Exploring the Spatial Correlation Network Structure of Green Innovation Efficiency in the Yangtze River Delta, China

Abstract

Green innovation is crucial to high-quality economic development and has become an important engine for green transformation development in the Yangtze River Delta region. In this study, we used the super epsilon-based measure (super-EBM) to measure the green innovation efficiency of 26 cities in the Yangtze River Delta region of China from 2003 to 2018. Secondly, on the basis of a modified gravity model, we determined the spatial correlation of the Yangtze River Delta region’s green innovation efficiency and built a relation matrix. Finally, using the Moran index, centrality analysis, and a block model, we investigated its spatial characteristics and empirically analyzed its influencing factors by applying the quadratic assignment procedure. The results show that (1) in spatial terms, the Yangtze River Delta’s green innovation efficiency is extremely unbalanced, and the spatial network association density is low, only 0.218; (2) in terms of block analysis, the green innovation efficiency of the Yangtze River Delta region’s cities can be divided into four blocks, with distinct blocks holding distinct responsibilities; and (3) in terms of influencing factors, geographic distance; the expansion of the difference in energy consumption and the environment pollution index; and narrowing the gap in economic development, the industrial structure, and green coverage will push forward the formation of spatial correlation at a significance level of 10%. Our findings expand the research on traditional innovation efficiency and provide theoretical guidance for formulating regional green innovation coordinated development policy. It is necessary to strengthen urban green innovation cooperation in the Yangtze River Delta and promote regional integrated development. Different policies should be adopted for cities with different spatial correlation patterns. From the perspective of block analysis, it is necessary to balance the acceptance relationship and spillover relationship between cities. In addition, to improve the spatial correlation of green innovation efficiency in the Yangtze River Delta, the allocation of green innovation resources in the Yangtze River Delta should be optimized.

1. Introduction

In recent years, rapid industrialization and urbanization have created a dilemma between sustained economic growth and the deterioration of the living environment [1,2]. At present, China’s environmental pollution, the greenhouse effect, and other negative effects are more prominent in cities [3,4]. In this period of economic transition, Chinese President Xi Jinping has put forward five principles of development: innovation, coordination, green development, openness, and sharing [5]. Innovation is the driving force of development, and green development is beneficial to push forward humans and nature coexisting in harmony [6]. Therefore, for the purpose of improving green innovation efficiency, an effective approach is to bring green technology and environmental factors into the research framework of traditional innovation [7]. In September 2018, the first “China Green Innovation Conference” was held in Beijing. In this context, research on green innovation is increasing.

The Yangtze River Delta urban agglomeration includes Shanghai and several cities in the provinces of Jiangsu, Zhejiang, and Anhui. This region is the most economically developed urban agglomeration area in China. However, there are many high-pollution and high-energy consumption industries in the Yangtze River Delta region. Therefore, the Yangtze River Delta region is facing serious ecosystem damage, environmental pollution, and resource shortages, which limit high-quality economic development [8]. Studies have shown that green finance and fintech can affect high-quality economic development [9]. One paper used a spatial econometric model to conclude that environmental regulation, foreign direct investment, and their interactions are conducive to high-quality economic development [10]. And some studies showed that smart city policy can improve urban green total factor productivity by improving technical efficiency and achieve high-quality economic development [11,12]. Promoting green innovation and strengthening regional cooperation will also contribute to high-quality economic development [13]. Therefore, the application of green innovation in economic development can serve as a boost to promote the high-quality economic development of cities in the Yangtze River Delta. The Yangtze River Delta urban agglomeration development plan, released in 2016, encouraged economic transformation and innovative upgrading [14]. The development of the Yangtze River Delta should be included in the national strategy [15]. In this condition, taking the Yangtze River Delta urban agglomeration as an example, researching the spatial differentiation and spatial association characteristics of green innovation efficiency is typical and representative [16]. Such research is beneficial for the healthy and orderly development of the economy and innovation in the Yangtze River Delta region. In addition to expanding the study on traditional innovation efficiency, such research provides theoretical guidance for formulating regional green innovation-coordinated development policy.

Existing studies on green innovation activities mostly focus on the provincial level [17,18,19]. Empirical research at the urban scale is relatively scarce. In fact, cities are the best carriers of innovation ecosystems and the most effective unit for the government to carry out environmental governance [20]. The Yangtze River Delta region is one of the most developed regions in China [21]. Therefore, it is more operable to boost the integrated development in the Yangtze River Delta at the urban scale. In terms of the measurement of green innovation efficiency, most studies adopt radial models (such as the CCR model and the BCC model) or non-radial models (such as the SBM model) [22,23]. A radial model requires the input–output variables to be reduced and expanded in the same proportion, ignoring the different characteristics of variables. Although a non-radial model can consider the difference of various input–output variables, the initial proportion relationship of variables may be lost, which affects the accuracy of measurement results to a certain extent. In 2010, Tone and Tsutsui proposed the EBM model with both radial and non-radial characteristics, which made up for the deficiencies of both the radial model and the non-radial model [24]. In addition, as innovation has become the core driver of regional economic development [25], research on green innovation efficiency needs to be enriched. Specifically, in this study, we first used a super-EBM model to measure the green innovation efficiency of 26 cities in the Yangtze River Delta from 2003 to 2018. Then, we analyzed the spatial heterogeneity of green innovation efficiency through the Moran index, constructed the spatial correlation network through the modified gravity model, and analyzed its characteristics. Finally, QAP was used to analyze the relevant influencing factors of green innovation efficiency.

The contributions of this study compared with the existing literature are mainly reflected in the following four aspects. First, the super-EBM model was constructed to measure green innovation efficiency at the city level, which made up for the defects of traditional DEA models and improved the accuracy of efficiency measurement results. Secondly, we built a modified gravity model to identify the spatial correlation of the Yangtze River Delta region’s green innovation efficiency from the viewpoint of the network, and further analyzed its evolutionary laws in depth. Thirdly, we introduced a block model into the spatial association analysis, dividing the Yangtze River Delta region’s green innovation efficiency into four blocks to analyze the spatial association between the different blocks. Finally, quadratic assignment procedure (QAP) regression was used to examine the factors influencing the spatial association network with the ability to overcome the multicollinearity problem between independent variables, making the regression results more scientific and reasonable.

The rest of this paper is organized as follows. Section 2 is a literature review. Section 3 describes the research approaches of this study. Section 4 explores the spatial heterogeneity of green innovation efficiency. Section 5 evaluates the spatial correlation network of green innovation efficiency and analyzes its influencing factors. Section 6 summarizes and proposes policy implications.

2. Literature Review

Many scholars focus on green technology innovation because it is regarded as the vital method of reducing global pollutant emissions [26,27]. Various studies have researched themes related to green innovation concepts, green innovation efficiency measurement, drivers, and determinants [28,29].

Scholars have different understandings of the concept of green innovation [30]. Some scholars define it as innovation conforming to the trend of environmental improvement [31]. In addition, some scholars believe that green innovation is innovation that can realize the sustainable development of enterprises [32], or ecological innovation to improve environmental efficiency and promote sustainable development [33].

Existing studies have proposed a variety of green innovation efficiency measurement methods. Current research with regard to green innovation efficiency mostly evaluates efficiency by data envelopment analysis (DEA), and stochastic frontier analysis (SFA). The departure from the DMU frontier, according to SFA, is caused by stochastic disturbances and technical inefficiencies. The SFA was mainly applied to research enterprise efficiency and affecting factors, as well as economic research [34,35]. On the basis of the input and output of all DMUs in the sample, the DEA approach entails generating a minimal output possibility set that can adapt to all individual production modes. The input–output efficiency is then calculated using this set of possibilities. Carayannis et al. (2016) employed the multi-objective DEA model to assess the innovation efficiency of 185 regions in 23 European countries, and the results showed that innovation efficiency was not exactly the same in different regions and different stages [36]. Tao et al. (2016) measured China’s green economy efficiency through an SBM model considering CO₂ emissions and energy consumption [37]. Du, Liu et al. (2019) applied a two-stage network DEA with shared input to assess green technology innovation efficiency in regional enterprises [17]. They explored regional differences in industrial enterprises’ green technology R&D and the transformation efficiency toward green technology. Through a modified SBM model, Liu et al. (2020) measured the green technology innovation efficiency in China on the basis of innovation failures and environmental pollution [38]. Wang and Zhang (2021) studied the spatial characteristics of green innovation efficiency in 30 provinces in China from 2009 to 2017 based on the global super-EBM model [39].

Another focus area in the literature mainly researches the determinants or affecting factors of green innovation efficiency. In general, the affecting factors of green innovation efficiency can be divided into direct factors and indirect factors. Direct factors include labor quality, industrial structure, energy consumption, technological innovation and so on. Indirect factors include economic development, government funds, regional infrastructure, foreign direct investment, openness, and environmental regulation. Luo et al. (2019) explored the influence of international R&D capital technology spillover on green technology innovation efficiency by establishing a spatial model [40]. A scholar took the French automobile industry as an example to study the influencing factors of green innovation, and concluded that green innovation is affected by three aspects: the technical system, the demand condition system, and public policy [41]. One study found that external policy tools and internal enterprise factors play different roles in different types of green innovation in the U.K. [42]. Sun et al. (2021) applied data of 24 innovating countries from 1994 to 2013 to investigate the influence of technological innovation in some countries on the energy efficiency performance of neighboring countries. The results showed that knowledge spillovers can improve the energy efficiency performance of neighboring countries [43]. Long et al. (2020) analyzed the evolution and driving factors of green innovation efficiency in the Yangtze River Economic Belt through establishing the super-SBM model and the spatial Durbin model [44]. Research shows that environmental regulation is a significant driving force of green innovation [45]. Yuan and Xiang (2018) confirmed that environmental regulation has improved the energy efficiency and environmental efficiency of the manufacturing industry in the short term, and attracted more R&D investment in the long term [46]. Hong et al. (2019) found that cooperative innovation among organizations, governments, and institutions has a significantly positive impact on innovation performance [47].

To sum up, studies on green innovation efficiency at home and abroad mainly focus on theoretical elaboration, efficiency evaluation, and influencing factors. Spatial agglomeration and spillover effects are significant characteristics of green innovation efficiency [48]. However, there is still a lack of research on the spatial relationship of green innovation efficiency, and the research methods are mainly Moran’s I index, the spatial Durbin model, and other spatial econometric theories and technical methods [49,50,51], which tend to attach importance to the attribute relationship between data. The investigation and application of relational data are ignored. The social network analysis method can investigate relational data and network relations, which can effectively supplement the research on spatial relationships of green innovation efficiency. Here, we used the super-EBM model to measure the green innovation efficiency of the Yangtze River Delta region, and further used the social network analysis method to build the green innovation efficiency of a spatial association network. In addition, this we described the characteristics of the spatial association network and analyzed its formation mechanism in order to provide theoretical guidance and decision-making reference for the sustainable development of the Yangtze River Delta region.

3. Methodology

3.1. Super Epsilon-Based Measure Model with Undesirable Outputs

The DEA method was applied to measure the green innovation efficiency of the cities in the Yangtze River Delta in this study. DEA is a mathematical process using linear programming to evaluate the efficiency of DMUs. It can solve the problem of multiple inputs and outputs in a unified framework. From a measurement approaches point of view, classical DEA models can be classified into two types [52]. The first type is represented by the Charnes–Cooper–Rhodes (CCR) model based on the radial measure. It was used to evaluate the scale and technical effectiveness of DMUs. The radial CCR model measures the efficiency score of input-oriented factors under the premise of keeping the output unchanged. However, due to overly strict assumptions, all input factors must be reduced in equal proportions, and this deviates from the real economy. The second type of DEA model is a slack-based measure (SBM) based on a non-radial measure [53]. The efficiency measure of an SBM model contains non-radial relaxation variables, which avoids the assumption that the input factors are decreased by the same proportion; this not only solves the problem of input–output slack, but also solves the problem of efficiency evaluation with undesired output [54]. However, the cost of this optimization is the loss of the original proportion information of the projected decision-making unit (DMU). In addition, to solve the linear programming, the SBM model exposes its shortcoming that there is a significant difference between the optimal relaxation of a zero value and a positive value. Thus, both radial and non-radial models have advantages and defects. To effectively solve the questions of the CCR model and the SBM model in measuring efficiency scores, Tone and Tsutsui (2010) proposed an Epsilon-Based Measure (EBM) model with radial and non-radial characteristics [24]. The EBM model is expressed in the following calculation formulas:

$$\begin{gathered} {\gamma }^{*}=\min \theta -{\epsilon }_x{{\displaystyle \sum }}_{i=1}^m\frac{w_i^{-}{s}_i^{-}}{x_{ik}} \\ s.t.\{\begin{array}{c}{\displaystyle {\displaystyle \sum }_{j=1}^n}{\lambda }_j{x}_{ij}+s_i^{-}=\theta x_{ik},i=1,2,\dots ,m\\ {\displaystyle {\displaystyle \sum }_{j=1}^n}{\lambda }_j{y}_{rj}\ge y_{rk},r=1,2,\dots ,s\\ {\lambda }_j\ge 0, s_i^{-}\ge 0\end{array} \end{gathered}$$

(1)

Assume that there are n DMUs, and each DMU includes m input f, x = (x_1j, x_2j, …, x_mj) and generates s outputs, y = (y_1j, y_2j, …, y_sj), where $w_i^{-}$ represents the weight of input i and meets the condition of ${{\displaystyle \sum }}_{i=1}^m{w}_i^{-}=1\left(w_i^{-}\ge 0\right)$.

When an undesirable output is considered, it is necessary to extend the EBM model of Equation (1) to a non-oriented EBM model on the basis of undesirable outputs. This non-oriented EBM model is expressed in the following calculation formulas:

$$\begin{gathered} {\gamma }^{*}=\min \frac{\theta -{\epsilon }_x{{\displaystyle \sum }}_{i=1}^m\frac{w_i^{-}{s}_i^{-}}{x_{ik}}}{\phi +{\epsilon }_y{{\displaystyle \sum }}_{r=1}^s\frac{w_r^{+}{s}_r^{+}}{y_{rk}}+{\epsilon }_b{{\displaystyle \sum }}_{p=1}^q\frac{w_p^{b-}{s}_p^{b-}}{b_{pk}}} \\ s.t.\{\begin{array}{c}{\displaystyle {\displaystyle \sum }_{j=1}^n}{\lambda }_j{x}_{ij}+s_i^{-}=\theta x_{ik},i=1,2,\dots ,m\\ {\displaystyle {\displaystyle \sum }_{j=1}^n}{\lambda }_j{y}_{rj}-s_r^{+}=\phi y_{rk},r=1,2,\dots ,s\\ {\displaystyle {\displaystyle \sum }_{p=1}^n}{\lambda }_j{b}_{ij}+s_p^{b-}=\phi b_{pk},p=1,2,\dots ,q\\ {\lambda }_j\ge 0, s_i^{-},s_r^{+},s_p^{b-}\ge 0\end{array} \end{gathered}$$

(2)

where q denotes the number of undesirable outputs; b = (b_1j, b_2j, …, b_pj); $\lambda$ is the intensity vector; $s_r^{+}$ stands for the non-radial output slacks; $w_r^{+}$ stands for the weight of output r; $s_p^{b-}$ represents the non-radial undesirable output slacks; $w_p^{b-}$ stands for the weight of the undesirable output p; $\epsilon$ denotes the relative significance of the non-radial slacks over the radial $\phi$; and $\epsilon \in \left[0, 1\right]$. If $\epsilon =0$, the model denotes a radial model, and if $\epsilon =1$, this model represents an SBM model. To make it comparable to a DMU with an efficiency of 1, the super-efficiency model is introduced to the EBM model. The core idea of the super-efficiency model is to replace the input–output of the k-th DMU with the input–output of all other DMUs when calculating the efficiency of the k-th DMU, so that the k-th DMU is eliminated. Combining the merits of the EBM model and the super-efficiency DEA model, we measured the green innovation efficiency using this super-EBM model.

3.2. Spatial Association Analysis

In essence, spatial association analysis is used to analyze the spatial interaction features between distinct areas. On the basis of Tobler’s First Law of Geography, the closer the geographic distance between two areas, the stronger their spatial dependence. For the purpose of measuring the spatial dependence between 26 cities, Moran’s I was adopted to study the spatial association. The spatial association is classified into global and local spatial autocorrelation. In this study, global and local Moran’s I were used to research the global and local spatial autocorrelation. They are expressed in the following calculation formulas:

$${\mathrm{Global}\mathrm{Moran}}^{’}\mathrm{s}I=\frac{n{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^n{w}_{ij}\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^n{w}_{ij}{{\displaystyle \sum }}_{i=1}^n{\left(x_i-\overline{x}\right)}^2}$$

(3)

$${\mathrm{Local}\mathrm{Moran}}^{’}\mathrm{s}I=\frac{n\left(x_i-\overline{x}\right){{\displaystyle \sum }}_{j=1}^n{w}_{ij}\left(x_j-\overline{x}\right)}{{{\displaystyle \sum }}_{i=1}^n{\left(x_i-\overline{x}\right)}^2}$$

(4)

where $w_{ij }$ stands for the spatial weight matrix, $n$ represents the number of cities, $x_i$ and $x_j$ represent the sample values. The value of the global and local Moran’s I indexes are both between −1 and 1.

3.3. Social Network Analysis

3.3.1. Spatial Association Network

A spatial association network was used to show the internal relations within the green innovation efficiency network of the Yangtze River Delta. The dots in the spatial association network stand for the Yangtze River Delta region’s 26 cities, and the lines in the network denote that the green innovation efficiency between these cities has spatial associations. This combination of dots and lines forms a spatial association network. Specifically, a modified gravity model was used to establish the spatial association network, and we used UCINET software to explore the features of the overall network and each individual cities’ characteristics in the network. Then, the QAP approach was applied to analyze which factors can affect the formation of the spatial association network.

Social network analysis is an important approach applied to research on the structure and pattern of social relations, with graph theory and algebra as the theoretical basis and the relation among social actors as the basic analytical unit. This method has been widely applied in sociology, economics, and other fields. Social network analysis works through constructing an association matrix between points. In general, the social network is identified by the Granger causality test based on the VAR method or the gravity model. However, the VAR method cannot be used to construct an annual association matrix. We aimed here to determine the dynamic trends of the features of the whole network in the sample research period. Thus, we selected the gravity model to identify the association relationship. The traditional gravity model is expressed in the following calculation formulas:

$$F_{ij}=\frac{G{M}_i{M}_j}{D_{ij}^2}$$

(5)

where F_ij stands for the attraction of the city i to city j; G denotes the coefficient of attraction between cities, which is generally 1; M_i and M_j represent some specific factors; and D_ij represents the space linear distance between city i and city j. In this paper, the traditional gravity model is modified, and the formula is as follows:

$$S_{ij}=K_{ij}\frac{E_i{E}_j}{D_{ij}^2}, K_{ij}=\frac{G_i}{G_i+G_j}$$

(6)

where E_i and E_j denote the green innovation efficiency and G_i and G_j denote the gross domestic product of city i and city j, respectively.

3.3.2. Indicators of the Spatial Association Network

(1)

Network Density

It can measure the strength of the association between individual cities in the network. In a spatial association network, the number of associations has a significantly positive association with network density, i.e., the greater the number of relationships, the greater the network density. The network density is expressed in the following calculation formulas:

$$D=\frac{L}{N\left(N-1\right)}$$

(7)

where L denotes the number of existing associations in the network, and N denotes the number of cities in the network, $D\in \left[0,1\right]$.

(2)

Centrality Analysis

The concept of centrality is related to the role of the social network in society. A city with a higher centrality indicates a close relationship with other cities. It is determined by three indicators: closeness, betweenness, and centrality. The degree centrality is an indicator to show the number of ties that are related to a city within the spatial association network:

$$D_e=\frac{n}{N-1}$$

(8)

where D_e denotes the degree centrality, n represents the number of cities correlated with the city, and N denotes the number of cities within the spatial association network. The betweenness centrality (B_c) represents the degree to which a city is at the center of the spatial association network. Cities with a larger B_c are more closely connected to other cities, and their positions in the spatial association network are more concentrated. The closeness centrality (C_c) indicates the degree to which a city is not influenced by other cities in the spatial association network. Cities with larger C_c have a more pronounced effect on other cities and are more likely to be central participants within the spatial association network.

3.4. Block Models

White et al. (1976) first put forward the block model to research social network relationships [55]. The situation of the individual city within a spatial association network can be easily researched on the basis of the block model theory. The positional relationship might take four different forms. Firstly, the bilateral spillover block; the most distinguishing feature of this block is that it has both sending and receiving relationships with other blocks, as well as more relationships with its own member. The main benefit block follows the second. The receiving connection from both within and outside the block is significantly bigger than the overflow to the outside of the block, which is the block’s fundamental attribute. The net spillover block is the third one. This block’s most distinguishing feature is that it sends out more connections than it receives from other blocks. The broker block is the fourth block. The obvious feature of it is that it has both receiving and sending relationships, unlike other blocks, and it has a small number of internal links within the block. Based on the above, a particular classification of the four primary blocks was determined, as illustrated in

Table 1. Specific classification of the four major blocks.

Internal Relationship Ratio	Received Relationship Ratio
	≈0	>0
≥(gk − 1)/(g − 1)	Bilateral spillover block	Main/net benefit block
<(gk − 1)/(g − 1)	Net spillover block	Broker block

.

In Table 1, assuming that block G_k includes g_k cities, the total number of possible relationships in block G_k is g_k(g_k − 1). The spatial association network includes g cities; thus, the number of relationships that may exist in each city of block G_k in the spatial association network is g_k(g − 1). Therefore, the expected internal relationship ratio is g_k(g_k − 1)/g_k(g − 1) = (g_k − 1)/(g − 1).

3.5. Variables and Data Sources

3.5.1. Input Variables

“Green” attributes for calculating green innovation efficiency were evaluated based on the measurement of innovation efficiency. Due to green innovation efficiency evaluating innovation efficiency from the standpoint of inputs, the energy input must also be addressed when assessing innovation inputs. R&D personnel and R&D capital are two of the innovation inputs. The full-time equivalent of research and experimental development personnel (RDP) were selected to measure the R&D people input, and the internal spending of research and experimental development expenditure (RDE) as R&D. The energy input was calculated by converting energy consumption into 10,000 tons of standard coal.

3.5.2. Output Variables

Output variables are classified into expected output and undesired output. The expected output includes economic output, innovation output, and ecological income. The GDP after constant price adjustment in 2003 was used to represent the economic output. The number of patent applications was used to stand for the innovation output. The green coverage rate of the urban built-up area was used to stand for ecological income. The creation of an innovative output frequently results in some level of environmental pollution during the manufacturing process. Therefore, the comprehensive index of environmental pollution stands for the degree of environmental pollution, which was obtained by an entropy method of calculating industrial wastewater discharge, solid waste output, and waste gas discharge. The green innovation efficiency measurement framework is shown in Figure 1.

3.5.3. Data Sources

According to the development plan for the Yangtze River Delta urban agglomeration, the research object area includes 26 cities: Shanghai; Nanjing, Wuxi, Changzhou, Suzhou, Nantong, Yancheng, Yangzhou, Zhenjiang, and Taizhou in Jiangsu Province; Hangzhou, Ningbo, Jiaxing, Huzhou, Shaoxing, Jinhua, Zhoushan, and Taizhou in Zhejiang Province; and Hefei, Wuhu, Maanshan, Tongling, Anqing, Chuzhou, Chizhou, and Xuancheng in Anhui Province. The research period of this paper is from 2003 to 2018, a total of 16 years. In consideration of the authority, accessibility, and continuity of data, all the data come from the statistical yearbooks (2002–2019) and statistical bulletins (2002–2019) of the 26 cities, and from the China Urban Statistical Yearbook (2002–2019), the China Energy Statistical Yearbook (2002–2019), and the State Patent Office. The vector map data for the Yangtze River Delta were from the National Basic Geographic Information Center.

4. Spatial Heterogeneity Analysis of Green Innovation Efficiency in the Yangtze River Delta Region

4.1. Analysis of Spatial Distribution Characteristics

The super-EBM model was used to measure the cities’ green innovation efficiency in the Yangtze River Delta region from 2003 to 2018, which was calculated using MaxDEA.

Because of space constraints, this paper only includes the analysis of 2003, 2011, and 2018, as shown in Figure 2. In 2003, the green innovation efficiency of Shanghai, Suzhou, Jinhua ranked highly, while Hefei, Chizhou, and Tongling ranked low. In 2011, the green innovation efficiency of Suzhou and Shanghai were at the forefront, while Chizhou, Tongling, and Xuancheng had the lowest green innovation efficiency. In 2018, the cities on the frontier were Zhoushan, Shanghai, and Wuxi, and the green innovation efficiency of Tongling, Anqing, and Chizhou ranked low.

4.2. Spatial Autocorrelation Analysis of Green Innovation Efficiency

4.2.1. Global Spatial Autocorrelation Analysis of Green Innovation Efficiency

The global Moran’s I of the Yangtze River Delta region’s green innovation efficiency from 2003 to 2018 was calculated by ArcGIS 10.7; the results are shown in

Table 2. Global Moran index of the green innovation efficiency in the Yangtze River Delta.

Green Innovation Efficiency	Moran’s I	Z-Value	p-Value
2003	0.074	3.287	0.001
2004	0.265	6.322	0.000
2005	0.094	3.562	0.000
2006	0.068	2.813	0.005
2007	0.111	3.400	0.001
2008	0.138	3.945	0.000
2009	0.115	3.561	0.000
2010	0.099	3.089	0.002
2011	0.135	3.920	0.000
2012	0.161	4.406	0.000
2013	0.146	3.925	0.000
2014	0.158	4.111	0.000
2015	0.171	4.379	0.000
2016	0.130	3.521	0.000
2017	−0.006	0.745	0.456
2018	0.153	4.067	0.000

.

It can be seen from Table 2 that from 2003 to 2018 (except in 2017), at a significance level of 1%, the global Moran’s I of the green innovation efficiency in the Yangtze River Delta was greater than 0. This shows that the green innovation efficiency in the Yangtze River Delta was significantly spatially positively autocorrelated.

4.2.2. Local Spatial Auto-Association Analysis of Green Innovation Efficiency

The global Moran’s I index only reveals whether there is spatial autocorrelation of green innovation efficiency on the whole, but cannot scientifically reveal the local spatial autocorrelation of each direction. Therefore, it is necessary to further use local Moran’s I index, Moran’s I scatter plot, and LISA agglomeration plot to investigate the possible local autocorrelation of green innovation efficiency and the contribution degree of each city to the global spatial autocorrelation, and identify high–high agglomeration areas and low–low agglomeration areas. A local Moran scatter figure of the Yangtze River Delta region’s green innovation efficiency was generated by MATLAB software. The patterns of spatial association were determined. Because of the limitation of space, this paper only includes results for 2003, 2011, and 2018, as indicated in Figure 3.

Figure 3 displays the local Moran’s I scatter diagrams of the Yangtze River Delta region’s green innovation efficiency in 2003, 2011, and 2018. In the local Moran’s I scatter diagram, the first quadrant represents cities that have high green innovation efficiency surrounded by cities that have high green innovation efficiency (H-H), which is a cluster with a positive spatial correlation. In the second quadrant, cities that have low green innovation efficiency are surrounded by cities that have high green innovation efficiency (L-H), which is a cluster with a negative spatial correlation. In the third quadrant, cities that have low green innovation efficiency are surrounded by cities that have low green innovation efficiency (L-L), which is a cluster with a positive spatial correlation. In the fourth quadrant, cities with high green innovation efficiency are surrounded by cities that have low green innovation efficiency (H-L), which is a cluster with a negative spatial correlation. In 2003 and 2011, the green innovation efficiency of 21 cities had a positive spatial association; there were nine cities in the first quadrant, showing an H-H spatial association mode, and 12 cities in the third quadrant, showing an L-L spatial association mode. In 2018, there were 18 cities with a positive spatial association in the Yangtze River Delta region’s green innovation efficiency. Ten of these cities were in the first quadrant, showing an H-H spatial association mode. Eight cities were located in the third quadrant, showing an L-L spatial association mode. This shows that green innovation efficiency in the majority of cities had a positive spatial association with geographic space.

Figure 4 shows the evolutions in the temporal and spatial distribution of the Yangtze River Delta region’s green innovation efficiency in 2003, 2011, and 2018. The spatial dynamic transition process of green innovation efficiency was deeply explored using the time–space transition measurement approach [56]. The different types of green innovation efficiency dynamic transitions are shown in Figure 4, illustrating that cities in the east of the Yangtze River Delta had an H-H spatial association mode, while those in the west had an L-L spatial association mode. The appearance of this spatial association indicates that green innovation efficiency in the Yangtze River Delta region remains spatially unbalanced and has significant spatial dependence characteristics.

5. Analysis of the Green Innovation Efficiency Spatial Association Network

5.1. Establishment of the Spatial Association Network

Based on the modified gravity model, we identified the spatial association of the Yangtze River Delta region’s green innovation efficiency and established a relation matrix. UCINET software was used to depict the construction of the spatial association network, as shown in Figure 5, which shows that the spatial association indicates a representative network structure.

In Figure 5, the spatial spillover channel of the Yangtze River Delta region’s green innovation efficiency was 137, indicating that the Yangtze River Delta region’s green innovation efficiency was generally connected between different cities.

5.2. Spatial Association Network Characteristics

Theoretically speaking, the maximum number of relationships between the Yangtze River Delta region’s 26 cities was 650. The real number of spatial relationships in the Yangtze River Delta region’s cities was 137. Therefore, the association network density in each city was 0.2108. Compared with the maximum value of 1, the network density value is low, indicating that the association relationship formed in green innovation activities between cities is sparse, and the network belongs to the type of low-density network.

There are two situations in the relationships among cities: the overflow relationship, and the benefit relationship.

Table 3. Centrality analysis of the spatial association network of green innovation efficiency.

City	Spillover	Benefit	Total	City	Spillover	Benefit	Total
Shanghai	8	4	12	Huzhou	6	13	19
Nanjing	6	3	9	Shaoxing	3	8	11
Wuxi	4	7	11	Jinhua	8	3	11
Changzhou	8	7	15	Zhoushan	7	4	11
Suzhou	5	7	12	Taaizhou	6	3	9
Nantong	9	4	13	Hefei	9	0	9
Yancheng	7	1	8	Wuhu	3	5	8
Yangzhou	2	7	9	Maanshan	4	6	10
Zhenjiang	2	8	10	Tongling	3	4	7
Taizhou	3	7	10	Anqing	2	4	6
Hangzhou	3	5	8	Chuzhou	6	6	12
Ningbo	9	3	12	Chizhou	2	4	6
Jiaxing	5	9	14	Xuancheng	7	5	12

shows that Huzhou has the most association relationships, 19 overall, with 13 beneficiary relationships and six overflow relationships. Anqing and Chizhou had the fewest association relationships, at six. Therefore, in general, Huzhou benefited, while Nantong, Ningbo, and Hefei were overflowing.

On this basis, we perform a central analysis of the spatial association network of green innovation efficiency of the 26 cities in the Yangtze River Delta region. The degree, closeness, and betweenness were calculated and are shown in Figure 6. The degree centrality of Huzhou, Changzhou, and Jiaxing were higher than others. In the green innovation efficiency spatial association network, most associations were directly connected to these cities from the core of the spatial association network. The degree centrality of Tongling, Anqing, and Chizhou are lower than others, indicating that those cities had fewer associations with other cities. The closeness of Huzhou, Xuancheng, and Changzhou ranked highly, which indicates that these cities were influenced by other cities in the green innovation efficiency spatial association network. The closeness of Tongling, Anqing, and Chizhou ranked low, which meant that they had comparatively weak spatial associations with other cities and stayed away from the center of spatial association network. In terms of the betweenness centrality, Xuancheng and Huzhou ranked highly, and Hangzhou, Taizhou, Tongling, Anqing, and Chizhou ranked low. The central standard deviation of the Yangtze River Delta region’s green innovation efficiency spatial association network was 6.783, which is comparatively large, meaning that there was a large imbalance in the Yangtze River Delta region’s green innovation efficiency spatial association network.

According to the centrality analysis, Huzhou and other cities located in the center of the Yangtze River Delta region have higher centrality, which may be due to the fact that central cities, as bridges connecting surrounding cities, are closely connected with other surrounding cities. These cities play important roles in the spillover and beneficiary relations in the spatial association network. However, Tongling, Anqing, and other cities located at the edge of the Yangtze River Delta region generally have a low degree of centrality, which may be due to the fact that it is not easy for them to obtain spillover effects from surrounding cities.

5.3. Spatial Association Network Block Model

The CONCOR approach was applied to conduct the block model analysis of cities in the Yangtze River Delta in this paper. The maximum segmentation depth was 2, and the concentration standard was 0.200. The 26 cities were divided into four blocks, with the cities contained in each block shown in

Table 4. City distribution of the four blocks.

Blocks	City
First	Shanghai, Suzhou, Wuxi, Changzhou, Nantong (five cities)
Second	Shaoxing, Jiaxing, Huzhou, Ningbo, Jinhua, Hangzhou, Taaizhou, Zhoushan (eight cities)
Third	Yancheng, Zhenjiang, Taizhou, Yangzhou, Nanjing, Chuzhou (six cities)
Fourth	Hefei, Wuhu, Tongling, Anqing, Maanshan, Chizhou, Xuancheng (seven cities)

. The first block includes Shanghai and five cities in Jiangsu Province. The eight cities in Zhejiang belong to the second block. The remaining five cities in Jiangsu and Chuzhou in Anhui constitute the third block. The fourth block includes seven cities in Anhui Province.

Table 5. Analysis of spillover effects among the four blocks.

Blocks	Receiving Relationship				Number of Members	Expected Internal Relation Ratio (%)	Actual Internal Relation Ratio (%)	Received External Relations	Feature
	First	Second	Third	Fourth
First	17	9	7	1	5	16%	50%	12	Bilateral spillover
Second	8	38	0	1	8	28%	81%	10	Broker
Third	4	0	18	4	6	20%	69%	14	Main benefit
Fourth	0	1	7	22	7	24%	73%	5	Net benefit

shows that of the 137 spatial correlations, there were 95 internal associations inside the four green innovation efficiency blocks and 42 associations between the four green innovation efficiency blocks. The first block issued 34 relationships, with 17 of them belonging to the block and 12 to other blocks. In this block, internal relationship ratios were projected to be 16%, whereas the real internal relationship ratio was 50%. As a result, “bilateral spillover” was a feature of the first green innovation efficiency block. The second block issued 47 relationships, with 38 of them belonging to the block and 10 to other blocks. Internal relationship ratios were projected to be 28%, but they were 81%. Therefore, the second green innovation efficiency block functioned as a standard “broker”, which acts as a link and mediator. The third green innovation efficiency block issued a total of 26 relationships, 18 of which belonged to the block and 14 of which were received from other blocks. Internal relationship ratios were projected to be 20%; however, the real internal relationship ratio was 69%. As a result, the third green innovation efficiency block is the “main spillover”. The fourth green innovation efficiency block issued a total of 30 relationships, 22 of which belonged to the block and six of which were received from other blocks. Internal relationship ratios were projected to be 24%, but they were in fact 73%. Therefore, the fourth green innovation efficiency block primarily generated spillover effects outside the blocks, while receiving fewer relationships from other blocks, for a “net spillover” effect.

On the basis of the distinct associations between green innovation efficiency blocks, a density matrix between each block was calculated in this paper. In addition, we transformed the multi-valued density into an image matrix. The 26 cities’ aggregate spatial association network density was 0.2108. The density level was more than the total association network level when the green innovation efficiency block density was greater than 0.2108, and it was awarded a value of 1. Otherwise, it was given a value of 0. The results are shown in

Table 6. Density matrix and image matrix.

	Density Matrix				Image Matrix
	First Block	Second Block	Third Block	Fourth Block	First Block	Second Block	Third Block	Fourth Block
First Block	0.850	0.225	0.233	0.029	1	1	1	0
Second Block	0.200	0.679	0.000	0.018	0	1	0	0
Third Block	0.133	0.000	0.600	0.095	0	0	1	0
Fourth Block	0.000	0.018	0.167	0.524	0	0	0	1

. The density matrix shows the distribution of green innovation efficiency spillover effects. The magnitude of the value and the intensity of the spillover impact was proportionate in the density matrix. It shows that the first block’s spillover effect was primarily reflected in the interiors of the first, second, and third blocks, and the spillover effect on the second block was relatively small. The second, third, and fourth blocks’ spillover effects were mostly reflected in those interior blocks. In the image matrix, the first block was the diagonal and assigned a value of 1, and other positions were assigned a value of 1, as well. Thus, it can influence the interior of the green innovation efficiency block and have a spillover effect on other blocks. However, other blocks were only the diagonal assigned a value of 1, indicating that they merely influenced the interior of the green innovation efficiency block. This illustrates that the green innovation efficiency in the block has a significant association and presents an obvious “club” effect.

Figure 7 shows the internal links between the Yangtze River Delta region’s green innovation efficiency blocks. Among them, the driving force of the green innovation efficiency is the first block. The fourth block acts as a bridge, and it is a mediation in the transport mechanism. The existence of such a transmission mechanism shows the significant “gradient” features between green innovation efficiency blocks.

5.4. Factors Influencing the Spatial Association Network

5.4.1. QAP Association Analysis

QAP association analysis is a nonparametric test that compares the similarity of the values in two matrices using the permutation of matrix data. The association coefficient between the geographical association matrix of green innovation efficiency and its affecting factors was calculated using UCINET software, which randomly permuted the matrix data 5000 times.

Table 7. QAP correlation analysis between the spatial association of green innovation efficiency and its influencing factors.

Independent	Coefficient	Significance	Std. Dev.	Min	Max	Prop ≥ 0	Prop ≤ 0
Distance	0.553	0.000	0.047	−0.127	0.214	0.000	1.000
GDP	−0.230	0.000	0.045	−0.213	0.150	1.000	0.000
Energy consumption	−0.158	0.000	0.042	−0.179	0.162	1.000	0.000
Industrial structure	−0.093	0.029	0.045	−0.210	0.134	0.972	0.029
Patent applications	−0.236	0.000	0.044	−0.204	0.156	1.000	0.000
Green coverage	−0.099	0.015	0.044	−0.172	0.145	0.985	0.015
Environment pollution	−0.082	0.032	0.043	−0.172	0.154	0.969	0.032

shows the specific results, with “Min” and “Max” denoting the minimum and the maximum association coefficients acquired by the matrix permutation, respectively. “Prop ≥ 0” and “Prop ≤ 0” denote the proportion of the association coefficients acquired by the matrix permutation greater than or equal to and less than or equal to the actual association coefficient, respectively.

In Table 7, the association coefficient between the geographic distance and the spatial association is significantly positive at the 1% level, which indicates that there was a positive association between them. QAP association results display that the association coefficient between the other influencing factors and the spatial association of regional development is less than zero and that it is significant at the level of 1%. This indicates that economic development, the consumption of energy, the industrial structure, the number of patent applications, the green area coverage, and the environmental pollution index have significantly negative effects on the spatial association, among which the absolute value of the association coefficient of distance is 0.553. This value is obviously higher than other variables, which shows that the distance between cities can affect the spatial association of the green innovation efficiency network.

5.4.2. QAP Regression Analysis

The regression relationships among numerous independent variable matrices and a single dependent variable matrix are investigated through QAP regression. The above seven influencing factors with substantial association coefficients were selected as the independent variables for QAP regression analysis based on the results of the QAP association study. Then, we permuted data 2000 times randomly using UCINET software. The QAP regression results are displayed in

Table 8. QAP regression analysis of the factors influencing the spatial association of green innovation efficiency.

Independent	Unstandardized Coefficient	Standardized Coefficient	Significance	Proportion as Large	Proportion as Small
Intercept	0.430	0.000
Distance	0.041	0.000	0.002	0.002	0.999
GDP	−0.065	−0.498	0.003	0.998	0.003
Energy consumption	1.608	0.892	0.000	0.000	1.000
Industrial structure	−0.004	−0.063	0.089	0.912	0.089
Patent applications	−0.000	−0.594	0.106	0.894	0.106
Green coverage	−0.042	−0.111	0.020	0.981	0.020
Environment pollution	0.236	0.071	0.094	0.094	0.906

.

The regression coefficients of the relation matrix variables were then examined, as well as their significance test findings. In Table 8, “Proportion as large” refers to the proportion of the absolute value of the regression coefficients acquired by random permutation that is not smaller than the observed regression coefficient, and “Proportion as small” refers to the proportion of the absolute value of the coefficients acquired by random permutation that is not larger than the observed coefficient. The results demonstrate that the geographic distance coefficient, as well as disparities in energy consumption and the environmental pollution index, are all considerably positive at the 10% level, which indicates that they play significant roles in boosting the formation of the spatial association network. However, the coefficient of economic development, the industrial structure, and the green area coverage are considerably negative at the 10% level, meaning that the decline in the differences between these factors is beneficial to form the spatial association network.

6. Conclusions and Policy Implications

On the basis of existing research, we applied spatial autocorrelation analysis, social network analysis, and QAP association analysis to research the spatial association of the Yangtze River Delta and establish an association network of the region’s green innovation efficiency using panel data from 2003 to 2018. First, the super-EBM model was applied to calculate the green innovation efficiency of 26 cities in the Yangtze River Delta, and the global spatial autocorrelation of the green innovation efficiency was researched, applying the global Moran index. Second, the local agglomeration features of the green innovation efficiency were explored by local Moran index scatter diagrams. Third, a spatial association network in the Yangtze River Delta region’s green innovation efficiency was constructed through a modified gravity model, and its overall features were analyzed to identify the position and function of each city in the spatial association network. Fourth, we researched the division of the Yangtze River Delta region’s green innovation efficiency and the internal association mechanism by block model analysis. Finally, the QAP was applied to explore the factors that influenced the formation of the spatial association network. Specifically, we came to the following conclusions.

(1) The spatial distribution of the Yangtze River Delta region’s green innovation efficiency varies greatly. Shanghai was at the forefront every year; Suzhou and Wuxi were also at the forefront for several years. The cities with lower green innovation efficiency were Anqing, Xuancheng, Tongling, and Chizhou.

(2) The analysis of local spatial autocorrelation showed that the cities in the east of the Yangtze River Delta region were mainly of the H-H spatial association mode, but those in the west had an L-L spatial association mode. This showed that the green innovation efficiency in the Yangtze River Delta region is spatially imbalanced and has significant spatial dependence.

(3) The features of the spatial association network showed that the degree centrality, betweenness centrality, and closeness centrality of Huzhou and Changzhou are higher than other cities. In the green innovation efficiency spatial association network, Huzhou and Changzhou also had the most relationships. In addition, Huzhou had the most beneficiary relationships.

(4) The Yangtze River Delta region’s cities’ green innovation efficiency was separated into four blocks using block model analysis, with each block holding various responsibilities. Cities with a high level of innovation and economic development made up the first block. These cities were the driving force behind the four blocks, accounting for the majority of the green innovation efficiency spillover. The second block was a broker, assuming the role of a middleman in green innovation efficiency. The third block mainly benefited from other blocks. Cities in the fourth block reaped a net gain as a result of spillover linkages from previous blocks.

(5) QAP regression analysis showed that geographic distance; the expansion of the difference in energy consumption and the environment pollution index; and narrowing the gap in economic development, the industrial structure, and green coverage will boost the formation of spatial association.

Based on the above conclusions, we put forward some policy recommendations as follows:

(1) The green innovation cooperation among the cities in the Yangtze River Delta should be strengthened to push forward the integrated development of the region. Shanghai, Jiangsu, and other provinces with better development of green innovation should strengthen their cooperation with Zhejiang and Anhui, expand the scope of their influence as the driving force of green innovation, and gradually narrow the disparity in green innovation efficiency between cities. The free and orderly flow of green innovation input elements should be promoted in the cities in the Yangtze River Delta in order to allow full play to the comparative advantages of each city and enhance the green innovation ability and comprehensive competitiveness of the Yangtze River Delta urban agglomeration through collaborative development.

(2) H-H cities should further improve green innovative technologies, and gradually abandon or upgrade industries with high pollution and investment; L-H cities are more suitable for capital-intensive development. The innovation development of H-L cities is at the forefront, but it also faces environmental problems. Such development should realize the coordinated development of green and economy by improving conversion efficiency. L-L cities have low R&D efficiency, and the upgrading of industrial structure is difficult. Therefore, they should pay attention to the adjustment of industrial structure and accelerate the improvement of R&D efficiency.

(3) From the perspective of block analysis, it is necessary for cities to reach a balance in the receiving relationship and spillover relationship to promote the balanced development of green innovation in various regions. In addition, the green innovation efficiency of the transmission mechanism between blocks must be optimized, promoting regional linkage between blocks and cooperative promotion.

(4) In order to improve the spatial correlation of green innovation efficiency in the Yangtze River Delta, the allocation of green innovation resources in the Yangtze River Delta should be optimized. The green innovation mode should be changed by relying on the expansion of the urban scale and economic input, developing and attracting technology-related talent, and adjusting industrial structure. In addition, the government should gradually include the environmental quality of the Yangtze River Delta region in the evaluation system of local officials, and the introduction of industries with high pollution and high energy consumption should be reduced to improve the management efficiency of resource utilization. Moreover, efforts should be made to narrow the gap in economic development between cities and promote green innovation and sustainable development in the Yangtze River Delta.

In view of the focus of the research and the limitations of objective factors, there are still some deficiencies in this paper, which need to be further studied in the future. First of all, due to the availability of data, the urban-scale green innovation efficiency evaluation system constructed in this paper is still limited. Further research is needed to obtain more reasonable indicators in the future. Secondly, on the basis of the spatial association study, the spatial panel model can be further used to study the influencing factors of green innovation efficiency and explore the dynamic mechanism of green innovation efficiency evolution.