Knowledge Spillover and Spatial Innovation Growth: Evidence from China’s Yangtze River Delta

Abstract

This article explores the relationship between knowledge sources at different levels and corporate innovation from the perspective of urban cluster, with a focus on enterprises. This paper conducted an empirical analysis of 375 listed companies in 27 cities within the Yangtze River Delta urban cluster in China from 2009 to 2019. The findings showed that: (1) Local scientific knowledge spillovers, mediated by industry relevance, positively influence firms’ innovation performance. This study verifies how spatial knowledge is dimensionally reduced from scientific spillovers to industrial technological innovation. (2) Emerging industries acquire relevant scientific knowledge for transformation from a broader regional scope. Regional knowledge creation in the Yangtze River Delta urban cluster has stimulated industrial innovation across various sectors, thereby enhancing the overall innovation capacity and level of the urban cluster. (3) Regional diversity significantly affects the process of transforming knowledge into innovation. This paper supports the existence of a unified spatial innovation network among heterogeneous spatial economic entities and emphasizes the innovation synergy from lower to higher levels within heterogeneous hierarchical innovation networks. Developing urban agglomeration strategies that leverage the resource advantages of industrial clusters and adjust industrial layouts is an important approach to promote innovation and economic growth.

1. Introduction

Innovation is widely recognized as the main driver of long-term economic growth [1,2]. Factors affecting innovation are multifaceted, including government behavior [3], entrepreneurial ecosystem [3], technological cluster [4], and changes in the application of tools [5]. In the context of the knowledge economy, the relationship between innovation and science has become even more intertwined. Knowledge is the source of innovation and plays a fundamental role in enhancing scientific and industrial capabilities [6], serving as an important input for driving economic development [7]. The accumulation and dissemination of knowledge creation constitute the endogenous driving force for economic innovation-led growth [8]. Therefore, emphasizing the importance of knowledge generation and transformation is a prerequisite for innovation development.

Regional innovation systems have been a recent area of interest for scholars studying knowledge creation and transformation. Regional innovation systems, networks composed of businesses, research institutions, governments, and other relevant stakeholders promote the development of innovation through collaboration and knowledge sharing. Such systems can provide the resources and environment needed for innovation, strengthen cooperation and coordination among different stakeholders, and improve the efficiency and quality of innovation [9]. Enterprises have the opportunity to tap into knowledge resources offered by various stakeholders, through knowledge transformation, research and development activities, and technological innovation, providing crucial support for their competitiveness [10]. Therefore, in research and practice on innovation and sustainable development, the focus on knowledge transformation and innovation development in businesses within regional innovation systems is indispensable.

However, our understanding of how innovation systems evolve and support business innovation remains limited. Aghion et al. (2014) argue that enterprises often face path dependence when developing and adopting innovation, due to strong network effects and high conversion costs [11]. For example, businesses are likely to select innovations that complement their existing assets within the organization or its supply network. Innovation that deviates from the previous path may weaken existing positive network externalities and incur higher conversion costs. This requires bridging with local knowledge bases or networks. We need more research as we explore the innovation network system of urban clusters, which go beyond the boundaries of individual cities.

The Yangtze River Delta (YRD) urban agglomeration in China boasts the highest economic aggregate among urban agglomerations in the country, making it a highly representative and ideal sample for studying regional knowledge spillovers and innovation growth. It is characterized by advanced economic development, strong industrial innovation capabilities, notable intercity innovation synergy, and clear economic growth mode transformation [12]. Significant variations exist among the 27 cities in terms of economic development, scientific and technological education strength, and industrial development characteristics. Each city plays a crucial role in the innovation growth of the YRD urban agglomeration and contributes to an interconnected network of innovation. The region has a unified innovation network that connects all cities, as well as distinct subdivided innovation networks that emerge through knowledge overflow. These dynamics establish a cohesive and hierarchical innovation network system, resulting in significant synergistic innovation effects.

This paper has three main contributions. First, this study investigates how local knowledge creation and firm innovation behavior are related in urban agglomerations, taking into account the spatial distribution of firms. This helps us explore how knowledge is created and transferred across space. Second, this study examines how scientific knowledge creation and industrial innovation capabilities grow dynamically in the YRD urban agglomeration. It provides a comprehensive analysis of how cities at different hierarchical levels create and innovate knowledge and industry, revealing that scientific knowledge creation in the region where firms are located is crucial for firm innovation. Third, unlike existing research that only discusses firm innovation or regional innovation separately, this study integrates local knowledge and urban agglomeration knowledge into a unified analytical framework. This provides new validation for a deeper understanding of how knowledge is transferred in regional innovation systems, especially from the perspective of urban agglomerations. It also identifies possible mechanisms in the knowledge transfer process.

The paper is structured as follows. The second part presents the theoretical framework on spatial knowledge spillovers. The third part presents the research hypotheses. The fourth part provides characteristic facts based on the Yangtze River Delta innovation network and knowledge spillover. The fifth part describes the data and empirical strategy. The subsequent section presents the empirical results and the last part concludes the paper and discusses policy implications.

2. Theoretical Lines: A Survey of Spatial Knowledge Spillovers

Marshall (1890) creatively points out that “the mysteries of the trade become no mysteries, but are as it were in the air” [13]. Krugman (1994) also argues that “knowledge flows are invisible; they leave no paper trail by which they may be measured and tracked” [14]. Polanyi (1966) distinguishes between explicit knowledge, which can be codified and easily disseminated, and tacit knowledge, which is difficult to codify and share [15]. For example, basic scientific knowledge is more codable and easier to disseminate compared to applied technical knowledge. Some know-how on process flows cannot be transmitted in a simple, explicit way, but only through experiential learning.

Arrow (1962) suggested early that knowledge is endogenous in economic growth. He argued that individual activities can create social benefits, as pioneers learn by doing and latecomers benefit from their knowledge [16]. Knowledge spillover is the process through which newly created knowledge is recognized and absorbed by other individuals or entities. This concept implies that there is a “free lunch” in terms of innovative growth. The trajectory of knowledge diffusion and re-creation across multiple dimensions and levels becomes intricate. These perspectives highlight the importance of knowledge spillover in economic growth and the challenges in fully understanding and harnessing its potential [17].

However, in the field of spatial economic growth, knowledge spillovers still have a black box that has not been fully opened. Duranton and Puga (2004) introduced the micro-framework of sharing, matching, and learning, which suggests that proximity is a prerequisite for spatial agglomeration [18]. In the context of proximity, various forms of knowledge sharing, information matching, and mutual learning take place. These include the integration and coordination of workers and positions, as well as the exchange of scientific and industrial technological knowledge and even technological processes. The increasing complexity of scientific and industrial technical knowledge leads to the emergence of new and deep divisions of labor. Within this diverse knowledge and complex industrial division of labor, different types of knowledge, such as explicit, tacit, shared, and differential knowledge, become increasingly important.

Knowledge spillover is an unstable and uncertain process. The difficulty of matching knowledge between entities continues to increase, creating a strong geographical limitation for knowledge spillover. This leads to innovation clusters, which are geographic concentrations of innovation activities that facilitate knowledge flow within a shared space [19], facilitating the flow of knowledge in a unified space. Clusters, which are groups of closely located firms or organizations, serve as the material carriers for adjacent sharing, matching, and learning. Innovators with specialized skills can spill over knowledge more easily within clusters. In summary, the complexity of knowledge creation activities determines the spatial stickiness of knowledge, meaning that knowledge tends to be localized and concentrated in specific regions [20]. Knowledge spillover is a spatial phenomenon that is determined by the uneven occurrence of knowledge creation and aggregation.

The distance advantage of spatial proximity can potentially be an advantage in knowledge sharing, but it is not always guaranteed. Several factors come into play, with knowledge relevance being a prominent issue. In some situations, even if the space distance between an enterprise and a university is very close, the correlation between the scientific knowledge of the university and the local industry may not be high. This means that knowledge overflow, or the transfer of knowledge from the university to the industry, may not easily occur [21]. Another scenario is when the university system itself is not perfect, facing constraints in knowledge spillover. This could be due to various factors, such as a lack of effective channels or mechanisms for transferring knowledge from academia to industry. In such cases, the last mile constraint becomes a challenge, making it difficult for knowledge spillover to be realized.

Knowledge spillover also confronts the intricate transformation from fundamental science to applied technology. This transformation can be challenging, and even if knowledge spillover occurs, it may fail in addressing the last meter of technological problems. The intertwining of general-purpose technology and proprietary technology adds more complexity to the process, resulting in a diverse and multi-level mix of interactive knowledge spillovers [22]. Knowledge spillover is the process of overflowing from one subject to another, and this point-to-point process can be accessed by many subjects simultaneously, evolving into a line-to-line process. The higher the frequency of interaction, and the stronger the accumulation and self-reinforcement effects of knowledge spillover. Cooperation between innovative subjects generates positive feedback [23], thus creating an innovation network that expands from line to network [24].

The network characteristics of connectivity, flow, collaboration, and Incentives contribute to determining the magnitude and level of innovation within the network. However, identifying and measuring network effects can be challenging [25]. In conclusion, although the advantage of spatial proximity in terms of distance can potentially facilitate knowledge sharing, it is not a guaranteed advantage. Various factors, including knowledge relevance, the effectiveness of the university system, market competitiveness [26,27,28], and knowledge transformation complexity, can affect the success of knowledge spillover and the realization of network effects.

Greater agglomeration facilitates the efficient dissemination of knowledge and ideas, leading to increased intellectual creativity [29]. Moretti (2021) analyzed high-tech clusters in the top ten USA urban agglomerations from 1980 to 2010 and found that they remained innovative and grew larger [23]. For instance, the top ten urban agglomerations had 70%, 79%, and 59% of inventors in computer science, semiconductors, biology, and chemistry, respectively. Inventors who moved to these top ten high-tech clusters produced more influential patents than those in smaller clusters, showing high innovation efficiency. Innovative companies and creative scientists also tended to cluster geographically by discipline or industry. Notably, Silicon Valley and San Francisco attract a concentration of highly creative computer scientists due to their proximity to top engineering departments at Stanford or Berkeley and the presence of top computer science graduates who choose to remain in the area.

Spatially, R&D clusters tend to be adjacently nested. Buzard et al. (2020) found that patent citations, a measure of knowledge spillovers, decreased with distance from R&D laboratories in the USA [30]. A network of knowledge spillovers emerges from the R&D clusters, following the distribution of manufacturing industries. Knowledge flows from high-density to low-density areas and from near to far locations. Knowledge creators gather in the network, which carries knowledge overflow. The more connected the network is, the more sharing, matching, and learning happens.

Local innovation networks are formed by integrating specialized labor and diverse elements, such as local talents, production processes, and professional information. Diverse resources, knowledge, talents, and innovation activities help industries develop in different ways, resulting in cities with diverse network connections that further attract diverse talents and innovation activities. Knowledge spillovers can take different forms, but they are all influenced by the innovation network of the innovator and other related organizations. The innovation network increases capital and knowledge creation, and strengthens the interaction among knowledge spillovers, industrial agglomeration, and economic growth. Knowledge spillover effects are not uniform. Cities and enterprises in the same innovation network may have different knowledge stock, comparative advantages, and technological innovation capabilities. However, they can all benefit from the collective wisdom of “the predecessors plant trees, and the descendants enjoy the shade”.

Berliant and Fujita (2009) used a Two Person model (TP model) to examine how two types of heterogeneous subjects can share and cooperate on their independently accumulated knowledge to increase knowledge output [31]. This model can also apply to knowledge spillover among universities, enterprises, or both. To create new knowledge, two economies, i and j, need to work together on three components: their shared knowledge, the knowledge that flows from i to j, and the knowledge that flows from j to i. The growth rate of novel knowledge hinges on shared knowledge and the respective proprietary knowledge of each subject. The common knowledge stock of i and j, along with the knowledge stock that flows between them, determines the total social knowledge spillover rate. This implies that existing knowledge stock influences knowledge spillover, and that the level of spillover from sharing, matching, and learning determines the growth rate of new knowledge output when the stock is constant. The growth rate of subject i’s output depends on both its own knowledge growth rate and the spillover of specialized knowledge from subject j.

This study aims to explore the relationship between distance, knowledge dissemination, and knowledge spillover, and extends the TP model accordingly. The distance between subject i and j, denoted as ${\mathrm{d}}_{\mathrm{i}\mathrm{j}}$∈[0, n], affects the knowledge spillover effect, which decreases as the distance increases. The parameter λ represents an individual’s ability to absorb knowledge spillover, influenced by “distance”, with ${\mathsf{\lambda }}_1$ to ${\mathsf{\lambda }}_{\mathrm{n}}$ indicating a diminishing effect as the distance becomes greater. The ease of knowledge dissemination is reflected in the impact of the connection density between subjects i and j in the network on the level of knowledge spillover. The multi-dimensional influence mechanism of spatial proximity, knowledge correlation, and network association plays a role in the spillover of newly created knowledge within a specific space at a low cost.

Knowledge creation and spillover form a dynamic process of cyclic accumulation, with knowledge creation being endogenous through knowledge spillover. The knowledge spillover effect is influenced by knowledge correlation and individual differences, as individuals with knowledge correlation can benefit from knowledge spillover in networks that involve knowledge sharing and learning. However, the acquired knowledge spillover effects vary due to differences in knowledge stock and the level of proprietary technology. The improvement of knowledge creation and innovation ability is influenced by the capacity to absorb knowledge.

3. Research Hypothesis

3.1. Scientific Knowledge and Local Firm Innovation

The perspective on the sources of urban innovation in current research has undergone a gradual shift. Initially, there was a focus on the introduction of external knowledge, but greater emphasis is now placed on local knowledge structures and the significant influence of a city’s innovation capabilities. Researchers have extensively studied how to enhance the innovation performance of firms, considering them as active agents in urban innovation. Previous studies have shown that firms’ innovation activities are influenced not only by their own investments [32,33], but also by the innovation environment of their location, such as the human capital within the city [34].

Arora et al. (2021) found that the proximity-based knowledge spillover effect significantly contributes to improving corporate innovation performance, as evidenced by their analysis of 800,000 corporate publications and patent citation data [35]. Local universities play a significant role in promoting innovation by facilitating knowledge spillover and academic collaboration through their research activities, expertise sharing, and collaborative projects with local enterprises. This reduces search costs and attracts scientific talent and knowledge [30,36]. Universities are major contributors to scientific knowledge creation, and their collaboration with local enterprises serves as a key channel for knowledge spillover and innovative growth in developing countries [37].

The proximity between firms and universities significantly impacts knowledge spillovers [38]. Audretsch and Lehmann (2005) conducted a study in Germany with a small sample size, testing the knowledge spillover effect of universities on neighboring enterprises [39]. The empirical results indicated that university students contribute as innovative labor and directly drive knowledge spillover. Scientific papers, which represent the knowledge output of universities, serve as a source of corporate knowledge. New types of engineering universities are particularly effective in addressing application problems, as their knowledge spillover effects tend to be more direct. Additionally, the openness or conservatism of a university, as well as its attitude towards the development of new disciplines, can impact corporate innovation activities [40,41].

Although spatial proximity, knowledge relevance, and network connections offer theoretical support for knowledge spillovers at specific spatial scales, further research is needed to understand the spillover processes of knowledge creation in cities at the micro-level of firms. Knowledge spillovers occur when firms with specialized expertise share knowledge with other firms. Firms aim to benefit from knowledge spillovers while leveraging their own knowledge stock, unique know-how, and industry-specific characteristics to integrate into the shared learning network. These firms seek to leverage locally available scientific knowledge that is relevant to their operations, using it to improve their production processes, develop new products, or enhance their services. In turn, this association with network innovation enables firms to access new ideas and resources, fostering the dynamic expansion of local industry clusters. Based on the analysis presented above, this paper proposes the following hypothesis:

Hypothesis 1 (H1).

Local scientific knowledge spills positively influence the innovation performance of firms.

3.2. Urban Cluster Innovation Networks and Firm Innovation Performance

Spatial proximity is important, but it does not automatically result in knowledge spillovers and sharing within a cluster [42]. The newly created knowledge needs to pass through the knowledge filter and join the relevant knowledge network to enable long-distance exchange and spillover of tacit knowledge in space. The knowledge network often provides an effective mechanism for mutual feedback between creating and spilling over knowledge. Innovation interest chains, formed by various stakeholders such as innovation agents and related organizations, play a significant role in this regard. They enhance economies of scale in capital and knowledge creation and strengthen the interaction among knowledge spillovers, industrial agglomeration, and innovation-driven growth.

Innovation networks exhibit spatial characteristics, spanning from city to city, university to industry, and industry to industry. Knowledge spillovers occur across cities through various proximity channels, including geographic, relational, social, cognitive, and technological factors [43,44,45]. Universities play a vital role in regional innovation as they serve as both knowledge generators and incubators of talent. The relationship between universities and industries is crucial for knowledge transfer and the commercialization of research findings, particularly in facilitating radical innovations [46].

This paper focuses more on the multi-layered nature of knowledge creation and spillover in the Yangtze River Delta, and seeks to discover the mechanisms by which spillover and diffusion occur between cities and industries at multiple levels. The integration strategy of urban agglomerations and the growing innovation linkages among cities at various levels have encouraged the formation of industry-related clusters that transcend administrative boundaries. This higher degree of industrial integration promotes knowledge spillovers and creation, particularly in strategic emerging industries [47]. Establishing innovation networks with complementary technologies and shared resources facilitates the transfer and diffusion of innovative knowledge in emerging industries [48]. Knowledge correlation is an important factor for emerging industries to cross physical distances and administrative boundaries to facilitate knowledge exchange and diffusion on a larger spatial scale. Building upon the aforementioned context, this paper proposes the following hypothesis to examine the contribution of a unified innovation network, conditioned by relevant scientific knowledge in the context of urban cluster, to the innovation performance of emerging firms:

Hypothesis 2 (H2).

The unified innovation network positively contributes to the innovation performance of firms.

3.3. Moderation of Scientific Knowledge for Innovation by Diverse Industrial Agglomerations

There is an interactive relationship between agglomeration and knowledge creation and spillovers, and spatial agglomeration of industries promotes the formation of cluster innovation networks and the growth of innovation output. Different types of industrial agglomeration have varying effects on knowledge spillovers. Jacobs argues that diversification and competitive market structures are more conducive to the creation and spillover of new knowledge [49,50]. Glaeser [51] and Boschma [52] argue that breakthrough innovation often occurs through the convergence and reorganization of knowledge across industries, and clustering diverse industries within cities promotes the flow and spillover of cross-industry knowledge, contributing to economic growth.

The choice of location and industrial agglomeration determines the spatial distribution of enterprises. This decision is influenced by various factors, including market access, availability of skilled labor, infrastructure, and proximity to suppliers or customers. A strong industrial base and knowledge stock in a city effectively promotes the clustering of related enterprises in the spatial scope, further promoting the creation and spillover of knowledge. Clearly, the diverse clustering of industries plays an important role in knowledge creation, spillovers and innovation growth. Based on this, this paper proposes the following hypothesis:

Hypothesis 3 (H3).

Industrial diversification agglomeration has a positive moderating effect on the knowledge spillover process.

Figure 1 shows the hypothesized research framework of this study.

4. Factual Characteristics

The Yangtze River Delta urban agglomeration, known for its robust economic development, comprehensive innovation elements, and leading scientific knowledge creation and industrial innovation capabilities, spans multiple provinces in China. In 2019, the YRD’s GDP reached 23.7 trillion CNY, accounting for 2.2% of the country’s land area, and the GDP created accounted for about 24%. In addition, there are 197 universities and 52 state key laboratories in the YRD, encompassing prestigious universities, including those designated as 24,985 and 211 universities, as well as 173 general undergraduate institutions. These universities have contributed significantly to scientific research, generating approximately 220,000 scientific papers and 120,000 invention patents. The region exhibits notable diversification in its industrial structure. In 2019, the region was home to 1042 listed companies with ongoing enhancements in innovation capabilities (see

Table 1. Invention patents of listed companies in the Yangtze River Delta from 2009 to 2019.

Industry	2009	2019	Growth Multiple
Computer, communications, and other electronic equipment manufacturing	63	1342	21
Software and information technology service industry	23	1270	55
Chemical raw materials and chemical products manufacturing	47	500	11
Chemical fiber manufacturing	1	34	34

Data source: Organized according to Wind and CSMAR data.

).

The Yangtze River Delta has a diverse urban hierarchy and functions. There are not only super cities such as Shanghai that integrate science, innovation, manufacturing, and financial centers, but also science centers such as Hefei and Nanjing. At the same time, it also includes newly industrialized cities such as Suzhou, Wuxi, Ningbo, and Taizhou. Using Python, this study analyzes scientific papers published in the YRD between 2009 and 2019, revealing rapid advancements in disciplines such as materials science and biomedicine. These fields have contributed significantly to the generation of new scientific knowledge, leading to a substantial increase in the number and proportion of scientific papers. As shown in Figure 2, the national key laboratories are mainly concentrated in Shanghai, Nanjing, Hangzhou, and Hefei. These four cities possess diverse scientific knowledge creation capabilities and concentrated innovation resources, facilitating knowledge spillover.

The innovation capabilities of cities in the Yangtze River Delta have converged notably, with a well-balanced spatial distribution of innovation output. The generation of scientific knowledge and the deepening of network density have increased the knowledge spillover effects, leading to this convergence [53]. As shown in

Table 2. Comparison of publications, innovation output, and GDP in 27 cities in the Yangtze River Delta.

City	2009			2019
	GDP (100,000,000)	Publications	Patents	GDP (100,000,000)	Publications	Patents
Shanghai	14,875.8	24,228	5980	38,156	68,219	22,593
Nanjing, Hangzhou, Suzhou, Wuxi, Hefei, Ningbo, 6 cities	2961.7	5491	1219	9599.5	20,773	7947
Remaining 20 cities	580.4	278	138	1958.5	1393	1650

Data source: National Bureau of Statistics, WOS database, and CSMAR data.

, a few cities such as Shanghai, Nanjing, and Hangzhou dominated the patent output of the YRD in 2009. Shanghai alone had 81.5% of the YRD’s patents, while Nanjing and five other major cities had 16.6%, and the rest of the cities had only 1.9%. By 2019, however, Shanghai’s share of patents dropped to 70.1%, while the six major cities rose to 24.9%, and the remaining cities rose to 5.1%. The share of publications followed the same trend. A multi-core innovation pattern emerged, with a higher interconnection density between cities. This challenges the previous pattern of one-way outward knowledge spillover observed in core cities and highlights the significant occurrence of mutual knowledge spillover among cities.

The spatial heterogeneity of industrial innovation has been further reinforced. Cities such as Nanjing and Hangzhou have leveraged their competitive advantages in industries such as information technology and communication equipment manufacturing. Late-developing cities such as Taizhou and Shaoxing have transitioned from chemical raw materials and product manufacturing to biomedicine. This intertwining of knowledge creation and industrial innovation across cities of different levels has fostered a collaborative innovation network with a high degree of agglomeration. The sharing of innovation resources between cities, the level of collaborative innovation, and the overall innovation output have witnessed significant improvements. Combined with their own scientific and industrial innovation capabilities, cities in the Yangtze River Delta have forged distinctive and specialized innovation paths.

Pharmaceutical manufacturing, information technology, and software services are emerging industries that rely heavily on knowledge and contribute significantly to regional development. This study uses pharmaceutical manufacturing as an example to show how knowledge-driven innovation matters. A text analysis of the main business scope of listed companies using Python revealed that the Yangtze River Delta has many biomedical companies as well as many non-biological pharmaceutical companies that have entered the field of biomedicine by expanding their production and operations.

Table 3. Non-biopharmaceutical companies whose main business involves pharmaceuticals.

Company	Location	Industry	Main Business
ABA Chemicals	Shanghai, Nantong	Chemical raw materials and chemical products manufacturing	Production: raw materials (operating with a license); pharmaceutical intermediates
Shanghai Haixin Group	Shanghai	Textile industry (leather, fur, down, and products)	Development and production of chemical fibers, animal and plant blended yarns, and related textiles; processing, manufacturing, and sales of chemical and traditional Chinese medicine raw materials and various preparations, biological products, health products, medical devices
Sihuan Bioengineering	Suzhou	Pharmaceutical manufacturing company (formerly known as Chengdu Parts Factory)	Large-volume injections, small-volume injections (including antineoplastic drugs), tablets (including antineoplastic drugs); wool textiles, cashmere products, knitwear
Insigma Technology	Hangzhou	Software and information technology service industry	Research and development of computer and network systems, computer system integration, and electronic engineering; biopharmaceutical development
Shanghai Jahwa United	Shanghai	Chemical raw materials and chemical products manufacturing	Development and production of cosmetics, daily chemicals, and raw and auxiliary materials; pharmaceutical research and development and technology transfer
Nantong Acetic Acid Chemical	Nantong	Chemical raw materials and chemical products manufacturing	Acetic acid derivatives. Production of hazardous chemicals and feed additives; production and sales of basic organic chemical raw materials and pharmaceutical intermediates
Zhejiang Jianfeng Group	Jinhua	Building materials industry	Manufacture and sales of cement and cement concrete; pharmaceutical intermediates
Zhejiang XinNong Chemical	Taizhou, Hangzhou	Manufacturing of pesticides and chemical raw materials	Production, processing, and sales of chemical pesticides; production, processing, and sales of pharmaceutical raw materials and intermediates
Wanbangde Pharmaceutical Holding Group	Hangzhou	The non-ferrous metal smelting and rolling processing industry	Production of the first class of medical devices; stripping non-ferrous metal calendering processing; pharmaceutical production
Zhejiang Garden Bio-chemical High-tech	Jinhua	Food manufacturing	Production and sales of food additives, production of feed additives; pharmaceutical intermediates
Greattown Holdings	Shanghai	Real estate	Comprehensive development of the real estate, construction, and sales of commercial housing; production, processing, and sales of biological products, pharmaceutical raw materials, and preparations, health care products
Pengxin International Mining	Shanghai	The non-ferrous metal smelting and rolling processing industry	Manufacturing of basic chemical raw materials; sales of chemical raw materials and products; manufacturing and sales of medical raw materials; special polymer new materials
Ovctek China INC	Hefei	Special Equipment Manufacturing	Three types of medical photoelectric instruments, medical testing, artificial organs, and medical equipment; drug research and development

Data source: Wind database.

provides examples of the transformation and upgrading of 13 companies in the biomedicine sector, categorized into related industry transformation and cross-industry transformation. The first category includes six companies that have transitioned from underlying technology or product production in related industries to the biomedicine industry. These non-pharmaceutical manufacturing companies have a relevant scientific knowledge base in biomedicine, which largely explains their transition. For instance, companies in the textile or chemical raw materials industries have used their chemical science knowledge to upgrade to pharmaceutical manufacturing. Ovctek is transforming based on its expertise in special equipment production and testing.

The second category covers six enterprises that have switched to the biomedicine industry from diverse sectors such as real estate, building materials, machinery, metal processing, and information technology. These enterprises have successfully shifted to new industries despite the significant industry span, low knowledge correlation, and associated uncertain risks. They have done so because urban agglomerations can promote the supply of scientific and industrial technology and stimulate market demand to guide businesses into new fields. The Yangtze River Delta has a robust chemical industry foundation and has developed chemistry, biology, and materials science. This has fostered a multi-level knowledge dissemination network and cross-industry knowledge spillover channels for cross-sector enterprises. Even in Zhejiang, where the historical foundation of chemistry and the chemical industry is not particularly strong, the presence of research institutes such as Zhejiang University, with the Hangzhou National Bioindustry Base as its core, has facilitated the expansion of the biological industry cluster to cities such as Hangzhou, Huzhou, Taizhou, Shaoxing, and Jinhua, forming the main Hangzhou Bay biological industry cluster.

In general, the 27 cities in the Yangtze River Delta exhibit variations in terms of economic development, scientific and technological education, and industrial characteristics. However, each city forms an interconnected subset within the urban innovation network of the YRD. Within this network, knowledge creation is shared and disseminated through innovation networks and industrial associations, leading to spatial diffusion of regional innovation and economic growth. As the process unfolds, the urban agglomeration in the YRD is transitioning from industrial integration to innovation integration, driven by scientific leadership, industrial innovation, and knowledge overflow.

5. Research Design

5.1. Empirical Models

In this section, we introduce an econometric model that aims to estimate the impact of regional knowledge on spatial innovation at both the city level and the Yangtze River Delta level. Equation (1) listed below are estimated with ordinary least squares (OLS) regression and standard errors are clustered at the firm-year level to correct for correlation within groups of observations [54].

$${lnpat}_{i,c,t}={\alpha }_0{+\beta }_1{lnRlpub}_{i,c,t}{+\beta }_2{lnRypub}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(1)

${lnpat}_{i,c,t}$ represents the natural logarithm of the number of innovations in firm i in city c in year t. The core explanatory variable ${lnRlpub}_{i,t}$ represents the local output of scientific knowledge which is correlated with the firm i in year t, and ${lnRypub}_{i,t}$ represents the YRD output of scientific knowledge which is correlated with firm i in year t. $C_{i,t}$ is a set of firm-level control variables, and $E_{c,t}$ is a set of city-level control variables, $w_i$ is a firm-level fixed effect, while ${\phi }_t$ is a time fixed effect. Finally, ${\epsilon }_{i,c,t}$ denotes the error term.

In order to address possible endogeneity issues in the above model, we start by finding the two instrumental variables (IV) for the endogenous variables, estimated using two-stage least squares estimation(2SLS) [55]. The models are as follows:

$${lnRlpub}_{i,c,t}={\alpha }_0{+\beta }_1{lnRlpu{b}_{IV1}}_{i,c,t}{+\beta }_2{lnRlpu{b}_{IV2}}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(2)

$${lnpat}_{i,c,t}={\alpha }_0{+\beta }_1{lnRlpu{b}_{hat}}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(3)

$${lnRypub}_{i,c,t}={\alpha }_0{+\beta }_1{lnRypu{b}_{IV1}}_{i,c,t}{+\beta }_2{lnRypu{b}_{IV2}}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(4)

$${lnpat}_{i,c,t}={\alpha }_0{+\beta }_1{lnRypu{b}_{hat}}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(5)

where ${lnRlpu{b}_{IV1}}_{i,c,t}$ and ${lnRlpu{b}_{IV2}}_{i,c,t}$ are instrumental variables for ${lnRlpub}_{i,c,t}$. ${lnRlpu{b}_{hat}}_{i,c,t}$ is an estimate of ${lnRlpub}_{i,c,t}$. Similarly, ${lnRypu{b}_{IV1}}_{i,c,t}$ and ${lnRypu{b}_{IV2}}_{i,c,t}$ are instrumental variables for ${lnRypu{b}_{hat}}_{i,c,t}$. ${lnRypu{b}_{hat}}_{i,c,t}$ is an estimate of ${lnRypub}_{i,c,t}$. Equations (2) and (3) are the 2SLS regression model for local scientific knowledge on innovation, and Equations (4) and (5) are the regression model for scientific knowledge on innovation in the Yangtze River Delta. The two sets of equations follow the same principle, i.e., in the first stage, the endogenous variables are estimated using the instrumental variables, i.e., Equations (2) and (4), and in the second stage, the estimates from the first stage are used to carry over into Equation (3) and Equation (5), respectively.

In this paper, we gate how diversity affects the impact of local related scientific knowledge on firm innovation. Diversified agglomeration promotes cross-industry knowledge flows and spillovers [56], which reduces the costs for firms to search for and acquire complementary resources [57] and thus is more conducive to innovation. We introduce the interaction term between diversity and local scientific knowledge to capture this moderating effect [58]. The models take the following forms:

$${lnpat}_{i,c,t}={\alpha }_0{+\beta }_1{lnRlpub}_{i,c,t}{{+\beta }_2{lnCitydiv}_{i,c,t}+\beta }_3{lnRlpub}_{i,c,t}\ast {lnCitydiv}_{i,c,t}+\lambda \ast C_{i,t}+\gamma \ast E_{c,t}+w_i+{\phi }_t+{\epsilon }_{i,c,t}$$

(6)

where ${lnCitydiv}_{i,c,t}$ is city diversity. The other variables are consistent with Equation (6) as described earlier. To calculate the other variables, we follow the procedures outlined in the subsequent section of the paper. These variables are selected based on their relevance to the research question and are derived using appropriate methodologies.

5.2. Indicators and Measurement

5.2.1. Dependent Variable

Measuring innovation is challenging, and there are various measures of innovation in the literature [59]. This paper focuses on the outcome of innovation, rather than the inputs to innovation activities or related indicators. We use patent data as a proxy indicator for innovation, following recent studies [60,61,62]. Patents do not measure all innovative activities (Smith, 2006) [8] or cover all inventions [63], as argued in the innovation literature. However, they are still reliable indicators of innovative activities [64]. Patent authorizations reflect the innovation level better than patent applications, as they receive recognition from the State Intellectual Property Office [65]. As mentioned earlier, the Yangtze River Delta (YRD) urban agglomeration is a leading region for innovation development in China, and this paper uses authorized invention patents of listed companies in the YRD as a proxy variable for firm-level innovation performance.

5.2.2. Independent Variable

The number of scientific paper publications has become a mainstream measure of scientific knowledge in recent years [66,67]. Scientific paper publications may reflect R&D investments, but firms can also use the latest contemporary scientific knowledge to achieve large technological leaps [68]. Scientific papers, as the knowledge output of universities, are a source of knowledge for firms [39]. Paper publication data enable us to visualize the temporal dynamics of regional scientific knowledge production, and provides a comparable and replicable indicator. Therefore, we follow Balland and Boschma (2022)’s work using scientific publications to measure scientific knowledge [69].

Taking one step further, we incorporate industry-related scientific knowledge outputs at the city level, the YRD level, and from 26 cities outside Shanghai in this paper. To calculate these indicators, we follow the steps outlined below.

Firstly, the scientific papers are meticulously classified. Retrieve scientific paper data from 2009 to 2019, pertaining to 27 cities in the YRD, using Web of Science (hereinafter referred to as WOS) and Python. Categorize dissertations based on the subject classifications provided by WOS. Presently, the WOS subject classification encompasses 13 major categories and 254 sub-disciplines. The scientific papers generated within the YRD urban agglomeration between 2009 and 2019 encompassed a total of 242 sub-disciplines.

Secondly, ascertain the precise industry classification to which each enterprise belongs. Jaffe et al. (1993) pioneered the research paradigm based on patent citations and verified knowledge spillovers by matching patent citations with geographical locations [70]. This study draws inspiration from their approach and utilizes text analysis to examine the top three primary industries associated with each enterprise, and further refine the industry classification based on its production activities.

Finally, the number of papers related to the industry associated with the enterprise’s production activities is individually matched, and ultimately we determine the number of papers related to the industry within the Yangtze River Delta.

5.2.3. Moderating Variable

This paper considers city diversity as the moderating variable. The diversification of industries within urban areas plays a crucial role in promoting knowledge spillovers, facilitating cross-industry knowledge exchange, and fostering innovation among firms. Agglomeration is measured using the generic HHI, while the inverse of HHI, based on the distribution of employees across different industries within the city, serves as an indicator of industrial diversification.

5.2.4. Control Variables

To mitigate the potential omitted variable bias and capture the impact of related knowledge on urban innovation, this article considers the availability of relevant data and includes controlling for relevant influencing factors at both the city and firm levels as comprehensively as possible. The selected control variables encompass the following factors:

• R&D Expenditure (Enrd): Innovation and R&D expenditure in enterprises exhibit a significant correlation. Agglomeration stimulates innovation by attracting talents and capital to specific locations. Hence, regional population density is included as a proxy for agglomeration economies [71].
• Firm Age (Age): The degree of spatial concentration of universities and research institutions is associated with knowledge creation, and applied research in university education fosters urban innovation [39].
• Financial structure (Intangible): Intangible assets are recognized as crucial catalysts for knowledge creation and innovative growth [72]. Companies emphasizing innovation frequently prioritize investment in research and development (R&D) to foster inventive breakthroughs. Inventions arising from R&D endeavors augment a firm’s intangible asset portfolio, bolstering its innovation capabilities and future growth prospects.
• Tobin’s Q (TobinsQ): Tobin’s Q ratio is commonly utilized as an indicator of a firm’s investment opportunities and growth prospects, making it more appealing to external investors. To sustain and strengthen this competitive advantage, firms often prioritize innovation as a means to develop novel products or services.
• City R&D expenditure (Cityrd): Government financial support influences the physical infrastructure of a region. Furthermore, it has an impact on how much local government supports innovation in terms of funding. The fiscal expenditure share of GDP is what we measure with this indicator.
• Enterprise location (perGdp): The level of economic development in a region is strongly associated with its innovation capacity. A higher level of economic development makes it easier and faster to attract innovation factors. Therefore, this paper uses the indicator of GDP per capita to represent the level of economic development [73].

All variables used in this paper and their measurements are displayed in

Table 4. Variables and their measurements.

Variable	Implication	Symbol	Measure	Source
Dependent variable	Enterprise innovation performance	$Inpat$	Authorized invention patent, logarithm	Wind, CSMAR
Independent variables	Relevant scientific knowledge in the city	$lnRlpub$	Industry-related papers in this city	Matching based on WOS and Wind databases
	Relevant scientific knowledge in the urban agglomeration	$lnRypub$	Papers related to the YRD industry	Calculated based on WOS and wind database
	Scientific knowledge related to 26 cities except Shanghai	lnPubexpsh	Industry-related papers in 26 cities	WOS
Moderate variable	Diversification of urban industries	$lnCitydiv$	City diversity HHI Index	Calculated based on Wind database data
Control variables	R&D Expenditure	$lnEnrd$	Enterprise’s annual R&D investment	Wind, CSMAR
	Age	$lnAge$	Establishment time	Wind
	Financial indicator	$Intangible$	Intangible asset rate	Wind
		$TobinsQ$	Tobin’s Q
	City R&D Expenditure	$lnCityrd$	City R&D expenditure	City Statistical Yearbook and Statistical Bulletin
	Enterprise space location	$lnperGdp$	City per capita GDP	City Statistical Yearbook

.

5.3. Data Collection

Financial data of listed companies, including main business income, business scope, and financial level, were obtained from the Wind database for this study. Invention pa-tents and R&D expenditures were sourced from the State Intellectual Property Office and the Guotaian database (CSMAR). Economic data for each city were collected from the ‘China City Statistical Yearbook’ and annual statistical bulletins, while data on urban scientific papers were sourced from the Web of Science scientific literature database. In case of abnormal values in a few indicators within the sample, the statistical yearbooks of the respective cities were corrected, calibrated, and manually verified. Furthermore, indicators that were not available for certain enterprises were excluded from the analysis. After applying the aforementioned procedures, the analysis was conducted using data from 325 enterprises spanning the period from 2009 to 2019.

6. Results and Discussion

6.1. Descriptive Statistics

Descriptive statistics are shown in

Table 5. Descriptive statistics.

Variable Name	Count	Mean	Sd	Min	Max
Pat	3575	9.62	41.13	0.00	788.00
Rlpub	3575	1144.35	2008.48	0.00	15,942.00
Rypub	3575	8579.18	9117.63	1.00	57,685.00
Pubexpsh	3575	5664.54	6165.62	1.00	40,853.00
Enrd	3189	5.72 × 1011	3.42 × 1013	1097.5	2.04 × 1015
Age	3575	13.94	6.63	0.00	29.00
Intangible	3575	0.04	0.06	0.00	0.79
TobinsQ	3575	2.45	2.45	0.60	69.67
Cityrd	3575	2.64 ×1010	3.95 × 1010	7.44 ×107	1.52 × 1011
perGdp	3575	98,397.93	33,026.81	10,611.00	180,044.00

. The logarithmic values of the variables are used in the model, and the standard deviation of most variables is smaller than the average value. However, the standard deviation of the number of invention patents is slightly higher than the average value. The data for industry-related papers in the Yangtze River Delta exhibit relative stability, with a smaller standard deviation compared to industry-related papers in this city. Other variables, such as R&D expenditure of enterprises, have some missing values, and their standard deviation is more significant relative to the average value. The correlation coefficients among the main variables are reported in

Table A2. Correlation matrix.

	lnpat	lnRlpub	lnRypub	lnCitydiv	lnEnrd	lnAge	Intangible	TobinsQ	lnperGdp
lnRlpub	0.146 ***	1.000
	(0.000)
lnRypub	0.371 ***	0.410 ***	1.000
	(0.000)	(0.000)
lnCitydiv	−0.097 ***	0.677 ***	−0.163 ***	1.000
	(0.000)	(0.000)	(0.000)
lnEnrd	0.385 ***	0.106 ***	0.389 ***	−0.151 ***	1.000
	(0.000)	(0.000)	(0.000)	(0.000)
lnAge	0.044 ***	0.264 ***	0.034 **	0.305 ***	−0.123 ***	1.000
	(0.009)	(0.000)	(0.039)	(0.000)	(0.000)
Intangible	−0.030 *	−0.067 ***	−0.004	−0.092 ***	−0.037 **	0.031 *	1.000
	(0.070)	(0.000)	(0.829)	(0.000)	(0.026)	(0.064)
TobinsQ	−0.109 ***	0.039 **	−0.031 *	0.068 ***	−0.057 ***	−0.008	0.080 ***	1.000
	(0.000)	(0.020)	(0.064)	(0.000)	(0.001)	(0.639)	(0.000)
lnperGdp	0.138 ***	0.468 ***	0.150 ***	0.356 ***	0.049 ***	0.396 ***	−0.050 ***	−0.005	1.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.004)	(0.000)	(0.003)	(0.749)
lnCityrd	0.040 **	0.586 ***	0.015	0.733 ***	−0.075 ***	0.490 ***	−0.077 ***	0.003	0.644 ***
	(0.016)	(0.000)	(0.380)	(0.000)	(0.000)	(0.000)	(0.000)	(0.862)	(0.000)

Notes: *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.

.

6.2. Hypothesis Testing

Table 6. Basic model result.

	(1)	(2)	(3)
	lnpat	lnpat	lnpat
lnRlpub	0.078 ***		0.097 ***
	(2.63)		(2.89)
lnRypub		0.030	−0.058
		(0.53)	(−0.91)
lnEnrd	0.003	0.003	0.003
	(0.62)	(0.66)	(0.62)
lnAge	0.375 ***	0.377 ***	0.376 ***
	(3.00)	(3.02)	(3.01)
Intangible	1.139	1.119	1.123
	(1.62)	(1.59)	(1.59)
TobinsQ	−0.014	−0.014	−0.014
	(−1.14)	(−1.16)	(−1.16)
lnperGdp	0.531 **	0.546 **	0.529 **
	(2.31)	(2.38)	(2.31)
lnCityrd	0.087 ***	0.088 ***	0.086 ***
	(2.78)	(2.83)	(2.77)
_cons	−8.428 ***	−8.477 ***	−8.038 ***
	(−3.12)	(−3.13)	(−2.99)
N	3575	3575	3575
adj. R2	0.221	0.220	0.222
CityFE	Yes	Yes	Yes
YearFE	Yes	Yes	Yes
Control	Yes	Yes	Yes

Notes: t statistics in parentheses. ** and *** denote significance at 5% and 1% levels, respectively.

presents the baseline regression results, which examines the relationship between relevant scientific knowledge and enterprise invention patents. In columns (1) and (3), a significant positive correlation is observed between the stock of industry-related scientific knowledge in this city and the growth of enterprise innovation, verifying hypothesis 1. This finding suggests a transfer relationship of knowledge spillover between basic science and industry. The micro-level mechanism of enterprise operation with knowledge spillover is manifested through enterprises digesting and absorbing relevant local scientific knowledge, improving their production and management, or creating new products and processes to promote innovation output. Furthermore, the positive impact remains stable and significant when considering the establishment time (lnAge) of listed companies as an indicator of corporate maturity. The location factor of the city also influences the innovation performance of enterprises. Higher levels of economic development or cities with higher urban scientific research expenditure provide a more conducive environment for innovative activities by enterprises.

The reason why the output of scientific papers from the city itself is more beneficial to the output of urban invention patents than the scientific papers of the YRD as a whole could be attributed to the limited dissemination of scientific knowledge by distance. It is crucial to explore the reasons behind the distance limitations in knowledge transmission. Are different industries subject to varying knowledge spillover effects? In the chain of transforming science into industrial technology, manufacturability, and application functionality are difficult to jump. From abstract to concrete, from pure theoretical knowledge to applied knowledge, scientific research needs to go through multiple rounds of integration of science and industry. Transforming basic research into industrial technology involves not just explaining the world, but also changing it. High technical efficiency and satisfactory product utility are essential for successful applied research.

Moving from basic science to applied research requires multiple levels of integration. In the case of an important invention patent, there are stratifications at the enterprise, university, and city levels. If a company is not located in Shanghai but knows that a top-ranked university in Shanghai has the necessary scientific research capabilities for the invention, distance constraints may hinder the seamless connection between scientists and enterprises in Shanghai, limiting the integration between science and industry. However, in the city where the enterprise is located, there may be a university or scientific research institution conducting applied research in the same field as universities in Shanghai and the Yangtze River Delta. This local university may have close academic ties with first-class universities and scientists in Shanghai. In this case, there is no “distance” of knowledge overflow in the field of codable basic scientific knowledge between the local universities and scientists. Although the level of basic scientific research may be lower than that of Shanghai, the practical ability in the field of manufacturability and functional utility research is more prominent.

To summarize, the theoretical logic underlying the intuitive description is the stepwise dimensionality reduction chain from basic theory to technical application in knowledge spillover. Difficult-to-code knowledge spillovers necessitate more face-to-face cooperation, leading to a diminishing effect of knowledge spillovers due to distance constraints. It is important to note that enterprises create invention patents by applying science and also deepen basic scientific research, which results in a series of positive feedback effects in knowledge creation and spillover. Consequently, the impact of local scientific papers on local invention patent authorization becomes more significant, likely due to the need for face-to-face application innovation. However, this face-to-face interaction depends on knowledge overflow between research universities and applied universities, which can lead to breakthroughs in relatively important invention patents by combining codable and non-codable knowledge through collaboration between research universities and applied universities. Moreover, industry heterogeneity plays a role, with knowledge-intensive industries being more reliant on science and experiencing more pronounced effects of knowledge spillover in the knowledge economy era compared to traditional industries.

To address potential endogeneity concerns in this study, several measures can be taken. Firstly, the panel two-way fixed effects model employed in this paper, incorporating firm and time fixed effects, helps to mitigate omitted variable endogeneity issues to some extent. However, it is important to acknowledge the possibility of endogeneity between urban scientific knowledge and firm inventions from both theoretical and practical perspectives. On one hand, there is a complementarity between scientific knowledge and technological inventions, as highlighted by previous studies [74,75,76,77]. For instance, Azoulay et al. (2009) found a positive effect of patents on publications by university researchers in the biotechnology field [78]. On the other hand, it is plausible that omitted variable bias may still exist, as innovation activities within firms are influenced by various factors. Despite controlling for several variables in the baseline section, there remains a possibility of omitting certain unobservable factors.

Based on this, this paper estimates the relationship between related scientific knowledge and firm innovation using an instrumental variable approach. In this paper, the interaction between the number of relevant books in modern China (The Full Text Database of Modern Chinese Books (1840–1949) contains over 120,000 titles) and the average number of city-level publications per year is chosen as the instrumental variable for relevant knowledge for the following reasons. The scientific knowledge within a firm’s location is influenced by its historical environment and cultural development. Regions with the highest number of recent Chinese publications are presumed to have advanced scientific knowledge. The exogeneity of this instrumental variable is supported by the lack of direct impact from the number of books published in modern China on firms’ innovation output. In addition, the manual matching of address and number of books in modern China, which includes six categories: agricultural science, industrial technology and applications, natural science, medicine and health, electronics and general technology, automation and computer technology, and environmental science and safety science, is matched to each firm’s main business to obtain the number of relevant books for each firm, but this is a variable that does not change over time.

This paper uses panel data as the research sample, so using only the number of recent books as an instrumental variable is not feasible due to fixed effects. To construct the panel instrumental variable, we introduce a time series variable following the approach of Nunn and Qian (2014) [79]. We adopt the method of multiplying the instrumental variables by the industry averages of the explanatory variables [80,81], which is a common international practice for choosing instrumental variables [82]. We construct two sets of instrumental variables by calculating the average values based on secondary and tertiary industry classifications and multiplying them by the number of relevant recent books. The regression results of the two-stage least squares (2SLS) method are reported in

Table 7. Endogeneity test results on local knowledge.

Variables	First Stage	Second Stage	First Stage	Second Stage
	lnRlpub	lnpat	lnRlpub	lnpat
lnRlpub		0.209 **		0.203 ** (2.323)
		(2.340)
lniv_lpaper	0.813 ***		0.802 ***
	(9.64)		(9.54)
lniv_lbook1	0.257 ***
	(4.21)
lniv_lbook2			0.286 ***
			(4.92)
N	3344		3344
R2	0.128		0.129
Kleibergen-Paap rk LM	79.84 (0.0018)		82.39 (0.000)
Kleibergen-Paap Wald rk F statistic	151.9		159.9
Hansen J statistic	2.537 (0.111)		2.435 (0.119)
FirmFE	Yes		Yes
YearFE	Yes		Yes
Control	Yes		Yes

Notes: Variables lniv_lpaper, lniv_lbook1 and lniv_lpaper, lniv_lbook2 correspond to ${lnRlpu{b}_{IV2}}_{i,c,t}$, ${lnRlpu{b}_{IV1}}_{i,c,t}$ in Equation (2) and Equation (3), respectively. T statistics in parentheses. ** and *** denote significance at 5% and 1% levels, respectively.

and

Table 8. Endogeneity test results on YRD knowledge.

Variables	First Stage	Second Stage	First Stage	Second Stage
	lnRypub	lnpat	lnRypub	lnpat
lnRypub		0.469 **		0.409 *
		(1.971)		(1.832)
lniv_ybook3	0.156 ***		0.111 **
	(2.66)		(1.99)
lniv_ypaper1	0.768 ***
	(10.27)
lniv_ypaper2			0.857 ***
			(12.56)
N	935		935
R2	0.208		0.209
Kleibergen–Paap rk LM	50.83 (0.0000)		53.22 (0.000)
Kleibergen–Paap Wald rk F statistic	239.9		308
Hansen J statistic	1.325 (0.250)		0.919 (0.338)
FirmFE	Yes		Yes
YearFE	Yes		Yes
Control	Yes		Yes

Notes: Variables lniv_ybook3, lniv_ypaper1 and lniv_ypaper3, lniv_ypaper2 correspond to ${lnRypu{b}_{IV2}}_{i,c,t}$, ${lnRypu{b}_{IV1}}_{i,c,t}$ in Equation (4) and Equation (5), respectively. t statistics in parentheses. *, **, and *** denote significance at 10%, 5%, and 1% levels, respectively.

.

The first stage regression results in Table 7 indicate that both selected instrumental variables have a significant positive effect on related scientific knowledge output. To test the validity of instrumental variables, common tests have been employed. The initial hypothesis of under-identification of instrumental variables, assessed through the Kleibergen–Paap rk LM test, is rejected at the 1% significance level. The weak instrument test, conducted using the Kleibergen–Paap Wald rk F statistic, confirms that the instrumental variables are not weak at the 10% significance level, as the test value (79.84) exceeds the critical value (19.93). These results indicate that the selected instrumental variables are reasonable and valid. Furthermore, the regression results of the instrumental variables method align with the previous findings, suggesting the robustness of the empirical findings.

6.3. Heterogeneity Tests

There is symmetry in the scale structure and functional structure among cities. Shanghai, the largest city in the YRD, serves as an industrial center and a scientific center city. It surpasses others in basic science and industrial innovation, boasting 10 “985” and “211” universities, as well as 24 state key laboratories. In the fields of biochemical and numerical physics, it concentrates 11 national double first-class basic disciplines, contributing to 35.6% of scientific papers in the YRD. These scientific achievements represent the core output of high-end frontier basic research in the region. To effectively assess the innovation impact of science center cities on a city group with a large economic and population scale and a wide geographical range, a group regression analysis was conducted on the emerging industries of enterprises in Shanghai and 26 other cities. Additionally, the scientific papers output from Shanghai was excluded from the overall scientific papers related to the YRD. The results of this analysis are presented in

Table 9. Regression results by groups.

	Shanghai		26 Cities		Emerging Industries in 26 Cities
	(1)	(2)	(3)	(4)	(5)	(6)
	lnpat	lnpat	lnpat	lnpat	lnpat	lnpat
lnRlpub	−0.001		0.109 ***
	(−0.03)		(2.85)
lnRypub		−0.031		0.088	0.425 *
		(−0.47)		(1.02)	(1.86)
lnPubexpsh						0.436 **
						(2.19)
N	1342	1342	2233	2233	715	715
adj. R2	0.108	0.108	0.275	0.272	0.332	0.332
FirmFE	Yes	Yes	Yes	Yes	Yes	Yes
YearFE	Yes	Yes	Yes	Yes	Yes	Yes
Control	Yes	Yes	Yes	Yes	Yes	Yes

Notes: t statistics in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

.

In Column (3), the coefficients of relevant scientific papers in 26 cities within the YRD (excluding Shanghai) exhibit significant positive values, verifying Hypothesis 2. This finding highlights the phenomenon of knowledge dimensionality reduction among cities within the YRD during the transformation of scientific knowledge into industrial technology. Science center cities and non-science center cities play distinct roles in basic scientific breakthroughs and industrial innovation functions [83]. In Column (4), it is observed that industry-related scientific papers in the 26 cities of the YRD exhibit positive knowledge spillover effects in emerging industries. Furthermore, in Column (6), the analysis excludes papers published in Shanghai while considering the relevant knowledge within the urban agglomeration. The regression results in Columns (1) and (2) indicate that neither local scientific papers nor YRD scientific papers have a significant impact on the invention patents of enterprises in Shanghai. It is important to note that these results do not provide evidence of a low spillover effect of scientific knowledge in Shanghai, but rather suggest that not all cutting-edge scientific output is effectively translated into industrial technology [84].

Furthermore, the conversion of cutting-edge scientific achievements into industrial applications necessitates a sequential dimensionality reduction process. Within this process, the involvement of innovative companies, industries with comparative advantages, high-end cutting-edge scientific discoveries from top research universities, as well as collaboration between science-centered and non-science-centered cities, is crucial in establishing a pathway for industrial technology transformation. This observation suggests that innovation is not solely dependent on the knowledge creation and spillover within individual cities, but rather on the overall capacity for knowledge creation and spillover within urban agglomerations. The impact of knowledge spillovers on enterprise, industry, and city development is contingent upon the adaptation of knowledge, which serves as the foundation for learning, cooperation, and sharing among economic entities. This paper reaffirms the theoretical finding of Jie Tang and Wenyue Cui (2021) that the development of city clusters is conducive to the establishment of collaborative networks and division of labor among cities, and confirms the findings provided by Arant et al. (2019) that urban cluster created more opportunities for knowledge spillovers and promoted the innovative development of city clusters through the cooperation and sharing among enterprises, industries, and universities (research institutions) [46,85].

The significance of developing urban agglomerations lies in the establishment of collaborative networks and division of labor between cities, thereby creating more opportunities for knowledge spillovers through cooperation and sharing among enterprises, industries, universities (research institutions), and the collaboration between enterprises and universities.

6.4. Mechanism of City Diversification on Firm Innovation

The significance of diverse agglomeration in urban development, as highlighted by Jacobs (1969) [86], has gained widespread recognition. Industrial diversification plays a crucial role in facilitating knowledge spillover. In practice, due to the variety of innovation sources and the diversity of innovation knowledge, enterprises must integrate multidisciplinary and multi-industry technologies when translating scientific advancements into industrial technologies. Additionally, they require “face-to-face” collaboration with individuals possessing multidisciplinary and multi-industry backgrounds.

Table 10. Mechanism analysis.

	(1)	(2)
	lnpat	lnpat
c_lnRlpub	0.083 **	0.075 **
	(2.59)	(2.41)
c_lnCitydiv	1.731 ***	1.365 ***
	(4.12)	(3.32)
c.c_lnRlpub#c.c_ lnCitydiv	0.174 *	0.155 *
	(1.86)	(1.70)
lnEnrd		0.002
		(0.42)
lnAge		0.373 ***
		(2.86)
Intangible		0.590
		(1.08)
TobinsQ		−0.016
		(−1.29)
lnperGdp		0.355
		(1.64)
lnCityrd		0.072 **
		(2.20)
_cons	0.531 ***	−5.713 **
	(12.32)	(−2.28)
N	3457	3457
adj. R2	0.222	0.234
FirmFE	Yes	Yes
YearFE	Yes	Yes
Control	No	Yes

Notes: t statistics in parentheses. *, **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

presents the results of moderating mechanism.

The regression results presented in Table 9 demonstrate that the interaction term of urban industrial diversification and scientific output significantly and positively impacts corporate invention patents. This finding indicates that industrial diversification plays a positive moderating role in the relationship between knowledge spillovers and innovation performance during the stage of technological innovation. The moderating effect of industrial diversification and agglomeration on the promotion of knowledge creation, sharing, and spillover for enterprise innovation performance is particularly pronounced in local spaces. This can be attributed to the fact that innovation input and industry knowledge during the stage of technological innovation are highly dependent on spatial factors. The regression results confirm H3 presented in this paper.

Thus, we have verified all the three hypotheses about innovation in the YRD proposed in this paper. The hypotheses and the verification results are summarized in

Table 11. Hypothesis test results.

Number	Hypothesis	Results	Position
H1	Local scientific knowledge spills positively influence the innovation performance of firms.	Supported	Table 6
H2	The unified innovation network positively contributes to the innovation performance of firms.	Supported	Table 9
H3	Industrial diversification agglomeration has a positive moderating effect on the knowledge spillover process.	Supported	Table 10

. Our findings show that both local scientific knowledge and scientific knowledge from other cities in the YRD city cluster have significant positive effects on the innovation output of firms, especially in the emerging industries. Our findings confirm that the innovation function of cities in the urban agglomeration is hierarchical, and it implies that policy makers should pay more attention to the synergy and division of labor among cities. We also found that industrial diversification and agglomeration positively influence the process of converting knowledge into innovation by creating more combinatorial opportunities for knowledge spillovers.

6.5. Robust Check

To ensure the reliability of the findings, this study conducted robustness tests using the method of shrinking tails and eliminating special values. Firstly, the core explanatory variables, namely the relevant knowledge of the city and the YRD, were reduced to 2.5% to mitigate the influence of discrete values on the empirical results. Secondly, seven cities in Anhui Province, which had relatively low connection density with the innovation network of the YRD, were excluded from the analysis. The results of these two robustness tests, namely the tailing test and the elimination of special values, revealed that the regression coefficient for the impact of relevant scientific knowledge on the innovation performance of enterprise invention patents remained significantly positive. This indicates that the spillover effect of innovation knowledge in the YRD is still evident (refer to

Table A1. Robust check.

	(1)	(2)
	Core Variable Indentation	Remove Outliers
	lnpat	lnpat
lnRlpub	0.096 **	0.087 **
	(2.34)	(2.56)
lnRypub	0.371 ***	0.373 ***
	(2.76)	(3.01)
N	3273	3190
adj. R2	0.222	0.192
FirmFE	Yes	Yes
YearFE	Yes	Yes
Control	Yes	Yes

Notes: t statistics in parentheses. ** and *** denote significance at 5% and 1% levels, respectively.

in Appendix A).

7. Conclusions, Limitations, and Policy Implications

7.1. Conclusions

This paper provides a comprehensive review and expansion of existing theoretical achievements on knowledge spillovers and conducts an empirical analysis of the Yangtze River Delta urban agglomeration, which is the most important economic, scientific, and industrial innovation center in China. The study reveals that the concept of heterogeneous subjects can be extended to cities and urban agglomerations with diverse industrial clusters through scientific research and industrial clusters in the spatial economy. The dissemination and diversity of knowledge serve as the foundation for continuous sharing and creation of knowledge among heterogeneous subjects.

The correlation between basic research and applied research is evident in the process of knowledge dimensionality reduction from science to industrial technology. Enterprises, as the main drivers of industrial technological innovation, require both strong innovation motivation and capabilities. Motivation arises from fierce competition among peers, while capabilities are fostered within a knowledge network environment that facilitates learning and sharing. Industrial diversification and knowledge correlation act as bridges between different domains of knowledge, resulting in a low-to-high innovation synergy within hierarchical innovation networks. This process represents a sustainable flow of knowledge and continuous creation, akin to “the predecessors planting trees, and the descendants enjoying the shade”.

The overall knowledge creation in the Yangtze River Delta and its hierarchical knowledge spillover represent important practical and forward-looking theoretical issues. The Yangtze River Delta has developed into an innovation network spanning 27 cities. Knowledge overflow has given rise to subdivided innovation networks with unique characteristics, forming a systematic and hierarchical innovation network system that produces significant collaborative innovation effects. Innovation growth and convergence among cities have become important trends. The empirical analysis of knowledge spillover in the Yangtze River Delta reveals that cities at different levels exhibit distinct innovation dynamics. Science-centered cities such as Shanghai have successfully disseminated cutting-edge scientific knowledge to industries through numerous application-oriented universities and research institutions. Concentrated and stable science centers with outstanding scientific achievements support emerging industrial centers and cities with strong industrial innovation capabilities, leading to accelerated economic growth.

7.2. Limitations

This paper studies the relationship between knowledge creation, spillover, and innovation growth in the Yangtze River Delta urban agglomeration from a new perspective of microenterprises. Several directions still need to be studied in the future. From the perspective of sample selection, this paper does not include the time period after the epidemic due to the measurements used, but this period is still an important element that can be continuously followed up and studied, and it is also possible to observe any other characteristics of the post-pandemic conclusions on the knowledge and innovation networks of the city clusters. Future research could also consider exploring publications from different databases such as CNKI, and examining the differential impact of internationalization and frontiers of knowledge on innovation. Additionally, other urban cluster in China can be selected to carry out comparative analysis and research on the selection of urban cluster, which can provide evidence of different urban clusters in the relationship between regional scientific knowledge and innovation.

A future research area of interest is to explore mechanisms other than diversification of urban industries that open the process of scientific knowledge toward innovation in urban agglomerations. This will unlock secrets in the black box of knowledge overflow and provide more channels for an in-depth understanding of scientific knowledge toward innovation. Finally, although this paper does not focus on the heterogeneity of industries, differences among industries have different paths for scientific knowledge toward innovation. Therefore, exploring knowledge spillover from different industries to industrial innovation is also an important direction for future research.

7.3. Policy Implications

The creation of scientific knowledge in the urban cluster is a significant contributor to innovation development. Therefore, the implementation of the innovation-driven development strategy should pay sufficient attention to basic scientific research, especially the spillover impact of cutting-edge and leading new ideas from universities and R&D institutions on innovation output. However, basic scientific research and innovation may not be suitable for all cities, and appropriate regional innovation policies are needed to promote functional complementarities among cities. Innovation policies should also encourage cities with better scientific foundations in the region to create cutting-edge and leading scientific knowledge, strengthen their ties with cities at different levels, and enhance their spillover effects to drive innovation and economic growth within the urban cluster.

To improve innovation performance, enterprises should actively seek knowledge-related innovation partners within and across cities, and through innovation collaboration and communication, they can enhance their innovation performance and growth level. From the policy perspective, the government should actively create a platform for innovation collaboration and communication, facilitating the collaboration and communication between intra-regional and inter-regional innovation actors, to foster the development of a multi-level innovation network in the region. By establishing innovation networks and enhancing the relevance of their knowledge, the geographic barriers of knowledge spillover can be overcome, the innovation resources in the region can be efficiently exploited, and the performance of local innovation can be increased, thus achieving the economic growth of the whole urban cluster.

For cities or enterprises with a weak scientific foundation, we should encourage them to pursue more technological innovations with high relevance and applicability, and simultaneously leverage the resource advantages of regional industrial agglomeration, optimize the industrial layout, and create diversified industrial clusters by using their own advantageous industrial resources. This would help them to absorb and transform the scientific knowledge spillover from the region, enhance the overall innovation capacity of the region, develop industrial clusters, and improve the overall innovation performance of the urban cluster.