Can the Digital Economy Really Narrow the Innovation Efficiency Gap Among Cities in China?—A Study from the Perspective of Triple Networks

Abstract

This study investigates how the digital economy empowers urban network intensity to address the dilemma of “low-efficiency lock-in” and to promote high-quality and balanced innovation development. Based on panel data from 264 prefecture-level and above cities in China from 2011 to 2022, the study adopts a multi-network perspective—covering innovation, information, and economic networks—and employs fixed effects and two-stage models to examine the impact and underlying mechanisms of the digital economy on disparities in urban innovation efficiency. The results reveal that the digital economy significantly reduces the gap in innovation efficiency across cities, primarily through the optimization of innovation networks and the strengthening of information networks. Moreover, the economic network positively moderates this relationship, amplifying the digital economy’s narrowing effect on innovation disparities. Threshold model tests indicate a nonlinear influence of the digital economy, showing an initial widening followed by a reduction in innovation efficiency gaps as innovation, information, and economic networks evolve. Heterogeneity analysis suggests that among the various dimensions of the digital economy, only digital industrialization plays a significant role in reducing efficiency disparities, while digital governance, digital infrastructure, industrial digitalization, and data valorization do not yet show statistically significant effects. Furthermore, the digital economy significantly reduces innovation efficiency gaps in southern cities, in regions southeast of the Hu Line, and in large cities, whereas in cities northwest of the Hu Line, digital economy development tends to exacerbate these disparities. This study provides both theoretical support for the coordinated improvement of innovation efficiency driven by the digital economy and practical implications for lagging cities aiming to leverage network effects to catch up in innovation performance.

1. Introduction

The overlapping effects of global value chain restructuring and the rising wave of trade protectionism have significantly increased factor market segmentation and transaction costs across countries, thereby intensifying the divergence in regional innovation efficiency and severely constraining sustainable economic development. For instance, in 2024, the Silicon Valley region accounted for 52% of total venture capital and 39% (Data source: Joint Venture Silicon Valley. 2025 Silicon Valley Index) of unicorn companies (unicorns refer to high-growth startups valued at over USD 1 billion and not yet publicly listed) in the United States, reflecting a spatially polarized pattern of innovation that further exacerbates regional development imbalances [1]. Pinheiro et al. (2025) [2] found that in Europe, advanced technological innovation hubs are primarily concentrated in southern Sweden, southern Germany, southeastern France, southeastern England, the Netherlands, Estonia, and Finland, while regions such as Norway (excluding the Oslo area), Denmark (excluding Copenhagen), northwestern Germany, and parts of Eastern Europe are relatively lagging. As the only country in the world that encompasses all industrial categories defined by the United Nations [3], China is facing a similar dilemma. In 2023, R&D personnel and funding in eastern China were approximately five times (Data source: China City Statistical Yearbook) greater than those in the western region, while the number of patent applications and unicorn companies was six times (Data source: CNIPA) and twenty-five times (Data source: 2023 China Unicorn Enterprise Rankings, China Commercial Industry Research Institute) higher, respectively. These figures indicate that high value-added innovation activities in China remain heavily concentrated in the eastern region. Such spatial imbalances in innovation reinforce a negative feedback mechanism—namely, the “innovation siphoning–low-end lock-in” cycle—through channels such as technology premiums and the selective migration of human capital, further widening regional disparities in innovation efficiency [4,5,6] and constraining the sustainability of economic development [7].

In recent years, the rapid development of the digital economy has emerged as a key driving force behind the sustainable advancement of new economic paradigms. According to The 2024 Digital Economy Report released by the China Academy of Information and Communications Technology (CAICT), the digital economy accounted for 60% (Data source: CAICT, 2024 Digital Economy Report) of global GDP in 2023. Meanwhile, Global Innovation Index 2023 published by the World Intellectual Property Organization (WIPO) highlights how the global R&D race triggered by artificial intelligence (AI) has propelled the innovation pace of the digital economy far beyond that of traditional economic models. Against this backdrop, countries around the world are actively formulating digital economy policy frameworks in an effort to foster new technologies and stimulate innovation momentum [8]. For instance, China has introduced the “Eastern Data, Western Computing” project to optimize national computing power infrastructure; the European Union has leveraged The Digital Markets Act to dismantle data monopoly barriers; and the United States has promoted digital technology R&D through The CHIPS and Science Act, strengthening its semiconductor ecosystem. These policy efforts consistently emphasize the role of data technologies in driving diffusion effects—such as digital infrastructure networks—and inclusiveness effects—such as the accessibility of digital finance—to enhance the knowledge and technology absorption capacity of peripheral innovation regions [9,10], thereby addressing spatial imbalances rooted in traditional growth models [11] and narrowing regional disparities in innovation efficiency. As the world’s largest developing country and a core growth pole of the digital economy, China faces considerable risks if technological collaboration between its eastern and western regions remains insufficient. Such disconnection could lead to fragmented markets and a mismatch between technological advancement and consumption upgrading. Combined with increasing external restrictions, these dynamics may amplify industrial chain vulnerabilities and undermine national goals such as carbon neutrality and common prosperity. Therefore, studying how China’s digital economy reshapes disparities in innovation efficiency is not only crucial for the country to overcome the middle-technology trap and reduce supply chain fracture risks, but also offers valuable practical insights for emerging economies striving for technological catch-up and sustainable development.

Urban networks, formed through physical linkages such as transportation and communication infrastructure, serve as interactive systems that facilitate the flow of production factors—including technology, capital, and population—across cities. In essence, they act as the spatial carriers of real economic activity [12]. Variations in either the number of nodes (cities) or the strength of inter-city connections directly influence the overall scale of urban networks, thereby generating nonlinear impacts on innovation efficiency through network scale effects. Within this framework, the digital economy, characterized by high permeability and near-zero marginal costs, reshapes the expansion pathways of urban network scale and exerts a dual influence on disparities in innovation efficiency [13,14]. On the one hand, the non-rival nature of digital technologies breaks the monopoly barriers surrounding access to information and knowledge [15], reconstructs the flow of technological innovation within innovation networks, redefines data exchange structures in information networks, and transforms factor coordination mechanisms in economic networks. This enables weakly connected entities to bypass traditional “core–periphery” hierarchies and acquire non-redundant technological information through weak ties [16], thereby reducing technology search and trial-and-error costs and facilitating convergence in innovation efficiency. On the other hand, regions functioning as network hubs with stronger linkages are more likely to benefit from the skill-biased effects of the digital economy, attracting high-skilled labor and capital, and accelerating the exchange and transformation of tacit knowledge. These regions, through frequent innovation interactions [17], achieve optimized reallocation of innovation resources, resulting in gains in innovation efficiency. In contrast, entities with weak linkages are more susceptible to falling into a “low-efficiency trap” [18], further exacerbating regional polarization in innovation efficiency.

This study adopts Schumpeter’s theory of innovation as its analytical framework [19], with a focus on breakthrough innovation driven by technological disruption and new combinations of production factors. It shifts the research perspective from the traditional focus on innovation scale disparities to the concept of innovation efficiency disparities—that is, from measuring differences in technological output volume to evaluating differences in the effectiveness of resource input and output. This perspective aims to transcend the limitations of focusing solely on the expansion of innovation scale and instead uncover the structural contradictions in resource allocation. Existing studies suggest that the development of the digital economy can enhance total factor productivity [20], accelerate green innovation outputs [21], and promote disruptive environmental technological innovation [22], thereby confirming its developmental effects. However, these studies have not sufficiently examined how the digital economy affects disparities in innovation efficiency, nor have they explored the role of complex network structures in this process. Some of the literature has investigated the relationship between the digital economy and innovation scale disparities, particularly in terms of its impact on patent counts and the output value of new products [23,24]. Yet, there remains no consensus on whether the digital economy helps narrow innovation scale disparities. Convergence theorists argue that the digital economy can promote technological catch-up in lagging cities by alleviating barriers to factor mobility [25], improving institutional environments [6,26], and accelerating knowledge spillovers [27]. In contrast, polarization theorists contend that the self-reinforcing nature of data factors and the positive feedback of network effects [28] allow technologically leading cities to leverage initial data and resource advantages to exert a siphoning effect on innovation factors [18], thereby intensifying technological decline in peripheral regions [18,29]. Other scholars have identified a dynamic, nonlinear relationship in which the digital economy initially widens but later narrows disparities in innovation scale [30]. Overall, the existing literature has primarily analyzed the digital economy’s influence through the lens of factor allocation and static endowments, with limited attention paid to the role of complex regional network structures in shaping disparities in innovation efficiency.

Existing research on complex networks has primarily focused on the evolution of network topologies [31,32] and the identification of urban network functions [33,34], presenting a structural orientation and a tendency toward static analysis, while lacking integration with other multidimensional factors. Some scholars have examined the relationship between the digital economy and urban networks, with most emphasizing how digital infrastructure can enhance inter-city interaction intensity and spatial organization patterns through virtual linkages [35,36,37]. However, these studies have yet to explore how this process affects disparities in innovation efficiency or to clarify the specific role played by complex networks. For example, Zhou et al. (2017) [38] argue that cultivating central network cities helps optimize the structure of innovation networks, thereby improving urban innovation capacity. Sheng et al. (2020) [39] further suggest that different structural characteristics of urban networks exert heterogeneous impacts on innovation efficiency. Overall, the existing literature has primarily focused on the digital economy’s influence on innovation efficiency and innovation scale disparities, while research on its impact on innovation efficiency disparities remains nearly absent. Moreover, most studies have examined the digital economy’s influence on either network structures or innovation activities in isolation, with few integrating economic, innovation, and information networks within a unified analytical framework. As a result, it is difficult to comprehensively understand the interplay among the digital economy, urban network structures, and disparities in urban innovation efficiency. In this context, adopting a complex network perspective to investigate the impact of the digital economy on innovation efficiency disparities not only reveals the dynamic optimization mechanisms of regional resource allocation driven by data-enabled network topologies but also addresses the “efficiency black box” created by traditional analytical frameworks that overlook variations in node linkage strength. This approach enriches the application of innovation theory in the context of the digital economy. Based on this, the present study seeks to address the following research questions: Can the digital economy help reduce disparities in innovation efficiency? What roles do innovation, information, and economic networks play in this process? How does the effect vary under different configurations of these networks? Are there regional differences in this influence? Answering these questions is crucial for promoting high-quality and coordinated regional innovation. Accordingly, this study utilizes panel data from 264 prefecture-level and above cities in China spanning 2011 to 2022 to empirically examine the impact and mechanisms of the digital economy on urban innovation efficiency disparities from the perspective of complex networks. Furthermore, the study explores the nonlinear effects of the digital economy on innovation efficiency disparities under different levels of development in innovation, information, and economic networks, thereby providing policy recommendations for addressing efficiency imbalances and achieving sustainable growth.

The potential marginal contributions of this study are as follows: First, unlike traditional frameworks that focus on regional attributes and factors, this study adopts a complex network perspective to explore the effects and pathways through which the digital economy influences innovation efficiency disparities under the development of innovation, information, and economic networks. It further identifies the optimization effect of innovation networks, the enhancement effect of information networks, and the moderating effect of economic networks. Second, in contrast to the existing literature that primarily examines disparities in innovation scale, this research focuses on innovation efficiency disparities. Efficiency disparities provide a more direct reflection of the capacity for resource allocation and collaborative effectiveness across regions, especially in contexts of tightening resource constraints. In such settings, pursuing convergence through efficiency is more sustainable than merely expanding factor inputs. Third, this study investigates the dynamic characteristics of the digital economy’s impact on innovation efficiency disparities at different stages of complex network development, with the aim of uncovering practical approaches to strengthen the convergence effects of the digital economy.

The structure of the following content is as follows: Section 2 provides a theoretical analysis. Section 3 presents the model and identification strategy. Section 4 examines the impact and mechanisms through which the digital economy influences innovation efficiency disparities from the perspective of complex networks, along with a series of robustness tests. Section 5 investigates the nonlinear effects of the digital economy on innovation efficiency disparities under the dynamic development of complex networks, followed by a further heterogeneity test. Section 6 concludes the study and offers relevant policy recommendations.

2. Theoretical Analysis

2.1. Direct Impact Mechanism

The rapid development of the digital economy not only facilitates the acceleration of innovation elements such as capital, technology, and information across spatial and temporal distances but also catalyzes a significant leap in the scale of innovation nodes through network effects, thereby influencing disparities in innovation efficiency. First, according to Metcalfe’s Law, the development of the digital economy promotes the dual expansion of innovation nodes in terms of quantity and scale, resulting in exponential growth in network value. This growth enables efficient communication and interaction among diverse innovation agents in different spatial and temporal contexts, thereby promoting multi-agent innovation collaboration. It also eliminates the hierarchical barriers of traditional innovation systems [40], alleviates information silos and administrative constraints, reduces intercity endowment gaps, and accelerates the convergence of innovation efficiency. Second, the network path compression effects caused by digital technologies such as cloud computing, the Internet of Things (IoT), big data, and artificial intelligence break the traditional constraints of geographic proximity, allowing implicit knowledge such as technology standards and R&D experiences to diffuse to low-linkage entities at lower friction costs [41], facilitating the aggregation and resonance of tacit knowledge across different spatial and temporal contexts, thus narrowing urban innovation efficiency gaps. Third, relying on algorithmic foundations and digital platforms such as open-source communities [42], the digital economy reduces the search and conversion costs for innovation agents to access cutting-edge technologies. The modular knowledge-sharing system in open-source communities [43], which maintains weak ties between different modules, effectively enhances the information intermediary position of low-linkage entities, promoting the diffusion of innovation potential from high-density clusters to long-tail regions, and accelerating the innovation catch-up of low-linkage entities [44,45]. Finally, the digital economy is gradually driving the transformation of corporate organizational models from traditional hierarchical vertical systems to flatter organizational structures, breaking down clear organizational boundaries and forming new types of corporate organizations [46]. This new organizational structure promotes cross-departmental, cross-module, and cross-functional communication within enterprises, reducing communication costs [47], improving management efficiency, and enabling companies to focus more on R&D activities, thus enhancing innovation efficiency and narrowing the innovation efficiency gap with advanced regions. Based on these observations, this study proposes the following hypothesis:

H1: 

The development of the digital economy can significantly reduce urban innovation efficiency disparities.

2.2. Indirect Impact Mechanism

2.2.1. Innovation Network Optimization Effect

In the context of the digital economy, urban spatial forms are undergoing a paradigm shift from “place-based space” to “flow space”. Based on the flow space theory [48], the functional positioning of cities has evolved into a critical node within the innovation network. In this context, the core of the innovation network lies in the continuous cross-border interaction between multiple urban stakeholders (such as enterprises, research institutions, universities, and government agencies), establishing “global pipelines” to acquire novel knowledge and advanced technologies beyond their local regions [49]. Through this cross-domain linkage mechanism, innovation elements such as knowledge and technology can flow and appreciate more rapidly, and innovation outcomes or outputs continue to emerge and diffuse, ultimately enhancing urban innovation efficiency [50]. On one hand, digital technologies, remote operation systems, and information distribution functions can break the geographical and administrative boundaries between cities, even leading to the phenomenon of “distance death” [51], accelerating the efficient and free movement of elements such as capital, labor, and technology [52]. This reduces the costs of searching, negotiating, and supervising cross-temporal and cross-spatial collaborations among network participants [53], helping innovation agents achieve immediate resource sharing and complementarity of advantages, thus facilitating the exchange and interaction of knowledge and technology within the network. On the other hand, the digital transformation of enterprises has led to the gradual shift in specialized knowledge, cultural products, and consulting services to a digital transaction model supported by cloud platforms. The previous geographical agglomeration has evolved into a virtual agglomeration centered around real-time data exchange, linking a broader range of innovation agents, thus enhancing the heterogeneity and connectivity of knowledge within the innovation network [54,55]. The strengthened network links inevitably improve the ability of innovation agents to access knowledge and technology, thereby enhancing urban innovation efficiency [56] and narrowing the innovation efficiency gap between cities. Moreover, digital infrastructure, through sensor devices, builds important channels for data transmission and technology transfer, breaking the separation between urban nodes and enabling cross-entity data and knowledge exchange and sharing. The “data productivity” effect brought about by massive amounts of knowledge can help innovation agents gain deep insights into the development prospects and technology cycles of their partners, reducing the likelihood of resource misallocation [57,58], and helping establish more stable external cooperation relationships [59], thus avoiding the conversion costs associated with seeking new partners, ultimately facilitating the catch-up of urban innovation efficiency.

It is important to note that cities occupy different positions within the innovation network structure, and their connections and interactions with other members of the network are distinctive. This, to some extent, directly influences the degree to which they can access and benefit from resources such as technology and knowledge [60]. Generally, cities with lower link strength within their innovation network face limited capacity to utilize, absorb, and digest heterogeneous knowledge. In such cases, the rapid development of the digital economy may exacerbate the innovation efficiency gap between cities. Specifically, the high pervasiveness of digital technologies provides ample momentum for the flow of innovation elements, accelerating the formation of an “inflow effect” and helping cities with stronger innovation linkages to enhance the diversity and specialization of their innovation elements, thus promoting the transition toward more intelligent and intensive production models. Furthermore, leveraging the scale advantages of data elements and utilizing the “high-multiplier” effect continuously improves value-creation capabilities [61], resulting in a “Matthew Effect”. In contrast, regions with underdeveloped innovation network linkages often struggle to achieve the timely reallocation of knowledge and technology across time and space [18,29], making it difficult for them to benefit from the innovation efficiency boost provided by the digital economy. This exacerbates the innovation disadvantages of regions with weaker network linkages and widens the innovation efficiency gap between cities. Based on this, the following hypotheses are proposed:

H2a: 

The digital economy influences the innovation efficiency gap by optimizing urban innovation networks.

H2b: 

As the innovation network develops, the impact of the digital economy on the innovation efficiency gap exhibits a nonlinear characteristic, initially expanding and then narrowing.

2.2.2. Information Network Driving Effect

The information network is a multi-level, dynamic system of urban node connections, supported by information and communication technology, formed through the cross-regional flow and interaction of information [62]. First, as the physical carrier of data elements, the information network supports the efficient circulation of data, a core production element in the digital economy. As the digital economy develops, it can effectively broaden the scope and depth of data elements, accelerate the development of information diversity, and reduce the risk of information redundancy, thus enhancing urban innovation efficiency [63] and narrowing the innovation efficiency gap between cities. Second, digital and intelligent technologies have both informational efficiency and collaborative functions. The widespread development of these technologies not only directly reduces the costs and time involved in the flow, communication, and exchange of information, but also reduces information asymmetry, improves the accessibility and acquisition of information resources, and accelerates the formation of information transmission cost effects [64]. These technologies can also help cities precisely match resources and partners that meet their production and innovation needs on the scale of information networks, optimizing resource allocation in key research and development stages, enhancing cross-city resource collaboration and sharing, and boosting urban innovation efficiency. Finally, the information network, supported by digital infrastructure, can guide labor, capital, and other production factors to cities with higher marginal returns, thus avoiding resource misallocation and enhancing the innovation efficiency of economically lagging cities [65,66]. Moreover, due to the inherent characteristics of information elements—such as “non-competitiveness”, “non-exclusivity”, and “low-cost infinite replicability”—multiple temporal and spatial combinations of factors can be formed and reused, promoting a multiplier effect in value creation [67], which helps relatively underdeveloped cities achieve innovation efficiency and “catch up” more rapidly.

Based on urban network theory, a city’s status and function depend on its connectivity or structural position within the network. The development of the digital economy further amplifies the spatial differentiation driven by network effects. Social network analysis shows that a city’s link strength (degree centrality) in the information network directly reflects its position in the network and its ability to access resources [68,69]. When a city has low link strength in the information network, its information flow channels are restricted, and cross-regional information circulation is limited. In such cases, the development of the digital economy may not bridge the innovation efficiency gap between cities; instead, it may exacerbate the problem due to barriers to data sharing and disruptions in information collaboration channels, making it difficult for elements to break through the “information silo” constraints. Digital technologies are more likely to be siphoned by regions with better infrastructure, leading to the concentration of capital, labor, and other production factors in cities with higher information network connectivity [70]. In such circumstances, the development of the digital economy could accelerate the loss of resources and the degradation of collaborative capacity in cities with low link strength in the information network, putting them at a competitive disadvantage in terms of innovation efficiency. This would ultimately strengthen the Matthew effect in regional innovation efficiency, deepening the innovation efficiency gap between cities. Based on this, the following hypotheses are proposed:

H3a: 

The digital economy influences the innovation efficiency gap by driving the development of urban information networks.

H3b: 

As the information network develops, the impact of the digital economy on the innovation efficiency gap exhibits a nonlinear characteristic, initially expanding and then narrowing.

2.3. Economic Network Structural Moderating Mechanism

The economic network structure, as the most fundamental link between cities, also known as the spatial interaction measure, is commonly used to assess the strength of economic connections between cities [62], directly reflecting a city’s capacity to absorb, utilize, and transform resources in economic exchanges. Generally, the more developed the economic network structure, the stronger the flow of goods, services, labor, and capital between cities, enhancing their ability to utilize and process resources, thereby strengthening the positive impact of digital economy development on reducing urban innovation efficiency disparities. On one hand, as the physical medium for advancing the digital economy, the improvement of digital infrastructure provides a crucial technological opportunity window for the enhancement of innovation efficiency in lagging cities. It not only reduces the transaction costs of acquiring external knowledge and technology for these cities, but also deepens inter-city connectivity, improving resource allocation efficiency [71], thus facilitating the innovation efficiency catch-up in lagging cities. Furthermore, actors within these cities can more keenly capture subtle changes in technological developments and market demands through the economic network, promptly establishing economic links with leading cities, facilitating the flow of innovation elements such as technology, information, and knowledge across cities, and significantly enhancing the diffusion of knowledge and technology, ultimately driving improvements in innovation efficiency. On the other hand, the digital economy’s inherent advantages in information technologies provide a fertile ground for enhancing urban innovation efficiency by reconstructing enterprise product systems, production processes, organizational structures, and business models through the combination of information, computation, communication, and linking technologies. This leads to value multiplier effects [72] and triggers improvements in enterprise innovation efficiency. In this context, cities with a relatively lagging economy, based on the stable and close relationships formed by various actors in the economic network through connections in goods, services, labor, or resources, can quickly respond to external changes in high-quality enterprise innovation models and characteristics. By facilitating the flow, sharing, and absorption of innovation elements, these cities can enhance knowledge exchange efficiency within the network, thereby driving innovation efficiency catch-up in relatively underdeveloped regions.

However, the degree of embeddedness or position within the economic network may influence the impact of the digital economy on innovation efficiency disparities. Specifically, marginal or relatively underdeveloped cities typically face disadvantages in infrastructure construction, capital accumulation, economic scale, and population concentration, leading to weaker economic interactions with other cities [73]. In such cities, micro-level actors struggle to access high-quality labor, capital, and material resources from external sources through “scale borrowing” [74]. Due to insufficient spatial interaction capacity, they find it difficult to overcome local resource constraints, weakening their economic linkages and collaborative capabilities with other cities [25]. In this context, the penetration of the digital economy may trigger resource siphoning and path dependence, suppressing innovation momentum and further expanding innovation efficiency disparities between cities. Based on this, the paper proposes the following hypotheses:

H4a: 

Economic network embeddedness enhances the role of the digital economy in reducing urban innovation efficiency disparities.

H4b: 

With the development of the economic network, the impact of the digital economy on innovation efficiency disparities exhibits a nonlinear pattern, initially widening and then narrowing.

3. Research Design

3.1. Variable Selection

3.1.1. Dependent Variable: Urban Innovation Efficiency Gap

The innovation efficiency gap between cities refers to the disparity in the transformation of regional innovation factor combinations into knowledge output and economic effectiveness, constrained by the given technological possibility frontier. It reflects the dynamic evolution of technological catch-up effects and frontier mobility effects [75]. Following the approach of Zhang and Yu (2024) [76], this study measures the city innovation efficiency gap using the deviation of each city’s innovation efficiency. Since a ratio closer to 1 indicates a smaller gap, we subtract 1 from it and take the absolute value to eliminate directional bias and standardize the gap measurement [77]. Specifically, given that the innovation we study is based on Schumpeter’s innovation theory, which focuses on technological breakthroughs and new factor combinations, innovation is formed through research and development (R&D) investment, leading to knowledge accumulation that is ultimately converted into market value [19]. Based on Edquist’s innovation systems analysis framework, we construct three input indicators—capital investment, human resources, and the innovation environment—corresponding to the resource supply, knowledge base, and institutional environment in Edquist’s framework. We also create two output dimensions—scientific and technological output and social system effectiveness output—to measure knowledge creation, the diffusion of new products, and the enhancement of social effectiveness resulting from innovative products [78,79]. Among these, scientific and technological output corresponds to the creation of new knowledge combinations and the marketization of new products as per Schumpeter’s innovation theory. Further, we extend Edquist’s innovation system framework by incorporating the impact of energy efficiency into the effectiveness of the innovation system [80], in response to the United Nations Sustainable Development Goals (SDGs) (UL Standards & Engagement. (25 September 2015). The UN Sustainable Development Goals and UL Standards & Engagement: A report. https://ulse.org/data-insights/un-sdgs-and-ul-standards-engagement-report (accessed on 20 April 2025)). Finally, following the works of Cruz-Cázares et al. (2013) [81], Bai and Jiang (2015) [82], Han et al. (2019) [83], Xie et al. (2020) [84], Liu et al. (2025) [85], and Mukhtar et al. (2025) [80], we construct an indicator system (see

Table 1. Index system of urban innovation efficiency.

Primary Indicators	Secondary Indicators	Tertiary Indicators	Quaternary Indicators
Innovation Input	Capital Input	Government Innovation Capital Investment	Proportion of Government Education Expenditure to GDP
			Proportion of Government Science Expenditure to Fiscal Expenditure
		Corporate Innovation Capital Investment	Proportion of R&D Internal Expenditure in Large-Scale Industrial Enterprises to GDP
	Human Capital Input	Educational Innovation Input	Number of Full-Time Faculty Members in Regular Higher Education Institutions
			Number of Higher Education Students per 100,000 People
		Research Innovation Input	Full-Time Equivalent R&D Personnel
			Number of Employees in Scientific Research, Technical Services, and Geological Survey Industries
	Innovation Environment	Innovation Platforms	Number of Regular Higher Education Institutions
			Number of Research and Development Institutions
		Economic Foundation	Gross Domestic Product per Capita
		Talent Environment	Average Years of Education
Innovation Output	Scientific and Technological Innovation Output	Knowledge and Scientific Research Innovation Output	Number of Invention Patents Granted
			Number of Scientific Papers Published by School
		Product Innovation Output	Sales Revenue from New Products
			Scale of High-Tech Industry Development
	Social System Effectiveness	Economic Efficiency	Labor Productivity
		Energy Efficiency	Comprehensive Energy Consumption Output Rate

). Based on this, the DEA-Malmquist non-parametric model is introduced to calculate city innovation efficiency. This choice is made because, compared to traditional DEA and stochastic frontier methods, the DEA-Malmquist non-parametric model does not require setting specific functional forms, avoiding errors caused by functional specification biases, and more accurately reflecting dynamic changes in relative efficiency. Finally, based on innovation efficiency, the innovation efficiency gap between cities is measured using the innovation efficiency deviation of each city. The calculation formula is as follows:

$$Ga{p}_{i,t}=|{\displaystyle \frac{Invtf{p}_{i,t}}{\overline{Invtf{p}_t}}}-1|$$

(1)

In Equation (1), i represents a city, t represents time, $Ga{p}_{i,t}$ denotes the innovation efficiency gap of city i in year t, Invtfp_i,t represents the innovation efficiency of city i in year t, and $\overline{Invtf{p}_{i,t}}$ denotes the mean innovation efficiency in year t.

3.1.2. Core Independent Variable: Digital Economy Development Level

Considering the availability, continuity, and comparability of relevant data at the city level, and drawing on the research of Zhao et al. (2020) [86], this study constructs a digital economy development level indicator system based on the definition of digital economy. The system includes five dimensions: digital infrastructure, digital governance, data value, digital industrialization, and industrial digitalization, selecting 26 representative secondary indicators (see

Table 2. Indicator system for the development level of digital economy.

Primary Indicators	Secondary Indicators	Tertiary Indicators
Digital Infrastructure	Traditional Infrastructure	Per Capita Postal Business Volume
	Digital Infrastructure	Optical Cable Density
		Internet Broadband Users per 100 People
		Mobile Switch Capacity per 100 People
		Digital Platforms
Digital Industrialization	Manufacturing of Computers, Communication, and Other Electronic Equipment	Number of Listed Companies in the Manufacturing of Computers, Communication, and Other Electronic Equipment
	Telecommunications Industry	Per Capita Telecommunications Business Volume
		Number of Mobile Phone Users per 100 People
	Broadcasting, Television, and Satellite Transmission Services	Number of Listed Companies in Telecommunications, Broadcasting, and Satellite Transmission Services
	Internet, Software, and Information Technology Services	Number of Listed Internet and Related Services Companies
		Number of Listed Software and Information Services Companies
	Digital Industry Vitality	Proportion of Employees in Computer Software
		Number of Unicorn Enterprises in the Digital Economy
Industry Digitalization	Smart Agriculture	Proportion of Administrative Villages with Internet Broadband Access
		Value Added in Agriculture, Forestry, Animal Husbandry, and Fishery
		Rural Electricity Consumption
	Smart Industry	Electricity Consumption in Industrial Production
		Number of Listed Smart Industrial Companies
		Industrial Robot Penetration Rate
	Digital Finance	Digital Inclusive Finance Coverage
		Digital Inclusive Finance Usage Depth
		Digital Inclusive Finance Digitalization Degree
	Digital Trade	Number of Listed E-Commerce Companies
Digital Governance	Government Awareness of Digital Economy Development	Frequency Statistics of Key Terms in Government Work Reports
	Proportion of Government Spending on Science and Technology	Expenditure on Scientific and Technological Affairs in Fiscal Budget/General Fiscal Budget Expenditure
Data Value	Data Trading Regulations	Number of Data Trading Centers

). This aims to comprehensively assess the development status of urban digital economies, and uses the entropy weight method (EW) to objectively assign weights to the indicators, denoted as Dige. The entropy weight method, as a weight determination method based on the principle of information entropy, effectively reflects the degree of dispersion of each indicator in the dataset, reduces the interference of subjective judgment, eliminates multicollinearity problems, and ensures the objectivity and accuracy of the evaluation results. Among them, urban digital finance is measured using the Peking University Digital Inclusive Finance Index (2011–2022) released by the Peking University Digital Finance Research Center. Since this index started being published in 2011, and some secondary indicator data are only updated until 2022, the time frame of this study’s data is defined as 2011 to 2022. Additionally, when measuring the digital governance dimension’s government awareness of digital economy development, this study refers to the research of Xiao et al. (2022) [87] and uses Python 3.12 programming to collect and analyze the frequency of terms related to the digital economy in municipal government work reports, maximizing the quantification of government support for digital economy policies.

3.1.3. Mechanism Variables

Urban Innovation Network

Based on the existing literature, the measurement of urban innovation networks primarily relies on social network analysis and complex network theory, quantifying the connections between nodes to reveal network characteristics. Micro-level patent collaboration data, due to its traceability, availability, and direct reflection of technological implementation, is widely used to measure urban innovation networks. In this regard, this study utilizes the incoPat patent database and Java distributed crawling technology to extract the address information of patent assignees. When the assignees of the same patent are located in two or more different cities, it is considered as a patent collaboration. An innovation association strength matrix between cities is constructed based on the number of collaborations. The absolute degree centrality is used to measure the innovation network strength of each city, and normalization is performed to eliminate differences in network size. This method captures the explicit connections of technological collaboration between cities through the geographical affiliation of micro-level patent assignees, highlighting the spatial topology characteristics of innovation collaboration across administrative boundaries.

The specific calculation formula is as follows:

$$Inven{t}_i={\displaystyle \frac{{\displaystyle \sum _{j=1}^{N}In{v}_{i,t}}}{N-1}}i\ne j$$

(2)

where $Inven{t}_{i,j}$ represents the centrality index of city i in innovation collaboration; N is the total number of nodes (cities) in the urban innovation network. The term $In{v}_{i,j}$ represents the number of patent collaborations between city i and city j.

Urban Information Network

The urban information network connects various city nodes through information flow, reflecting the intensity of information exchange and spatial structural relationships between regions [48,88]. Unlike traditional “local space”, the information network emphasizes the “flowing space” attribute, focusing on the rapid movement and real-time transmission of information. Currently, Baidu (Baidu.com) is the largest Chinese search engine globally, with a monthly active user industry penetration rate of 96% (Data source: 2023 China Search Engine Market Research Report). Therefore, using the Baidu Index to represent information flow can more intuitively reflect the level of attention paid by internet users in a city to a specific keyword, thus quantifying the intercity information flow. This has become a common indicator in academic research for measuring the strength of intercity information networks [89,90]. Based on this, this study compiles the search volume data for one city to search for another city’s name across the country and integrates it into an intercity information association matrix. By using the relative degree centrality indicator from social network analysis, the linkage strength of each city in the information network is calculated. The specific calculation formula is as follows:

$$Interne{t}_i={\displaystyle \frac{{\displaystyle \sum _{j=1}^N{I}_{i,j}}}{N-1}}i\ne j$$

(3)

where $Interne{t}_{i,j}$ represents the centrality index of city i in the information network; N is the total number of nodes (cities) in the urban information network. The term $I_{i,j}$ represents the Baidu Index of city i searching for city j.

3.1.4. The Moderating Variable: Urban Economic Network

The urban economic network refers to the network structure formed by the flow of economic factors such as labor and capital between cities, creating mutual connections and dependencies [91]. The strength of the economic network reflects the factor linkages and interactions between cities based on economic scale and geographical proximity, demonstrating the circulation of factors and resource complementarity between cities. Drawing on the work of Dou (2023) [92], this study constructs a modified gravity model to measure the strength of the urban network, which is proportional to the city’s GDP and population and inversely proportional to the geographical distance [93]. The calculation of the indicator is as follows:

$$Econe{t}_i={\displaystyle \frac{\sqrt{P_{i}GD{P}_i}\times \sqrt{P_{j}GD{P}_j}}{d_{i,j}^2}}$$

(4)

where $Econe{t}_{i,j}$ represents the network strength between city i and city j, $P_i$ and $P_j$ are the populations of city i and city j, $GD{P}_i$ and $GD{P}_j$ are the gross domestic products of city i and city j, and $d_{i,j}$ is the geographical distance between the two cities, measured in terms of latitude and longitude.

3.1.5. Control Variables

Based on the existing literature, this study includes relevant control variables to address omitted variable bias: (1) Economic Development Volume (LnDev): The real GDP is obtained by deflating the GDP price index of each province with 2011 as the base year, and the logarithm is taken to avoid the influence of heteroscedasticity. (2) Industrial Structure Rationalization (Indus): This is calculated by multiplying the output value ratios of the primary, secondary, and tertiary industries to GDP by the corresponding industry information. (3) Population Size (LnPeo): The natural logarithm of the city’s resident population for the given year is used. (4) Fiscal Decentralization (Findp): This is measured by the ratio of local government general budgetary revenue to general budgetary expenditure. (5) Financial Endowment Disparity (Fina): This is the ratio of the city’s year-end financial institution deposit balance to the year-end loan balance of financial institutions. (6) Foreign Direct Investment Level (Fdi): This is calculated as the ratio of the actual foreign investment utilized in the given year to GDP. (7) Technology Expenditure Level (Tech): This is the ratio of the city’s science expenditure to GDP for the given year. At the same time, this study also controls for city fixed effects and time fixed effects, aiming to eliminate the impact of uncontrollable factors such as economic fluctuations, policy regulations, and cultural and historical factors on urban innovation, thus improving the accuracy of the research results.

3.2. Data Sources and Descriptive Statistical Analysis

The data for urban innovation efficiency and digital economy indicators are sourced from the China National Intellectual Property Administration (CNIPA), CNRDS, CSMAR Database, the China Urban Statistical Yearbook, Provincial Statistical Yearbooks, and municipal statistical bulletins. The frequency of government work report terms was obtained through text segmentation and statistical analysis using Python. The Digital Inclusive Finance Index data comes from the Peking University Digital Finance Research Center, while the data on urban innovation networks is sourced from the incoPat Global Patent Database. Urban information network data are derived from Baidu Search Index. For missing values, linear interpolation was used for imputation. However, it is important to note that linear interpolation assumes linear changes between adjacent data points, which may overlook nonlinear regions or complex patterns in the data. To verify the reliability of the imputation, a Hausman test was conducted to determine if the imputation altered the statistical properties of the data. The test result yielded a p-value of 1 (greater than 0.05), indicating that the null hypothesis was accepted, meaning the imputation did not significantly alter the model structure. This data imputation method is thus deemed reliable for this study.

Table 3. Descriptive statistics.

Variable	N	Mean	SD	Min	Max
Gap	3168	0.2410	0.2050	0.00700	0.9170
Dige	3168	0.0770	0.0790	0.0200	0.4460
LnDev	3168	16.7500	0.9020	14.3900	19.9200
Indus	3168	0.2900	0.2070	−0.0430	1.2410
LnPeo	3168	5.9450	0.660	3.9910	8.0750
Findp	3168	0.4620	0.2150	0.0700	1.5410
Fina	3168	1.5490	0.5330	0.1780	16.7300
Fdi	3168	0.0170	0.0170	0.0000	0.1990
Tech	3168	0.0290	0.0270	0.0010	0.4150
LnEconet	3168	17.1200	1.3880	12.5900	22.8400
Internet	3168	0.4930	0.3670	0.0270	2.8370
Invnet	3168	0.0710	0.1020	0.0000	0.9160

reports the descriptive statistics of the variables, and following classical statistical theory, this method effectively identifies data distribution characteristics and potential outliers [94], providing foundational tests for subsequent econometric models.

To visually illustrate the regional innovation efficiency gap, this study utilizes ArcGIS 10.8 software, selecting the urban innovation efficiency data for 2011 and 2022 as sample data to create spatial trend maps of urban innovation efficiency (see Figure 1). Based on the trend surface analysis, it can be observed that, both in 2011 and 2022, the spatial distribution of urban innovation efficiency exhibits an unbalanced pattern, with higher efficiency in the south and lower efficiency in the north, as well as higher efficiency in the east and lower efficiency in the west. Furthermore, over time, the gap in urban innovation efficiency between the east and west, as well as the north and south, shows a gradual expansion.

3.3. Model Construction

To examine the impact of the digital economy on the disparity in urban innovation efficiency, the following benchmark model is constructed:

$$Ga{p}_{i,t}={\alpha }_0+{\alpha }_{1}Dig{e}_{i.t}+{\phi }_j{\displaystyle \sum _{j=1}^n{X}_{j,it}+{\mu }_i+{\delta }_t+{\epsilon }_{i,t}}$$

(5)

In this model, the subscripts i and t represent the region (city) and time (year), respectively. The dependent variable $Ga{p}_{i,t}$ represents the innovation efficiency gap of city i in year t, while the core explanatory variable $Dig{e}_{i,t}$ denotes the level of digital economy development in city i in year t. The coefficient ${\alpha }_1$ reflects the impact of changes in the digital economy on the innovation efficiency gap of the city.

Based on the theoretical analysis above, to verify the potential transmission paths of digital economy’s impact on the innovation efficiency gap of cities from a network perspective, we adopt a two-stage econometric model as proposed by Bertrand and Mullainathan (2001) [95]. Equations (6)–(8) collectively form the mechanism verification model for the impact of digital economy on the innovation efficiency gap of cities. Model (6) represents the first-stage regression, while model (8) is the second-stage regression.

$$M_{i,t}={\nu }_0+{\nu }_{1}Dig{e}_{i.t}+{\nu }_j{\displaystyle \sum _{j=1}^n{X}_{j,it}+{\mu }_i+{\delta }_t+{\epsilon }_{i,t}}$$

(6)

In this model, the dependent variable $M_{i,j}$ represents the mechanism variables (the effects of urban innovation network optimization and the effects of urban information network driving), which capture the impact of the digital economy on these mechanism variables. Based on model (7), a new estimated variable is generated.

$$\widehat{M_{i,t}}=\widehat{{\nu }_1}Dig{e}_{i,t}$$

(7)

According to Equation (8), the estimated value of the change in the mechanism variables caused by changes in the digital economy can be calculated. This estimated value is used to replace the core explanatory variable, digital economy, in the baseline model, resulting in the transmission model of how the mechanism variables affect the urban innovation efficiency gap:

$$Ga{p}_{i,t}=\widetilde{{\alpha }_0}+\kappa \widehat{M_{i,t}}+\widetilde{{\alpha }_1}{\displaystyle \sum _{j=1}^n{X}_{j,it}}+{\mu }_i+{\delta }_t+{\epsilon }_{i,t}$$

(8)

In model (9), the core coefficient $\kappa$ represents the impact of the change in the mechanism variables on the urban innovation efficiency gap. Since the construction of the mechanism variables in model (9) originates from model (8), $\kappa \widehat{M_{i,t}}$ can also be expressed as

$$\kappa \widehat{M_{i,t}}=\kappa \widehat{{\nu }_1}Dig{e}_{i,t}$$

(9)

In this model, the intuitive meaning of the $\kappa \widehat{M_{i,t}}$ is how digital economy influences the urban innovation efficiency gap through the mechanism variables, and accurately identifies the transmission channels through which the digital economy affects the urban innovation efficiency gap.

Furthermore, considering that urban economic networks can optimize the allocation of factor resources and the environment for collaboration through the circulation and connection of economic elements, thus regulating the intensity of the digital economy’s effect on the urban innovation efficiency gap, the following moderating effect model is constructed to test the macro-regulation effect of urban economic networks. The model is as follows:

$$Ga{p}_{i,t}={\beta }_0+{\beta }_{1}Dig{e}_{i,t}+{\beta }_{2}Dig{e}_{i,t}\times LnEcone{t}_{i,t}+{\beta }_{3}LnEcone{t}_{i,t}+{\varphi }_j{\displaystyle \sum _{j=1}^n{X}_{j,it}}+{\mu }_i+{\delta }_t+{\epsilon }_{i,t}$$

(10)

In the dynamic evolution of urban network structures, the polarization effect and network externalities induced by such structures may result in nonlinear impacts of the digital economy on the urban innovation efficiency gap under different network development environments. Therefore, this study sets urban innovation networks, urban information networks, and urban economic networks as threshold variables and constructs the following single-threshold effect model. Based on the results of the sample data tests, the model will be expanded into a dual-threshold or triple-threshold model if necessary.

$$Ga{p}_{i,t}={\gamma }_0+{\gamma }_{1}Dig{e}_{i,t}\times I(T{h}_{i,t}\le {\theta }_1)+{\gamma }_{2}Dig{e}_{i,t}\times I(T{h}_{i,t}>{\theta }_1)+{\gamma }_j{\displaystyle \sum _{j=1}^n{X}_{j,it}}+{\mu }_i+{\delta }_t+{\epsilon }_{i,t}$$

(11)

In this context, the threshold value is represented by $\theta$, and the indicator function $I( )$ is used to denote the threshold effect, with the coefficient ${\gamma }_i$ to be estimated.

4. Empirical Testing

4.1. Baseline Regression

Table 4. Baseline regression.

	RE		FE
	(1)	(2)	(3)	(4)
Dige	−0.2147 ***	0.0812	−0.2872 *	−0.3072 *
	(0.0706)	(0.1017)	(0.1643)	(0.1772)
LnDev		−0.0259 **		0.0004
		(0.0126)		(0.0266)
Indus		−0.0406		0.0227
		(0.0268)		(0.0375)
LnPeo		−0.0251 *		0.0016
		(0.0144)		(0.0638)
Findp		0.0514		0.0498
		(0.0315)		(0.0626)
Fina		0.0067		−0.0028
		(0.0087)		(0.0104)
Fdi		−0.5705 *		−0.6762 *
		(0.2957)		(0.3773)
Tech		−0.1963		0.1341
		(0.1938)		(0.2409)
Constant	0.2572 ***	0.8098 ***	0.2628 ***	0.2305
	(0.0084)	(0.1630)	(0.0130)	(0.4868)
R2	0.0031	0.0316	0.2604	0.2615
Year FE	NO	NO	YES	YES
City FE	NO	NO	YES	YES
N	3168	3168	3168	3168

Notes: *, **, *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively; the numbers in parentheses represent the standard errors.

reports the results of the random effects and fixed effects models. Since the Hausman test rejects the null hypothesis (random effects model) and, based on the R² value, the fit of columns (3) and (4) is clearly superior to that of columns (1) and (2), this study primarily focuses on the regression results of the fixed effects model. After incorporating both time and city fixed effects, the regression results are significantly negative at the 10% level, regardless of whether control variables are included. This indicates that the development of the digital economy can significantly reduce the urban innovation efficiency gap. After controlling for variables, each 1% increase in digital economy development leads to a significant reduction of 0.307 percentage points in the urban innovation efficiency gap. This also validates Hypothesis H1.

4.2. Robustness Check

Given that the double fixed effects OLS model relies on the assumption of homoscedasticity of the error terms, while in reality, there are significant differences between cities in terms of economic development levels, resource endowments, etc., which can lead to heteroscedasticity and thus underestimate standard errors, potentially affecting significance tests. Additionally, the presence of extreme values in the sample and the impact of the COVID-19 pandemic on the flow of factors and city networks may interfere with and bias the regression results. To enhance the robustness and explanatory power of the empirical findings, this study conducts multiple checks on the impact of the digital economy on the urban innovation efficiency gap. These checks include introducing multidimensional fixed effects, performing subsample regressions, excluding data from the pandemic years and employing feasible generalized least squares (FGLS) as an alternative to OLS for estimation.

4.2.1. Multidimensional Fixed Effects

Cities within the same province often share similar policy support, geographical environment, and cultural–economic characteristics. These factors may simultaneously influence both the level of digital economy development and the gap in urban innovation efficiency. However, if they are not captured by the model, this could lead to omitted variable bias. Therefore, in addition to controlling for city and time fixed effects, provincial fixed effects are further incorporated to isolate the influence of provincial characteristics on the estimated results. The specific results are presented in column (1) of

Table 5. Robustness and endogeneity tests.

	Robustness				Endogeneity
	(1)	(2)	(3)	(4)	(5)	(6)
	Multidimensional Fixed	Subsample	Excluding the COVID-19	FGLS	IV	GMM
L.Gap						0.0654 *
						(0.0365)
Dige	−0.3072 *	−0.3205 *	−0.3380 *	−0.3011 ***	−1.9662 *	−1.9013 **
	(0.1772)	(0.1851)	(0.1953)	(0.1039)	(1.1348)	(0.9470)
LnDev	0.0004	0.0011	−0.0225	−0.0049	0.0150	0.1655 **
	(0.0266)	(0.0270)	(0.0316)	(0.0184)	(0.0408)	(0.0746)
Indus	0.0227	0.0229	0.0083	0.0290	0.0852 *	0.2913
	(0.0375)	(0.0379)	(0.0434)	(0.0260)	(0.0493)	(0.2676)
LnPeo	0.0016	0.0023	0.0482	−0.0088	0.1560	0.3104
	(0.0638)	(0.0644)	(0.0820)	(0.0410)	(0.1433)	(0.3090)
Findp	0.0498	0.0488	0.0662	0.0011	0.0075	−0.0810
	(0.0626)	(0.0633)	(0.0694)	(0.0383)	(0.0811)	(0.2128)
Fina	−0.0028	−0.0028	−0.0052	−0.0048	−0.0091	0.1992 *
	(0.0104)	(0.0105)	(0.0108)	(0.0074)	(0.0108)	(0.1060)
Fdi	−0.6762 *	−0.6600 *	−0.9469 **	−0.5024 **	−1.3528 ***	0.4061
	(0.3773)	(0.3902)	(0.4046)	(0.2270)	(0.4716)	(2.1360)
Tech	0.1341	0.1405	0.3109	0.2361	0.4087	1.0756
	(0.2409)	(0.2437)	(0.2649)	(0.1778)	(0.4628)	(1.4091)
KP rk LM					42.5200 ***
KP rk Wald F					34.7700 [16.3800]
AR(2)						0.9070
Hansen test						0.9690
Constant	0.2305	0.2174	0.3409	0.3714
	(0.4868)	(0.4943)	(0.5829)	(0.3921)
R2	0.2615	0.2535	0.2918
Pro FE	YES
Year FE	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES
N	3168	3120	2640	3168	2286	2640

Notes: *, **, *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively; the numbers in parentheses represent the standard errors.

. The outcome remains significantly negative, consistent with the baseline results.

4.2.2. Subsample Regression

Considering that municipalities directly under the central government, as provincial-level administrative units, enjoy greater policy autonomy and fiscal resources, their digital economy policies are often more pioneering and specific. Additionally, municipalities typically have larger economies and more advanced industrial structures, and their innovation efficiency may be influenced by economies of scale or agglomeration effects. Therefore, this study excludes the samples from municipalities directly under the central government for robustness testing. The regression results are presented in column (2) of Table 5. The coefficient of digital economy remains significantly negative at the 10% level, consistent with the baseline results.

4.2.3. Excluding the COVID-19 Pandemic

The COVID-19 pandemic began to ravage the world at the end of 2019, and many countries and regions implemented lockdowns and restrictions, severely impacting business production and operations, with some even coming to a temporary halt. Business revenues plummeted, and R&D investments were significantly reduced, which in turn affected the innovation efficiency of the cities where these businesses are located. To eliminate the influence of the pandemic, this study excludes the samples from 2020 and 2021 and re-estimates the impact of digital economy development on innovation efficiency gaps. The specific results are shown in column (3) of Table 5. The coefficient of digital economy remains significantly negative at the 10% level, confirming the baseline regression results.

4.2.4. Replace the Model

Since the fixed effects model assumes that the error terms are homoscedastic and not autocorrelated, while in practice, panel data may exhibit heteroscedasticity or serial correlation due to the uneven development across cities, leading to an underestimation of standard errors and reduced efficiency in parameter estimation, this study introduces the feasible generalized least squares (FGLS) model to relax the strict assumptions of homoscedasticity and no autocorrelation in the fixed effects model. The results are presented in column (4) of Table 5. The regression results show that the development of the digital economy still significantly reduces the gap in urban innovation efficiency, further ensuring the consistency and reliability of the regression conclusions.

4.3. Endogeneity Test

4.3.1. Instrumental Variables Method

Given the potential issues of omitted variables in empirical analysis, as well as the possible reverse causality between the development of the digital economy and the innovation efficiency gap, i.e., the smaller the inter-city innovation efficiency gap, the more likely it is to drive the flourishing of the digital economy, this study employs an instrumental variable (IV) method to test for endogeneity. Drawing from the approach of Huang et al. (2019) [96], the interaction term between the number of fixed-line telephones per million people in 1984 and the number of international internet users from the previous year is used as an instrument. The internet technology underlying the digital economy began its development with the popularization of fixed-line telephones, and the lagged one-year internet user count in the interaction term reflects the temporal development trend of the digital economy, thus meeting the relevance requirement. Additionally, as the digital economy evolves, the influence of fixed-line telephone communication methods on the innovation efficiency gap can be neglected, satisfying the exclusion requirement. The two-stage least squares (2SLS) estimation results are presented in column (5) of Table 5. The LM test results significantly reject the null hypothesis at the 1% level, indicating the presence of at least one valid instrument, thus allowing for effective IV estimation. The Wald F-test result, greater than 10, passes the weak instrument test. After addressing the endogeneity issue, the impact of the digital economy on the innovation efficiency gap remains significant at the 10% level.

4.3.2. Dynamic Generalized Method of Moments (GMM)

This study also employs the dynamic Generalized Method of Moments (GMM) method to test for endogeneity, with the results presented in column (6) of Table 5. The p-value for AR(2) is 0.907, which accepts the null hypothesis that there is no second-order serial correlation in the differenced error terms. Additionally, the Hansen test’s p-value is 0.969, which accepts the null hypothesis that the instrumental variables are not correlated with the error terms. The lagged term of the urban innovation efficiency gap (L.Gap) is significant at the 10% level, and the development of the digital economy remains significantly negative at the 5% level.

4.4. Mechanism Test

Building upon the first-stage regression, this study employs Model (7) to capture, on a city-year basis, the dual transmission mechanisms through which the digital economy influences the gap in urban innovation efficiency—namely, the optimization of the innovation network $\widehat{Invent}$ and the reinforcement of the information network $\widehat{Internet}$. In the subsequent second-stage regression, these identified transmission effects are substituted for the core explanatory variable in the baseline regression, allowing for a clearer identification of how the digital economy affects disparities in innovation efficiency via these two channels.

Table 6. Results of mechanism test.

	(1)	(2)	(3)	(4)
Dependent Variable	Invnet	Gap	Internet	Gap
Dige	0.4629 ***		1.3395 ***
	(0.0231)		(0.0849)
$\widehat{Invent}$		−0.6637 *
		(0.3828)
$\widehat{Internet}$				−0.2293 *
				(0.1323)
Constant	−1.0479 ***	0.2305	−2.7809 ***	0.2305
	(0.0634)	(0.4868)	(0.2332)	(0.4868)
R2	0.9491	0.2615	0.9468	0.2615
Controls	YES	YES	YES	YES
Year FE	YES	YES	YES	YES
City FE	YES	YES	YES	YES
N	3168	3168	3168	3168

Notes: *, *** indicate statistical significance at the 10%, and 1% levels, respectively; the numbers in parentheses represent the standard errors.

presents the results of the mechanism analysis. First, the digital economy can significantly narrow the innovation efficiency gap among cities by optimizing the innovation network. Specifically, a 1 percentage point increase in digital economy development is associated with a 0.463 percentage point improvement in the innovation network, which in turn contributes to a 0.664 percentage point reduction in the innovation efficiency gap. This finding indicates that digital platforms can serve as effective conduits for communication among diverse innovation actors across cities, removing network linkage barriers, accelerating the efficient flow of technological and knowledge-based innovation factors, and ultimately fostering a more balanced development of urban innovation efficiency—thereby validating Hypothesis H2a. Second, the digital economy also contributes to narrowing the innovation efficiency gap by reinforcing the information network. A 1 percentage point increase in digital economy development leads to a 1.334 percentage point enhancement in information network strength, which subsequently reduces the innovation efficiency gap by 0.229 percentage points. This suggests that the continuous improvement of digital infrastructure significantly reduces the cost of intercity information exchange, deepens information network connectivity, and enhances access to external knowledge while mitigating information asymmetry. These improvements facilitate more efficient innovation and stimulate local “buzz” effects, thereby promoting synergies among internal innovation resources and further narrowing regional disparities in innovation efficiency. This finding supports Hypothesis H3a.

4.5. Moderating Effect Test

Table 7. Results of moderating effect test.

	Moderating Effect Test
	LnEconet
Dige	3.9390 **
	(1.9200)
LnEconet	0.0008
	(0.1065)
Dige × LnEconet	−0.2254 **
	(0.1015)
Constant	−0.3208
	(0.5887)
R2	0.2627
Controls	YES
Year FE	YES
City FE	YES
N	3168

Notes: ** indicate statistical significance at the 5% levels; the numbers in parentheses represent the standard errors.

reports the results of the moderating effect of urban economic networks. The interaction term between the digital economy and the urban economic network is significantly negative at the 1% level, indicating that embeddedness within economic networks significantly amplifies the impact of the digital economy in reducing disparities in urban innovation efficiency. The underlying reasons are twofold. On one hand, within a strongly connected economic network environment, the digital economy can not only optimize factor allocation and accelerate the circulation and sharing of innovation elements such as knowledge and technology—thereby shortening the spatial time lag in digital technology adoption—but also effectively enhance the Pareto efficiency of innovation resource allocation. This facilitates the release of digital dividends, helps to bridge the regional innovation gap, and further narrows the disparities in urban innovation efficiency. On the other hand, the optimization of economic structures encourages more cities to actively integrate into the broader economic network, thereby extending the network’s scale. In this context, the digital economy, leveraging the co-construction and sharing of digital infrastructure and the efficient reuse of digital resources, significantly reduces the marginal cost of connecting individual nodes to the economic network, while simultaneously promoting exponential growth in the increasing returns to scale of data elements. Consequently, regions with weak innovation efficiency can harness the inclusiveness and diffusion capacity of the digital economy to initiate an innovation catch-up effect, further reducing the disparities in innovation efficiency across cities. In summary, the “efficiency optimization effect” and “network scale effect” of economic networks positively moderate the role of the digital economy in narrowing innovation disparities, thereby validating Hypothesis H4.

5. Further Analysis

5.1. Threshold Effect Test

To identify potential nonlinear effects of the digital economy on disparities in urban innovation efficiency across different urban network structures, this study follows Hansen (1999) [97] and conducts 300 bootstrap replications using the “Bootstrap” method to derive the p-values of the test statistics, thereby determining the presence of threshold effects. The test results are reported in

Table 8. Threshold number test.

Threshold Variable	Threshold Selection	F-Statistic	p-Value
LnEconet	Single	20.6200	0.0267
	Double	5.9500	0.6433
	Triple	5.3100	0.6567
Invnet	Single	12.1900	0.0433
	Double	−2.5200	1.0000
	Triple	2.4000	0.6833
Internet	Single	13.3700	0.0600
	Double	9.0000	0.2400
	Triple	6.6800	0.5500

. Whether the threshold variable is defined as the urban economic network, innovation network, or information network, the p-values for the single-threshold model are all below 0.1, while those for the double- and triple-threshold models exceed 0.1. Therefore, a single-threshold effect model is adopted.

5.2. Threshold Model Regression Results

Table 9. Threshold regression results.

		Moderator Variable
		LnEconet	Invnet	Internet
Threshold Variable	$q_1$	15.8568	0.0100	0.1767
$Dige\times I\left(Th\le q_1\right)$		1.8398 ***	0.6795 **	0.8384 **
		(0.5313)	(0.3468)	(0.3711)
$Dige\times I\left(Th\ge q_1\right)$		−0.4100 **	−0.3762 **	−0.3879 **
		(0.1697)	(0.2261)	(0.1697)
Constant		0.1188	0.3784	0.3501
		(0.4411)	(0.4342)	(0.4347)
R2		0.0129	0.0104	0.0109
Controls		YES	YES	YES
Year FE		YES	YES	YES
City FE		YES	YES	YES
N		3168	3168	3168

Notes: **, *** indicate statistical significance at the 5%, and 1% levels, respectively; the numbers in parentheses represent the standard errors.

presents the regression results of the single-threshold model. It is evident that, when using the urban economic network, innovation network, and information network as threshold variables, the impact of the digital economy on urban innovation efficiency disparities exhibits a nonlinear pattern—first widening, then narrowing. Specifically, when the strength of the urban economic network is below the threshold value (15.778), the coefficient of the digital economy is 2.607. This may be attributed to weak intercity linkages in economic networks, which hinder the flow of factors such as labor, capital, and material resources. Under such structurally constrained networks, cities have limited attractiveness to essential inputs. Consequently, the uneven diffusion driven by the digital economy exacerbates innovation polarization, resulting in a phase of expanding innovation efficiency disparities. Once the economic network strength surpasses the threshold, the coefficient becomes negative at −0.551, indicating that the interaction among multiple nodes within the economic network facilitates the gradient reallocation of physical factors. As economic networks deepen, the expansion of “pipelines” for factor mobility enables peripheral or underdeveloped cities to access high-quality resources from core cities, thereby enhancing local innovation performance and narrowing intercity innovation efficiency gaps. This finding supports Hypothesis H4b.

Similarly, when the strength of the urban innovation network is below the threshold (0.005), the digital economy intensifies disparities in urban innovation efficiency. At this stage, core cities leverage their resource endowments in technology, talent, and infrastructure to establish horizontal innovation partnerships with peripheral cities. Through digital platforms, these cities accelerate the siphoning of innovation resources and amplify innovation potential. However, peripheral cities, constrained by limited resources and market size, struggle to establish equitable innovation exchanges, falling into a low-efficiency trap that widens innovation disparities. Once innovation network strength exceeds the threshold, the coefficient of the digital economy shifts to −0.485. This indicates that, when intercity innovation networks are well-connected, the development of the digital economy significantly reduces disparities in innovation efficiency. In this phase, horizontal cooperation channels are established, and the innovation network evolves into a “small-world” structure, breaking path dependence on single-node diffusion. Such a decentralized configuration enables peripheral cities to bypass traditional hierarchical diffusion and directly share and complement innovation resources. The advancement of the digital economy substantially reduces the cost of cross-regional collaboration, strengthens intercity innovation cooperation, and thus narrows innovation efficiency gaps. This finding confirms Hypothesis H2b.

Lastly, when the urban information network strength is below the threshold value (0.177), the regression coefficient of the digital economy is 2.103. In this scenario, the underdeveloped state of information networks hampers the cross-regional flow of data and information, limiting the effectiveness of resource allocation and reinforcing “information islands”. Innovation resources concentrate in areas with higher information network density, exacerbating disparities. However, when the strength of the information network surpasses the threshold, the coefficient turns negative at −0.455. At this point, improvements in intercity information transmission efficiency significantly alleviate information asymmetries caused by geographic distance. Leveraging the high connectivity of information networks, the digital economy facilitates the cross-border allocation of innovation factors such as knowledge, technology, and talent. This network externality, driven by a “trickle-down effect”, accelerates technology acquisition in low-efficiency regions and contributes to narrowing urban innovation efficiency disparities. These results validate Hypothesis H3b.

5.3. Heterogeneity Test

5.3.1. Heterogeneity of the Digital Economy

To further investigate the impact of different dimensions of the digital economy on urban innovation efficiency disparities, this study decomposes the digital economy into five dimensions: digital infrastructure, digital industrialization, industrial digitalization, digital governance, and data value. The results, presented in

Table 10. Heterogeneity of the digital economy.

	(1)	(2)	(3)	(4)	(5)
Digital Infrastructure	0.0460
	(0.0934)
Digital Industrialization		−0.3509 ***
		(0.1209)
Industrial Digitization			0.0650
			(0.0905)
Digital Governance				−0.0510
				(0.0759)
Data Value					−0.1101
					(0.0985)
Constant	0.4987	0.1074	0.4556	0.4014	0.5253
	(0.4845)	(0.4847)	(0.4716)	(0.4754)	(0.4769)
R2	0.2608	0.2628	0.2608	0.2608	0.2610
Controls	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
N	3168	3168	3168	3168	3168

Notes: *** indicate statistical significance at the 1% levels; the numbers in parentheses represent the standard errors.

, indicate that only the coefficient of digital industrialization is significantly negative at the 1% level, while the effects of digital infrastructure, industrial digitalization, digital governance, and data value are not statistically significant. This suggests that the narrowing effect of the digital economy on urban innovation efficiency disparities is primarily driven by digital industrialization. A possible explanation is that digital industrialization, as the core sector providing digital technologies, products, services, and infrastructure, enhances innovation efficiency not only directly through technological advancement but also indirectly via knowledge and technology spillovers. These spillovers facilitate intercity technological exchanges and cooperation, enabling relatively underdeveloped cities to more effectively absorb the experiences and achievements of advanced regions, thereby accelerating their innovation efficiency improvement. In contrast, the current development of China’s digital economy is characterized by a relatively low level of industrial digitalization and weak digital governance, which may partially constrain the effectiveness of these components in reducing innovation efficiency disparities [98].

5.3.2. Regional Heterogeneity

Heterogeneity of North and South Regional

Given the significant differences between northern and southern regions in terms of geographical characteristics, levels of economic development, digital infrastructure, and the distribution of innovation resources, the impact of the digital economy on urban innovation efficiency disparities may exhibit region-specific patterns. To explore this heterogeneity, this study divides the sample into southern and northern regions using the Qinling–Huaihe Line as the demarcation and conducts separate analyses. As shown in Columns (1) and (2) of

Table 11. Regional heterogeneity.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	North	South	Southeast Side of the Hu Line	On the Hu Line	Northwest of the Hu Line	Large Cities	Small and Medium-Sized Cities
Dige	0.1523	−0.5041 **	−0.3746 **	0.3914	6.3043 *	−0.5218 ***	1.4118 ***
	(0.3722)	(0.1980)	(0.1770)	(0.8391)	(3.1962)	(0.1922)	(0.5438)
Constant	1.1463	−2.2284 **	0.4030	−1.1852	5.2335	1.1322	−0.0854
	(0.7905)	(0.9747)	(0.5077)	(1.7302)	(4.0214)	(0.8737)	(0.6084)
R2	0.2598	0.2797	0.2618	0.2295	0.3926	0.2806	0.2586
Controls	YES	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES
N	1332	1836	2664	372	132	1176	1992

Notes: *, **, *** indicate statistical significance at the 10%, 5%, and 1% levels, respectively; the numbers in parentheses represent the standard errors.

, the development of the digital economy in southern cities has a stronger effect in reducing urban innovation efficiency disparities compared to northern cities. This outcome may be attributed to two main factors. First, since the launch of China’s reform and opening-up, many southern cities have served as pilot zones for institutional innovation, leading to a more diversified industrial structure encompassing a broader range of sectors and fields [99], which provides favorable conditions for the digital economy to mitigate innovation efficiency disparities. Second, in contrast to the South, northern cities retain stronger foundations in planned innovation systems and face greater institutional inertia in transitioning to a market-oriented model, resulting in fewer market-driven innovation entities [100]. This institutional rigidity impedes the efficient flow of data, capital, and labor resources [101], thereby limiting the vitality of the digital economy and the fluidity of elements within complex urban networks. Overall, the combination of early-mover advantages and market-oriented characteristics in the South contributes to a spatial gradient in the effectiveness of the digital economy in narrowing innovation efficiency disparities.

Heterogeneity of the Regional Endowment

Given the substantial lag in overall economic strength and availability of digital talent in the northwest region relative to the southeast, this study, following Ding et al. (2021) [102], adopts the Hu Line as a geographical boundary to divide the sample into three parts: southeast, northwest, and cities located along the line itself. As reported in Columns (3) to (5) of Table 11, the coefficient of the digital economy is significantly positive at the 10% level in the northwest of the Hu Line, significantly negative at the 5% level in the southeast, and statistically insignificant for cities situated on the line. A possible explanation is that in the southeast region, digital infrastructure is more comprehensively developed and information flows more efficiently, which effectively facilitates technological diffusion and resource sharing. This supports the formation of scale effects in the digital economy and promotes the balanced allocation of innovation factors. In contrast, the northwest side of the Hu Line suffers from a lower penetration rate of digital technologies and underdeveloped digital infrastructure, which impedes the efficient circulation of data, capital, and technology. Consequently, central cities in the northwest may siphon resources from surrounding areas, intensifying intra-regional “core–periphery” disparities and thereby widening urban innovation efficiency gaps.

Urban Population Size Heterogeneity

Given the potential disparities in both digital economy development and urban innovation efficiency across cities of varying population sizes, this study adopts the method of Wu et al. (2024) [103], using a population threshold of 4 million as the cutoff point to divide the sample into large cities and small-to-medium-sized cities. The empirical results are presented in Columns (6) and (7) of Table 11. It is observed that digital economy development in large cities significantly reduces intercity innovation efficiency gaps, while in small-to-medium-sized cities, the effect is significantly positive. The underlying reason may be that large cities, due to their greater population size, typically possess superior transportation systems, digital infrastructure, comprehensive administrative services, and a strong innovation atmosphere. These factors not only directly enhance technological absorption and recombination capacity but also facilitate real-time exchange of information and technology, thereby enabling more efficient urban connectivity and amplifying the digital economy’s network scale effects on narrowing innovation efficiency disparities. This finding aligns with the conclusions of Xiong et al. (2018) [104]. Conversely, small-to-medium-sized cities are often disadvantaged in terms of talent concentration, infrastructure, and access to advanced innovation equipment. These limitations hinder both digital economy development and the realization of leapfrogging effects in innovation, while also impeding their integration into broader urban networks, thereby restricting their ability to leverage the novel knowledge generated by large cities to enhance local innovation efficiency.

6. Conclusions and Policy Implications

6.1. Conclusions

Based on panel data from 264 prefecture-level and above cities in China spanning the period from 2011 to 2022, this study employs fixed effects and two-stage regression models to empirically investigate the impact and mechanisms through which the digital economy affects intercity innovation efficiency disparities, focusing on innovation, information, and economic networks. The main findings are as follows: (1) The digital economy significantly reduces disparities in urban innovation efficiency. This conclusion remains robust after a series of tests for endogeneity and robustness. The digital economy contributes to narrowing these disparities primarily by optimizing innovation networks and strengthening information networks, while economic networks play a positive moderating role in this process. (2) When innovation, information, and economic networks are treated as threshold variables, the impact of the digital economy on innovation efficiency disparities exhibits a nonlinear pattern: it initially widens the gap before ultimately narrowing it. (3) Heterogeneity analyses reveal that the narrowing effect of the digital economy on urban innovation efficiency disparities is mainly driven by digital industrialization. Other dimensions—namely digital infrastructure, industrial digitalization, digital governance, and data value—do not yet show significant effects. The digital economy exerts a stronger narrowing effect on innovation efficiency disparities in southern cities compared to northern cities. It significantly reduces such disparities in cities located southeast of the Hu Huanyong Line, while in the northwest region, it tends to widen the gap; in cities situated directly on the line, its impact is not statistically significant. Furthermore, in large cities, digital economy development effectively narrows innovation efficiency disparities, whereas in smaller cities, it paradoxically exacerbates them.

6.2. Policy Implications

Based on the findings above, this study offers the following policy recommendations:

1. Continue deepening the development of the digital economy and leverage its key role in narrowing the innovation efficiency gap between cities: According to the empirical results, China should prioritize advancing the dimension of digital industrialization while strengthening the other four dimensions of the digital economy to bridge the innovation efficiency gap between cities. First, develop integration across the upstream and downstream digital industry chains, focusing on supporting core technology enterprises such as cloud computing, big data, artificial intelligence, and blockchain in deep collaboration with manufacturing and service industries to cultivate innovative industrial clusters. Considering fiscal investments and technological maturity, it is recommended to initially set up “industry collaboration pilot zones” in core eastern cities such as Shanghai, Suzhou, and Hangzhou, creating replicable experiences. In the medium term, select cities such as Shenzhen and Ningbo to explore pilot projects for incorporating data assets into accounting systems, allowing qualified enterprises to account for data assets as intangible assets. Establish a data asset sandbox mechanism and conduct data value assessments in high-value data scenarios such as finance and healthcare to release data dividends and help narrow the innovation efficiency gap. Secondly, digital infrastructure, industrial digitalization, digital governance, and data value currently have limited effects on narrowing the innovation efficiency gap, and these should be prioritized for medium- to long-term development. For example, the government can collaborate with leading companies such as Huawei and Alibaba to develop smart transportation and energy management systems. Drawing on the practice of Hangzhou’s “City Brain” using AI to optimize traffic signals, this should be scaled up and replicated in cities like Zhengzhou and Xi’an with traffic bottlenecks. Leverage Alibaba Cloud’s “Feitian” scheduling system to dynamically match real-time demands in the industrial internet and encourage companies to continuously upgrade their industrial internet platforms, insisting on data integration to optimize the supply chain in the long term. Continuously improve digital governance and build a dynamic regulatory framework suited to new business models, establishing cross-domain governance conventions, unified data collection standards, and mutual recognition of law enforcement within urban clusters. This will accelerate the interoperability of patent pools and elastic scheduling of computing power resources, enhancing collaborative innovation across cities.

2. Enhance urban network connectivity and continue strengthening urban network construction to enable balanced development of innovation efficiency: First, continue advancing infrastructure construction to solidify the foundation for the flow of elements. Increase the construction of high-speed rail, highways, and aviation infrastructure, reinforcing strategic intercity transport networks, and implementing China’s “Eight Horizontal and Eight Vertical” high-speed rail network plan. Simultaneously, promote the construction of 5G networks, data centers, industrial internet platforms, supercomputing centers, and urban brain systems to increase the coverage and transmission speed of urban information networks. Second, strengthen the radiation and driving role of core cities and foster the development of secondary node cities. Continue promoting the development of urban clusters such as the Beijing-Tianjin-Hebei, Yangtze River Delta, and Pearl River Delta regions, building innovation communities and science and technology corridors within urban clusters or metropolitan areas. Guide leading enterprises in core cities to relocate production bases to secondary cities, accelerating the establishment of a vertical network system linked by industrial chains. Promote the construction of cross-city intellectual property transaction markets and encourage core cities to transfer patents to secondary centers through tax incentives, accelerating the diffusion of technologies. Improve the urban innovation talent cultivation system, establish talent introduction and retention mechanisms, and encourage the cross-city flow and exchange of core talents. Through optimized talent allocation, strengthen the resilience of urban networks and enhance the innovation catch-up capacity of weaker network nodes.

3. Implement differentiated regional digital policies to narrow the regional innovation efficiency gap: On the one hand, implement differentiated “digital tax” policies to provide tax reductions and other subsidies to digital economy enterprises in western cities, while offering innovation subsidies to eastern cities’ enterprises that transfer digital technologies to the west, promoting cross-city development of digital technologies and innovation efficiency. At the same time, strengthen talent cultivation and introduction in western cities to ensure a talent base for the development of the digital economy. In cities like Chengdu and Xi’an, establish digital technology transformation centers offering services such as patent evaluation and financing connections to accelerate the “late-mover advantage” of western cities and lay the digital foundation for improving innovation efficiency in the western region. On the other hand, focus on leading digital technologies and ecosystem building, supporting leading enterprises to build industrial internet platforms and digital transformation exchange centers, and promoting the R&D of cutting-edge technologies such as artificial intelligence and quantum information. Form a “core enterprises + SMEs” collaborative innovation network. Additionally, deepen the “Eastern Data, Western Computing” project, guiding the transfer of computing power to western hubs, optimizing resource allocation efficiency through cross-regional data flow, avoiding redundant construction and homogeneous competition, releasing the “innovation diffusion” from eastern cities, and accelerating the absorption of digital technologies in western cities to drive a gradient leap in innovation efficiency.

7. Limitations and Further Research

This study has some limitations. First, in terms of indicator setting, due to the availability of urban data, the human capital input indicators primarily focus on quantitative measures, which fail to fully capture differences in the quality of the subjects. Additionally, the innovation environment cannot yet quantify institutional culture and soft factors. Output indicators also do not account for intangible benefits such as knowledge spillovers. The use of the entropy weight method to calculate secondary indicator evaluation indices may fail to overcome the problem of excessive reliance on data distribution characteristics, which may lead to an underestimation of some theoretically important indicators. Furthermore, the DEA method is sensitive to data quality and indicator selection, and can also be influenced by outliers. Second, this study uses the Baidu Index to represent intercity information flow. However, the Baidu Index reflects “potential attention” rather than “actual behavior”, and is influenced by factors such as algorithm weight adjustments and search channel diversions (e.g., differences between mobile and PC platforms), making it difficult to accurately capture the actual intercity information flow. Finally, there are significant differences in the policy environment, industrial base, and social culture of digital economies across different countries. Although the discussion of China as an emerging economy can offer some reference for other developing or emerging economies, it is still necessary to base the analysis on the specific economic conditions of each country. A comprehensive comparative analysis of the similarities and differences in the relationship between digital economy and innovation efficiency gaps is required to extract universally applicable laws and experiences.

As for future research directions, first, the current study is based on urban-level data in China. Future research could delve deeper into the industrial or enterprise level in China, exploring how the digital economy influences the innovation efficiency gap through mechanisms such as cross-organizational collaboration and cross-regional cooperation from a more mesoscopic or microscopic perspective. Second, this study mainly investigates the direct effects of the digital economy on the innovation efficiency gap. Future studies could incorporate spatial econometrics to analyze the spatial spillover effects brought about by the flow of network elements, the diffusion paths, and impact range of networks, as well as identify the radiation or siphoning effects of core cities on the innovation efficiency of peripheral cities. Third, subject to data availability, this research could be extended to major countries globally. By comparing the differences in the role of digital economies on urban innovation efficiency at different stages of development and under different institutional contexts, the international applicability of the findings could be enhanced. Lastly, with the rise in geopolitical risks and trade protectionism, the risks associated with digital economies, such as cybersecurity and data monopolies, have become increasingly prominent. Future research could incorporate network resilience into the analytical framework, exploring how cities can build more digitally integrated urban network structures to maintain the stability and sustainability of innovation efficiency when facing external shocks.