nodes; dij is the shortest distance between enterprises i and j.
Betweenness centrality	${CC}_i={\displaystyle \frac{2\sum _j^n{\sum }_k^n{c}_{jk}\left(i\right)}{n^2-3n+2}},$ $j\ne i\ne k, j<k$	The degree to which an actor controls resources by bridging others. cjk(i) is the probability that i is on the shortest path between enterprises j and k, ranging from 0 to 1.

.

3.1.2. QAP Analytical Method

This study employs Quadratic Assignment Procedure (QAP) analysis to examine the influence of executive networks and supply chain networks on innovation networks. In contrast to the Exponential Random Graph Model (ERGM), which requires predefined assumptions about network formation rules, QAP regression retains the intrinsic topological characteristics of each network through matrix permutation tests. This approach provides a more accurate identification of the relationships between observed networks [47].

Before conducting the regression measurement of network influence, it is necessary to ensure a correlation exists between the explained variable and the explanatory variables, which requires a QAP correlation analysis. The procedure is divided into four steps: (1) Calculate the correlation coefficients between variables. Treat all values in the innovation network, supply chain network, and executive network as long vectors and compute the correlation coefficient between these two vectors. (2) Random permutation and resampling. Randomly permute the rows and corresponding columns of the innovation network, then calculate the correlation coefficient between the permuted network and the supply chain network. Repeat this process 2000 times to form a distribution of correlation coefficients. (3) Statistical inference. Compare the initially calculated correlation coefficient with the distribution obtained through random permutation. If the observed correlation coefficient is significantly higher than those obtained after random permutation, it can be concluded that a significant relationship exists between the innovation network and the supply chain network. (4) Analogous calculation. Based on the principles of steps (2) and (3), calculate the correlation between the innovation network and the executive network.

To measure the impact of the dual networks on the innovation network, QAP regression analysis is conducted based on the QAP correlation analysis. The steps are divided into five parts: (1) Data preparation and preprocessing. Convert non-binary data in each network into binary data using UCINET6.212 software. (2) Random permutation and coefficient estimation. Keeping the independent variable networks (supply chain and executive networks) unchanged, randomly permute the rows and columns of the dependent variable network (innovation network), then calculate the regression coefficients. Repeat this process 2000 times to obtain a random distribution of regression coefficients. (3) Significance testing. Compare the actually observed regression coefficients with the random distribution and calculate the p-value to determine their statistical significance. (4) Analogous calculation. Based on the principles of steps (2) and (3), calculate the regression relationship between the innovation network and the executive network. (5) Result interpretation and inference. Based on the regression coefficients, coefficient of determination, and significance levels, determine the degree of influence and significance of the explanatory variables on the explained variable, thereby inferring the influential relationships between the networks.

Following Su et al. [48], a cooperative innovation relationship matrix is constructed for the NEV industry, defined as:

$${Parent}_i=f\left({Sp}_i,{Ep}_i,R_i,D_i,O_i,{AE}_i\right) i\in (1,58)$$

(1)

where ${Parent}_i$ represents the patent collaboration matrix, ${Sp}_i,{Ep}_i,R_i,D_i,O_i,{AE}_i$ and denote individual network characteristic matrices, respectively.

This model supports two core analyses: QAP correlation analysis, which evaluates the interdependencies among independent variables to mitigate multicollinearity risks; and QAP multiple regression, which assesses the impacts of independent variables (e.g., SCN centrality and EN density) on the dependent variable (IN performance).

3.2. Data Sources and Processing

The new energy vehicle (NEV) industry serves as a typical example for studying multi-layer network dynamics. In the Chinese context, it exhibits distinct characteristics marked by strong policy guidance, rapidly evolving market competition, and deeply intertwined relational culture. National top-level design and mandatory technical regulations (such as the new national standards for power battery safety) have significantly raised R&D compliance thresholds, compelling firms to strengthen supply chain alliances (structural embeddedness) and collaborative innovation to mitigate risks. Meanwhile, intense market competition forces companies to rely on executive networks (relational embeddedness) to swiftly access cutting-edge information and opportunities, enabling them to rapidly embed themselves into innovation networks. Furthermore, the profound relational culture provides an informal governance mechanism based on trust for these interactions, reducing cooperation uncertainties through reputation and reciprocity, thereby making executive networks a key form of social capital for stabilizing supply chains and facilitating knowledge flow. These three contextual dimensions collectively form the systemic driving forces that shape the interactions among the three types of networks.

The data processing was performed in three steps. First, key enterprises were identified using the 2012 China Securities Regulatory Commission. (CSRC) industry code (C36) as the primary filter, yielding 4242 SCNs, 18,787 ENs, and 21,277 innovation networks from their respective alliance databases. Second, a core sample set was established by cross-matching the enterprise codes across the three networks. Firms flagged as ST/*ST-listed or lacking complete supplier/customer data were excluded, resulting in 58 core enterprises (

Table A1. 58 core enterprises full name and abbreviation.

Full Company Name	Acronym	Full Company Name	Acronym
Anhui Ankai Automobile Co., Ltd.	AAAG	Shanghai Motor Group Co., Ltd.	SMG
Anhui Ankai Futian Shuguang Car Bridge Co., Ltd.	AAFTSA	Shanghai Shenzhou New Energy Development Co., Ltd.	SSNED
Beijing Foton Daimler Automobile Co., Ltd.	BFDA	Shanghai Solar Energy Technology Co., Ltd.	SSETC
Beiqi Foton Motor Co., Ltd.	BFMC	Shanghai Xinpeng Metal Products Co., Ltd.	SXMP
Bosch Automotive Diesel Systems Co., Ltd.	BADSC	Shanghai Xinpeng Industrial Co., Ltd.	SXI
Changzhou Guangyang Bearing Co., Ltd.	CGBC	Shenzhen Terjia Technology Co., Ltd.	SZTJTC
Dongfeng Electronics Technology Co., Ltd.	DFET	Sichuan Chengfei Integrated Technology Co., Ltd.	SCIAMC
Dongfeng Motor Co., Ltd.	DFMC	Tianrun Crankshaft Co., Ltd.	TRCF
Fuao Auto Parts Co., Ltd.	FUAO	Wanxiang Qianchao Co., Ltd.	WXQC
Guangzhou Automobile Group Co., Ltd.	GZAG	Weichai Power Co., Ltd.	WCP
Guizhou Guihang Auto Parts Co., Ltd.	CGBC	Weichai Power Yangzhou Diesel Engine Co., Ltd.	WCPY
Haima Motor Co., Ltd.	HMM	Wuxi Weifu High-tech Group Co., Ltd.	WEFU
Henan Xixia Automobile Water Pump Co., Ltd.	HXAPC	Xuchang Yuandong Drive Shaft Co., Ltd.	XYDS
Henan Zhongyuan Internal Distribution Co., Ltd.		FAWHaima Automobile Co., Ltd.	FAWH
Hubei CheBridge Co., Ltd.	HBCB	China FaW Co., Ltd.	CFAW
Huayu Automotive Systems Co., Ltd.	HAS	Faw Car Co., Ltd.	FAW
Huaiji Moon valve Co., Ltd.	HJLMVC	Changchun Yidong Clutch Co., Ltd.	CYCC
Huaiji Dengyun Auto Parts Co., Ltd.	HDAP	Changchun FAW Fuwei Auto Parts Co., Ltd.	CFFAP
Jiangling Automobile Co., Ltd.	GJMA	Changchun FAW Sihuan Automobile Co., Ltd.	CFSHAC
Jiangsu Oliwei Sensor Hi-Tech Co., Ltd.	JASHC	Zhejiang Wan’an Technology Co., Ltd.	ZWTC
Jiangsu Pacific Precision Forging Technology Co., Ltd.	JPPFT	Zhejiang Wanfeng Aowei Steam Turbine Co., Ltd.	ZWAW
Jiangsu Xinquan Automobile Decoration Co., Ltd.	JXAT	Zhejiang Asia-Pacific Electromechanical Co., Ltd.	ZAPME
Jiangsu Yunyi Electric Co., Ltd.	JYNET	Zhejiang Yinlun Machinery Co., Ltd.	ZYMC
Kunming Yunnei Power Co., Ltd.	KYPC	Zhengzhou Coal Mine Machinery Group Co., Ltd.	ZCMMG
Inner Mongolia Shenzhou Silicon Industry Co., Ltd.	IMSSI	Zhengzhou Yutong Group Co., Ltd.	ZYG
Shandong Binzhou Bohai Piston Co., Ltd.	SBBP	Zhengzhou Yutong Bus Co., Ltd.	ZYGB
Shandong Longji Machinery Co., Ltd.	SDLJ	China Automotive Engineering Research Institute Co., Ltd.	CAERI
Shandong Xingmin Steel Ring Co., Ltd.	SDXM	Zhongtong Bus Holding Co., Ltd.	ZTBC
Shanghai Aerospace Vehicle Electromechanical Co., Ltd.	SAACMEC	Chongqing Changan Automobile Co., Ltd.	CCAC

) with robust multi-network profiles and distinct NEV industry characteristics. These firms form the primary nodes of the NEV industry network, enabling the construction of a high-quality relationship matrix. Finally, a 58 × 58 enterprise relationship matrix was built to capture multilevel network dynamics by incorporating variables such as joint patent collaborations, supply chain transactions, and executive interlocks. This matrix served as the foundation for subsequent analyses of the interactions between INs, SCNs, and ENs.

3.3. Variable Measurement

All variables were constructed as relational matrices to analyze the impact of multi-network interactions on innovation networks in the NEV industry. The detailed variable definitions and operationalizations are presented in Table 2.

Table 2. The variable measures and literature basis.

Variable Type	Variable	Measurement Method	Literature Basis
Dependent variable	Innovation network (Parenti)	Patent cooperation matrix	Su et al. [48]
Independent variables	Supply chain network (Spi)	Supply relationship matrix	Sun et al. [49] Liu et al. [50]
	Executive network (Epi)	Executive concurrent relationship matrix	Ding et al. [51]
Control variables	Enterprise scale similarity (Ri)	Business scale similarity matrix	Xiang et al. [52]
	Geographic proximity (Di)	Geoproximity matrix	Cao et al. [53]
	Ownership homogeneity (Oi)	Ownership difference matrix	Zhang and Qian [54]
	Enterprise age Heterogeneity Network (${AE}_i$)	Age differences matrix	Luo et al. [55]

Table 2. The variable measures and literature basis.

Variable Type	Variable	Measurement Method	Literature Basis
Dependent variable	Innovation network (Parenti)	Patent cooperation matrix	Su et al. [48]
Independent variables	Supply chain network (Spi)	Supply relationship matrix	Sun et al. [49] Liu et al. [50]
	Executive network (Epi)	Executive concurrent relationship matrix	Ding et al. [51]
Control variables	Enterprise scale similarity (Ri)	Business scale similarity matrix	Xiang et al. [52]
	Geographic proximity (Di)	Geoproximity matrix	Cao et al. [53]
	Ownership homogeneity (Oi)	Ownership difference matrix	Zhang and Qian [54]
	Enterprise age Heterogeneity Network (${AE}_i$)	Age differences matrix	Luo et al. [55]

Innovation Network: Following Su et al. [48], this variable was measured by joint patent applications, a widely used proxy for innovation. Nodes represent firms and edges indicate the frequency of patent collaboration. The adjacency matrix was weighted, with higher values reflecting stronger cooperation.

Supply chain network: Following Sun et al. [49] and Liu et al. [50], a weighted adjacency matrix was constructed using supply chain transaction data sourced from the CSMAR database. Nodes included the listed firms. Edges represent supply relationships, with weights corresponding to the annual transaction frequency between a firm and its supplier/customer (non-zero positive integers for existing ties, 0 for absent ties). The network was treated as undirected, assuming mutual dependence in buyer-supplier relationships. To ensure reproducibility, ties were established based on explicit transaction records in the CSMAR database, with no additional threshold applied. The matrix was constructed in UCINET using “company-supplier/customer” security code pairs, with edge weights aggregated from raw transaction counts.

Executive network: Adapted from Ding et al. [51], this weighted matrix captures interlocking directorates. Nodes represent firms, and edges indicate the number of shared executives (e.g., directors or senior managers holding concurrent positions in both firms). Edge weights correspond to the count of overlapping personnel between firm pairs (non-zero integers for connections, 0 for no shared executives). The network was undirected, as executive ties are symmetric. No minimum threshold was applied. Any firm pair sharing at least one executive was connected, with weights reflecting total shared personnel counts. The matrix was constructed in UCINET using “company-concurrent company” security code pairs, aggregating raw directorship records into edge weights.

Enterprise Scale Similarity: Following Xiang et al. [52], this variable is the ratio of the mean business scale between firms, which is defined as:

$${\mathrm{R}}_{\mathrm{i}\mathrm{j}}=\mathrm{mean}\left({\mathrm{N}}_{\mathrm{i}}\right)/\mathrm{mean}\left({\mathrm{N}}_{\mathrm{j}}\right)$$

(2)

where $N_i$ and $N_j$ represent firm size (total assets). Values range from 0 to 1, with 1 indicating identical scales.

Geographic Proximity: This binary variable indicates whether firms are registered in the same province (1 = yes, 0 = no), adopted based on Cao et al. [53].

Ownership Homogeneity: This binary variable indicates whether firms share the same ownership type (1 = yes, 0 = no), adopted following Zhang and Qian [54].

Enterprise age Heterogeneity Network: This binary network indicates whether a significant age difference exists between two firms (1 = yes, 0 = no), constructed following the approach of Luo et al. [55] by calculating pairwise age differences and applying a threshold set at the sample median.

4. Results

4.1. Network Topology and Structural Characteristics

4.1.1. Overall Network Topology

Inter-enterprise network matrices were constructed for the supply chain, executive, and innovation networks and visualized using Gephi 0.10.1. Network characteristics were analyzed at both global and node levels.

Global network analysis (Figure 2a–c) revealed distinct structural profiles, with visual patterns directly illustrating quantitative metrics and highlighting pivotal actors (see

Table 3. The overall network indicators.

Network	Node	Edges	Network Density	Cluster Coefficient	Average Path Length
SCNs	1569	2061	0.041	0.054	2.846
ENs	4804	5951	0.064	0.365	3.822
INs	1145	1054	0.04	0.020	3.502

). The SCNs (Figure 2a) exhibit a low density (0.041) and a short average path length (2.846), visually manifested as a star-like topology with core manufacturers acting as central hubs, indicating efficient information flow and reflecting the modular architecture of the NEV industry. However, the dependence on a star-like topology poses a systemic risk, as the network may lack the redundancy to compensate for disruptions at key manufacturing hubs. The ENs (Figure 2b) exhibit a higher density (0.064) and clustering coefficient (0.365), corroborated by tightly knit subgroups that foster cohesive decision-making communities. However, the longer average path length (3.822) reflects noticeable gaps between clusters, suggesting the presence of hierarchical governance. In contrast, the innovation network (Figure 2c) displays a dynamic and interconnected architecture, characterized by densely linked hubs that bridge distant nodes. The visual lack of modular boundaries, along with metrics such as structural hole bridging, underscores its role in facilitating rapid knowledge diffusion and technological breakthroughs.

4.1.2. Individual Network Analyses

This study evaluates node-level influence based on three centrality measures: degree, betweenness, and closeness. The top five enterprises ranked across these measures are presented in

Table 4. The top five enterprises in terms of centrality metrics.

Network	Top 5	Degree Centrality		Closeness Centrality		Betweenness Centrality
SCNs	1	WCP	10.526	CFAW	3.331	WCP	14.724
	2	FAW	8.772	CGBC	3.314	CFAW	6.266
	3	FUAO	7.018	WCP	3.316	CGBC	7.707
	4	CFAW	7.018	FUAO	3.333	FUAO	12.813
	5	CCAC	7.018	FAW	3.242	FAW	5.075
ENs	1	FAW	10.526	ZYG	2.995	ZYG	5.827
	2	WCP	8.772	HXAPC	3.008	HXAPC	8.271
	3	WEFU	8.772	CAERI	1.887	CAERI	0.376
	4	HXAPC	7.018	CCAC	3.038	WCP	10.307
	5	SBBP	7.018	ZAPME	3.014	CCAC	2.318
INs	1	CFAW	7.018	CFAW	1.886	CFAW	0.564
	2	CAERI	5.263	ZYG	1.851	CAERI	0.313
	3	KYPC	3.509	FAW	1.851	ZYG	0.188
	4	SSETC	3.509	FAW	1.786	KYPC	0.063
	5	ZYG	3.509	FAW	1.885	SSETC	0.251

Note: In this table, “CFAW” represents China Faw Co., Ltd.,(Changchun, China) whereas “FAW” corresponds specifically to FAW Car Co., Ltd., (Changchun, China) the official entity name during the observation period covered by this study. The full names and abbreviations of other companies mentioned are provided in Table A1.

. Weichai Power Co., Ltd. (WCP)(Weifang, China), occupies a central hub position in Figure 2a, with its network topology reflecting vertical integration strategies. Weichai ensures supply chain stability for key components by controlling both upstream battery suppliers and downstream manufacturers. The figure highlights its unique role as a core connector bridging multiple modules, while its 18% patent share in hydrogen fuel cells (the highest in the industry) further consolidates this position [56].

Zhengzhou Yutong Bus Co., Ltd. (ZYG) (Zhengzhou, China), dominates the central cluster in Figure 2b, with leading closeness (2.995) and betweenness centrality (5.827), reflecting its strategic focus on electric commercial vehicles. As the key interlock within the executive subgroup, Yutong’s board network spans 23 provinces, integrating R&D, market expansion, and policy compliance to rapidly scale charging infrastructure projects [57,58].

China FAW Co., Ltd. (CFAW) (Changchun, China), emerges as the largest hub node in Figure 2c, with its centrality metrics confirming its pivotal role in the NEV innovation ecosystem. This position is driven by two complementary factors: (1) technology leadership through its “Red Flag Innovation Ecosphere,” which integrates over 200 upstream and downstream partners to accelerate solid-state battery industrialization via industry-university-research alliances; and (2) policy–market synergy, as the firm leveraged its State-Owned enterprise status to secure “14th Five-Year” funding while achieving premium market penetration through strategic brand repositioning [59].

These findings highlight how heterogeneous network positions—innovation-centric, supply chain-anchored, and executive-mediated—reflect strategic adaptations to China’s unique policy–market dynamics. This divergence underscores the importance of tailored network strategies in rapidly evolving industries and offers insights for both theoretical development and practical applications.

4.2. QAP Correlation Analysis

Table 5. The QAP correlation analysis result.

Variable	Parenti	Spi	Epi	Ri	Di	Oi	AEi
Parenti	1.000
Spi	0.14 ***	1.000
Epi	0.168 ***	0.297 ***	1.000
Ri	−0.014	0.01	−0.024	1.000
Di	−0.029	−0.006	−0.004	−0.026	1.000
Oi	0.026	0.026	0.079 **	0.022	0.001	1.000
AEi	−0.041	−0.073	−0.031 **	−0.003	0.041	−0.016	1.000

Note: ** p < 0.01; *** p < 0.001.

presents the QAP correlation results, revealing three key patterns in multi-network interactions. SCNs exhibit a strong positive correlation with innovation networks (r = 0.140), indicating that closer supply relationships enhance collaborative innovation. Similarly, ENs exhibit a significant association with innovation networks (r = 0.168), demonstrating how interlocking directorates facilitate knowledge sharing and resource integration. A particularly robust correlation is observed between SCNs and ENs (r = 0.297). This significant interlinkage indicates a substantial overlap between the economic domain of supply chains and the social domain of executive relationships, hinting at a complex embeddedness structure that warrants further investigation.

In contrast, the control variables—geographic proximity (Di), enterprise scale similarity (Ri), and ownership homogeneity (Oi)—exhibit no significant direct correlations with innovation networks (all p < 0.05). These factors may indirectly influence innovation by moderating the effects of core networks and their relationships, a possibility further examined in subsequent regression analyses.

Overall, QAP correlations offer initial evidence of intertwined economic and social networks in innovation, justifying a multi-layer analytical approach. However, these bivariate associations merely indicate linkages—not causal or independent effects. Rigorous hypothesis testing requires the multivariate QAP multiple analysis presented next.

4.3. QAP Regression Analysis

UCINET 6.645 was used to conduct a QAP regression with 5000 permutations on the multi-network matrices of 58 NEV firms.

Table 6. The QAP multiple regression results.

Variable	Model 1	Model 2	Model 3	Model 4
Di	0.190 ***	0.174 ***	0.158 ***	0.152 ***
Ri	0.030 ***	0.030 ***	0.017 ***	0.028 ***
Oi	0.010	0.008	0.002	0.003
AEi	−0.040 *	−0.032	−0.040	−0.031
Spi		0.113 **		0.085 **
Epi			0.127 **	0.103 **
R2	0.039	0.052	0.054	0.061
R2 adjustment	0.038	0.050	0.052	0.059
p	0.000	0.000	0.000	0.000
Observations	3306	3306	3306	3306

Note: * p < 0.05; ** p < 0.01; *** p < 0.001.

summarizes the hierarchical regression outcomes.

In Model 1 (controls only), geographic proximity (β = 0.190) and enterprise scale similarity (β = 0.030) were found to significantly enhance innovation networks, aligning with industrial cluster theory and resource dependence theory. However, ownership homogeneity (β = 0.01) was not significant, likely reflecting the success of China’s mixed-ownership reforms in reducing institutional barriers within the NEV sector [60].

Building on these controls, Model 2 introduces an SCN, revealing strong positive effects (β = 0.115). Closer supply relationships facilitate resource sharing and risk mitigation, as evidenced by CATL’s (Contemporary Amperex Technology Co., Limited) joint R&D alliances with automakers, which shorten battery technology iteration cycles through streamlined material integration [61].

Similarly, Model 3 demonstrates that executive interlocks significantly boost innovation (β = 0.128), consistent with upper echelons theory. Cross-firm ENs enable strategic resource arbitrage, as exemplified by NIO’s (Nio Inc.) rapid resolution of the 2022 chip shortage crisis through coordinated supplier negotiations [62].

After combining both networks in Model 4, the SCN (β = 0.087) and EN (β = 0.103) ties remained significant despite reduced coefficients, indicating positive effects on innovation networks. This provides robust evidence for their complementary, rather than interdependent, roles in facilitating innovation. SCNs provide material foundations for innovation through tangible resource integration (e.g., raw material pooling), while ENs optimize resource allocation via information brokerage and strategic alignment (e.g., cross-sector R&D prioritization).

These results collectively validate a dual-driven framework, whereby economic transactions and social coordination jointly shape innovation ecosystems. These effects highlight the need for balanced network governance strategies in resource-constrained environments.

4.4. Sensitivity Verification

To verify the robustness of the QAP regression results, this study adopts a dual validation strategy [63]. On one hand, we adjust the binarization threshold of the relational matrix by testing the reproducibility of model results under different cutoff criteria (retaining top 10% and 20% of association edges). On the other hand, we vary the number of random permutations, conducting sensitivity analyses with successively set permutation magnitudes of 10,000 and 20,000. The results (see

Table 7. Robustness test of QAP regression results.

Variable	Top 10%	Top 20%	10,000	20,000
Spi	0.091 ** (0.004)	0.010 * (0.04)	0.091 ** (0.003)	0.091 ** (0.003)
Epi	0.217 ** (0.004)	0.049 * (0.05)	0.217 ** (0.004)	0.217 ** (0.004)
Di	0.012 * (0.031)	0.004 * (0.039)	0.012 * (0.031)	0.012 * (0.031)
Ri	0.003 * (0.045)	0.006 * (0.047)	0.003 * (0.045)	0.003 * (0.045)
Oi	0.017 (0.198)	0.017 (0.250)	0.017 (0.198)	0.017 (0.198)
AEi	0.004 (0.090)	0.004 (0.087)	0.004 (0.090)	0.004 (0.090)
R2	0.040	0.030	0.040	0.040
R2 after adjustment	0.039	0.028	0.039	0.039

Note: * p < 0.05; ** p < 0.01.

) demonstrate that across all robustness checks, both the direction and statistical significance of standardized regression coefficients remain stable, indicating that the empirical conclusions exhibit strong robustness to model parameter selection and that the core findings are reliable.

5. Discussion

In this study, a multi-network analysis was employed to uncover the mechanisms driving innovation cooperation in China’s NEV industry. Supply chain, executive, and innovation networks were found to exhibit distinct structural profiles: SCNs exhibited low density but short average path lengths, reflecting efficient resource flows in modular architectures; ENs displayed high density and clustering coefficients, indicating tight-knit decision-making communities; and innovation networks were found to balance intermediate density with high modularity, enabling specialized collaboration clusters. These findings are consistent with recent work by Ge et al. [64], who analyzed Chinese patent collaboration networks found that innovation networks tend to exhibit high modularity and multi-core structures, which facilitate knowledge sharing and coordinated innovation in technology-intensive industries.

QAP regression results reveal that both SCNs and ENs have significant and independent positive effects on innovation networks, even after controlling for each other. These findings support our proposed “resource-decision dual-embeddedness” framework. SCNs reflect the resource structuring mechanism, consistent with the resource-based view, by lowering transaction costs and improving innovation resource allocation, as also highlighted by Gu et al. [65]. In contrast, ENs embody cognitive and social mechanisms, fostering strategic alignment, coordinating technical roadmaps, and reducing uncertainty through executive interlocks and social ties. This aligns with Liu et al. [66], who found that in highly uncertain environments, social networks enhance trust and communication, thereby mitigating collaboration risks.

Having established their independent effects within the dual-embeddedness framework, we next examine how these networks interact in practice to jointly shape innovation outcomes. Executive networks act as an adaptive mechanism that counteracts structural rigidities—such as technological lock-in and path dependence—inherent in tightly-coupled supply chains. While vertical integration within supply chains achieves resource consolidation, it often does so at the expense of strategic flexibility, potentially locking firms into existing technological trajectories and limiting their ability to explore novel paradigms beyond the established value chain. Executive interlocks serve as a crucial counterbalance to this rigidity by infusing the decision-making process with external expertise and diverse perspectives, thereby facilitating exploratory innovation. Moreover, overlapping directorates build latent trust, helping manage collaboration risks like intellectual property disputes more effectively than contracts alone. Together, these networks complement each other: one supplies resource access, the other strategic adaptability—collectively enhancing the formation, durability, and recombinant capacity of inter-firm innovation.

The study finds that the influence of supply chain networks and executive networks on innovation networks remains significant after controlling for factors such as geographic proximity, firm size, ownership type, and firm age. Regarding control variables, geographic proximity and firm size similarity were found to have significant positive effects on innovation networks. These results are consistent with Liu et al. [66], who demonstrated that geographic closeness facilitates tacit knowledge spillovers and reduces coordination costs, while firms of similar size achieve higher levels of strategic alignment and trust, leading to more efficient collaboration. Conversely, ownership type and firm age differences were not significant predictors. This contrasts with findings in Western contexts (e.g., Wagner & Sutter, [67]), where ownership differences often shape collaborative behaviors, highlighting the unique institutional environment of China’s NEV industry.

6. Conclusions, Implications and Limitations

Based on a multi-layer network analysis encompassing supply chain, executive, and innovation networks within China’s new energy vehicle industry, this study reveals the mechanisms through which interconnected networks shape innovation cooperation.

The main findings are as follows: First, all three types of networks exhibit a generally sparse connectivity pattern, with the executive network demonstrating a higher degree of clustering. This indicates its critical bridging role in establishing informal trust relationships between firms and facilitating inter-firm cooperation. This finding is consistent with Mizruchi’s [68] view that executive social ties promote resource coordination, as well as with Zhu et al.’s [69] conclusions on executive relationships within China’s NEV industry. Second, the synergistic effects of supply chain networks and executive networks jointly enhance innovation collaboration. This not only validates the resource-based view proposed by Barney [70], which emphasizes the integration of supply chain resources to improve competitive advantage, but also extends social network theory by clarifying the function of executive interlocks as informal channels for information exchange and strategic coordination. Third, this study proposes a dual embeddedness mechanism that, from a theoretical integration perspective, reveals how formal and informal networks complement each other in driving innovation collaboration. This framework addresses the limitations of prior research (e.g., Hong & Zhou, [71]; Anwar, M. et al., [72]), which examined single networks in isolation, and provides empirical evidence at the network level for the theory of innovation ecosystems.

Based on the empirical findings, this study proposes the following policy implications for optimizing the innovation ecosystem in the NEV industry:

First, given the significant bridging role of executive networks in connecting firms, building trust, and facilitating cooperation, governments and industry associations should promote the establishment of cross-firm executive exchange platforms. Such platforms should not only facilitate information sharing and strategic coordination but also incorporate incentives for executive interlocking, thereby increasing the cooperative density of informal networks, reducing collaboration barriers, and enhancing the overall connectivity of the innovation network. Second, in light of the central role of supply chain networks in resource integration and innovation synergy, policymakers should introduce resource allocation-oriented incentive mechanisms into industrial innovation policies, such as R&D subsidies and tax incentives based on supply chain collaborative effects. By directing more resources to key node firms that can drive innovation across upstream and downstream partners, the resource integration function of supply chains can be fully leveraged to reduce innovation costs and enhance resource sharing. Third, building on the dual embeddedness mechanism proposed in this study, it is recommended to establish a cross-network collaborative governance system that integrates both formal (supply chain) and informal (executive) networks into a unified policy framework. This system should clarify the roles of government, core firms, SMEs, and industry associations in cross-network collaboration. Quantitative indicators such as a Structural Empowerment Index (SEI) and a Relational Bridging Coefficient (RBC) should be introduced to dynamically assess firms’ embeddedness and collaborative performance in innovation networks, thereby providing evidence for policy adjustment. Finally, given the strategic leadership of core firms in promoting innovation collaboration, policies should encourage “anchor firms” to fulfill their innovation ecosystem responsibilities. This can be achieved by establishing joint R&D platforms, promoting intellectual property sharing mechanisms, and collaborating with upstream and downstream partners to tackle common technological challenges—thereby transforming network advantages into collective innovation capacity to overcome key industrial bottlenecks.

This study has limitations that require explicit acknowledgment to contextualize its findings. This study’s findings should be interpreted in light of its focus on formal interlocking directorate networks, which inherently underrepresent the dynamics of China’s SME sector. While this sampling approach validly captures network effects among larger, formally organized enterprises, it necessarily excludes the more prevalent informal networking mechanisms that characterize SME operations. Methodologically, while QAP regression appropriately handles dyadic dependencies in director networks, potential limitations include omitted variable bias (e.g., unobserved regional institutional factors) and confounding effects (e.g., technology diffusion through non-network channels). The identified relationships thus principally reflect dynamics within China’s institutionalized corporate network sphere, with their generalizability to SME contexts requiring verification through alternative methodological approaches.

To further advance the findings of this study and address its limitations, future research can be systematically expanded along the following three directions:

First, deepening dynamic network analysis to identify causal mechanisms. Future studies should combine panel network data with advanced analytical techniques, such as the Separable Temporal Exponential Random Graph Model (STERGM), to dynamically track the co-evolution of supply chain, executive, and innovation networks during key technological transformation periods (e.g., the commercialization stage of next-generation battery technologies). Second, focusing on informal networks to enrich theoretical models in the Chinese context. Subsequent research should pay closer attention to the role of informal social networks—such as industry alliances and technological communities—on which small- and medium-sized enterprises (SMEs) heavily rely. By employing mixed research methods, scholars can explore how informal networks complement or substitute formal institutionalized networks. Third, event-study designs using the 2018 US–China tariff announcements as exogenous shocks, paired with the difference-in-differences (DID) analysis of supply chain rerouting, would quantify how geopolitical disruptions differentially impact firms based on ownership structure and network embeddedness. Together, these approaches would not only mitigate the selection biases of the current study but also transform its limitations into a diagnostic framework for understanding resilience across China’s stratified industrial base.