Study on the Spatio-Temporal Evolution and Influential Factors of Green Innovation Efficiency in Urban Agglomerations of China

Abstract

Promoting green innovation efficiency in urban agglomerations (UAs) can help to ensure the sustainability of China in a competitive but fragile post-industrialization era. This paper applies the Super Slacks-Based Measure model (Super-SBM) to measure the green innovation efficiency in 19 UAs of China from 2006 to 2018. Then, it examines the spatial-temporal evolution characteristics from the perspectives of geography and economics. Furthermore, the spatial econometric model is also established to explore the influential factors of green innovation efficiency, as well as its regional differences. The results reveal the following: (1) From the perspective of temporal differentiation, the green innovation efficiency of most UAs in China presents a fluctuated increase during the study period, and UAs located in the east are more ideal. (2) As for spatial differentiation, the number of UAs of a high value level is relatively stable, and the southeast coastal UAs performs as the core and a stepped pattern of “east > center > west” is clear. (3) A significant positive spatial spillover effect of green innovation efficiency does exist in UAs of China, and the effects of relative factors vary across regions. Differentiated measures should be taken to improve the green innovation efficiency in the UAs of China. This study provides significant guidance for realizing the goal of high-quality development in China, as well as fulfilling the international commitment of carbon peak and carbon neutrality.

1. Introduction

Since the beginning of the industrial revolution, innovations from all walks of life have greatly promoted the progress of human society. Especially under the continued trend of economic globalization, innovation capability has become a major cornerstone of economic competitiveness and comprehensive national power for a country. Alongside the progress of human society, the mounting problems of climate change and environmental pollution have been forcing people to rethink methods of development and to seek harmony with nature. However, the status quo is not ideal. The United Nations Environment Programme (UNEP, 2022) states that inadequate progress on climate action will mean that the rapid transformation of societies is the only option (https://www.unep.org/news-and-stories/press-release/inadequate-progress-climate-action-makes-rapid-transformation, accessed on 27 November 2022). Green innovation provides a breakthrough in realizing the goals of transformation and sustainable development.

Owing to the focus and innovation in the manufacturing industry, China surpassed Japan and became the world’s second largest economy in 2010. However, the economic-orientated and high-pollution traditions of manufacturing have made environmental problems such as water pollution, energy excessive consumption, the greenhouse effect, and haze increasingly prominent, which hinders the sustainable development of China. According to the data, China’s primary energy consumption in 2020 was 145.46 EJ, which is 1.657 and 8.556 times that of America and Japan, respectively (BP Statistical Review of World Energy, 2021. https://www.bp.com.cn/content/dam/bp/countrysites/zh_cn/china/home/reports/statistical-review-of-world-energy/2021/BP_Stats_2021.pdf, accessed on 27 November 2022). Due to an awareness of these severe problems, a series of environmental policies including launching an environmental tax and raising pollution emission standards have been implemented. However, under the competition of the world market, as well as the constrains of global climate change, how to achieve a “win-win” situation between economic growth and environmental protection in places where the conflicts are considerable is essential for China.

As the strategic core area and important spatial carrier of China’s economic development [1], UAs represent the convergence of labor forces, material and capital. After more than 40 years of development, a pattern of UAs referred to as “5 + 8 + 6” has emerged, which cover the eastern, central and western parts of China. It is estimated that about 80% of China’s population is concentrated in the 19 UAs, with a contribution to GDP of more than 90%. Yet, pressures on the environment also make them sensitive and potential disaster-prone areas of ecological environmental problems [2], which result in difficulties in providing adequate support for Chinese modernization. As the gathering place of innovation resources and activities, improving the green innovation efficiency of UAs plays an irreplaceably important role not only in achieving the goal of ecological civilization and high-quality development in China, but also in responding actively to global climate change and fulfilling the international commitment of carbon peak and carbon neutrality.

2. Literature Review

The existing papers related to the topic can be classified into three aspects, namely green innovation efficiency, evaluation methods, as well as its influencing factors.

2.1. Green Innovation Efficiency

Green innovation efficiency is a measure of sustainability that refers to how well innovation activity takes into account the ecological and economic benefits of a socioeconomic system. Green innovation is a critical path to promoting green development under the climate change background (Shuai et al. 2020; Yan et al. 2022) [3,4]. Studies are mainly based on the following two aspects. One focuses on the green innovation efficiency from a provincial or a regional perspective. Lv et al. (2020) calculated the green innovation efficiency of 30 provinces in China from 2006 to 2016 using the Slacks-Based Measure (SBM) and investigated its space-time transition and convergence trend [5]. Zhang et al. (2022) demonstrated a V-shaped cove in the temporal evolution of green innovation efficiency in the Yangtze River Economic Belt by using the SBM and Exploratory Spatial Data Analysis (ESDA) methods [6]. Xu et al. (2022) confirmed that there was a positive spatial spillover effect of green innovation efficiency in the Yellow River Basin with the method of the spatial econometric model [7]. Secondly, industries that have high environmental requirements or are closely related to environment pollution are specifically researched. Liu et al. (2018) employed the threshold regression model to conduct empirical tests of the nonlinear threshold effect of agglomeration on the green innovation efficiency of the tourism industry [8]. Wang et al. (2019) focused on the impact mechanism and action boundary of equity financing, debt financing and government subsidies on the green innovation efficiency of the manufacturing industry [9]. Fang et al. (2020) used the extended-DEA model to conduct an empirical test on China’s pollution-heavy industries [10]. However, research on the green innovation efficiency of UAs is notably lacking. UAs require green innovation efficiency, which can be achieved by utilizing environmental benefits, as well as economic benefits, to re-think the innovation activities and utilization chain and find implementable solutions.

2.2. Evaluation Methods of Green Innovation Efficiency

In terms of research methods, a parametric analysis based on stochastic frontier analysis (SFA) and a non-parametric method based on data envelopment analysis (DEA) are useful. For example, Liang et al. (2016) used DEA to evaluate green innovation efficiency and to understand the factors of China’s industrial enterprises [11]. Xiao et al. (2019) measured the green innovation efficiency and ecological welfare performance of 30 provinces in China from 2004 to 2015 based on the improved SFA mode [12]. However, due to limitations in dealing with multi-output efficiency problems of SFA, DEA and its extension forms are more widely accepted [13]. DEA is an effective tool that is used to measure the efficiency of a decision-making unit (DMU), which was first proposed by Charnes et al. (1978) and has many classical forms, such as CCR, BBC, etc. However, its single radial assumption often has a gap with the reality, resulting in inefficiency factors. Although the SBM proposed later avoids the bias and influence caused by the difference between dimension and angle selection, the lack of efficiency differentiation on the front surface is still not conducive to comparison and ranking among DMUs. Therefore, the Super-SBM model was conceived by Tone (2002), which provides a solution for the comparison of DMU on the frontier [14]. Furthermore, Chen et al. (2021) proved that ignoring undesired outputs may lead to a virtual height estimation of green innovation efficiency [15]. Therefore, we finally adopted the Super-SBM model with a comparable frontier DMU and undesired outputs to evaluate the green innovation efficiency in UAs of China.

2.3. Influencing Factors of Green Innovation Efficiency

When exploring the influencing factors of green innovation efficiency, scholars usually discuss two aspects of the external environment and internal drives. As for the external environment, policies and the marketization level are mainly of concern. By using the Tobit mode, Yang et al. (2018) discussed the influence of technology market maturity and market openness on green innovation efficiency [16]. Liu et al. (2021) confirmed that the implementation of carbon trading policies had a significant and continuous effect in promoting and improving green innovation efficiency in the pilot areas. Zhao et al. (2022) evaluated the impact of green finance and environmental regulations on the green innovation efficiency in China from 2011 to 2020 [17]. Zhang et al. (2020) [18], Xu et al. (2020) [19] and Chen et al. (2022) [20] quantitatively measured and analyzed the construction effects of a smart city, low-carbon city and national innovative city on green innovation efficiency in China, respectively. The internal drivers are closely related to enterprises that are the main body of technological innovation. Influencing factors represented by the inputs of research and development (R&D) [21], the inflow of foreign direct investment (FDI) [22] and the management level [23] are discussed. Current studies on the influencing factors are abundant but the spillover effects and mechanisms from spatial perspectives are often neglected. Therefore, a spatial econometric model is applied in this paper to overcome the limitations.

In summary, this paper attempted to bridge research gaps via the following three aspects. Firstly, based on the panel data of 213 cities from 2006 to 2018, the study of green innovation efficiency focuses on the perspective of UA, where the spatiotemporal differentiation characteristics were explored, namely time evolution and spatial evolution, and its evolution characteristics are comprehensively discussed. Secondly, to calculate the green innovation efficiency, the undesired outputs were taken into account and an improved SBM model was used to overcome the issue of DMU being incomparable on the frontier to obtain more accurate results. Thirdly, the spatial spillover effects of factors influencing green innovation efficiency in UAs of China were considered by using a spatial econometric model and possible reasons are discussed accordingly. This work provides theoretical support for the realization of high-quality development in China.

3. Research Design

3.1. Theoretical Framework

Green innovation involves a circularly promoted community that is both internally driven and externally responsive [24]. As we can see from the theoretical framework of Figure 1, it is composed of two circles, namely, the background elements and the core green innovation. The four background elements interact with each other, which provides corresponding basic conditions for green innovation. Specifically, natural elements provide necessary natural resources, economic elements can provide corresponding assistance, social elements affect the accessibility and utilization of innovation elements, and the role of policy elements can be seen in the implementation of regulations and regional planning. Meanwhile, green innovation will also generate green welfare through technological innovation; that is, it will decrease resource utilization and increase the waste utilization rate. Furthermore, the benefits of green innovation will promote the high-level development of policy elements, and then stimulate enterprises to carry out innovative activities. Moreover, just as in traditional innovation activities, green innovation inevitably contributes to green economic growth. Lastly, a society which enjoys development under green innovation guidance is undoubtedly optimized and of a high quality.

3.2. Research Methods

3.2.1. Super-SBM

Referring to Tone (2002), the Super-SBM model was employed to measure the green innovation efficiency of UAs in China in this study, which can help to overcome the shortcomings caused by the difference in these dimensions and angle selection and makes the frontier comparable. The method is shown in Equation (1).

$$\begin{array}{l}\min \varphi =\frac{1+\frac{1}{m}{\displaystyle \sum _{i=1}^m{s}_i^{-}/x_{ik}}}{1 - \frac{1}{q_1+q_2}\left({\displaystyle \sum _{j=1}^{q_1}{s}_j^{+}/y_{j0}+{\displaystyle \sum _{l=1}^{q_2}{s}_l^{b-}/b_{lk}}}\right)}\\ s.t.\{\begin{array}{l}{\displaystyle \sum _{r=1,r\ne k}^n{x}_{ir}}{\lambda }_r-s_i^{-}\le x_{ik}\\ {\displaystyle \sum _{r=1,r\ne k}^n{y}_{jr}}{\lambda }_r+s_j^{+}\ge y_{jk}\\ {\displaystyle \sum _{r=1,r\ne k}^n{b}_{lr}}{\lambda }_r-s_l^{b-}\le b_{ik}\\ \lambda ,s^{-},s^{+}\ge 0\\ i=1,2\dots ,m;j=1,2\dots ,q;\\ r=1,2\dots ,n(r\ne k)\end{array}\end{array}$$

(1)

where $\varphi$ is the green innovation efficiency value and $\lambda$ is the weight vector. $x, y, b$ denote the input, desired output, and undesired output, respectively. $s^{+}, s^{-}, s^{b-}$ represent the slack variables of expected output, input, and undesired output of innovation utilization factors, respectively. $m$, $q_1$, $q_2$ are the quantity of input factors, expected outputs and non-expected outputs. For green innovation efficiency, the DMU has not reached the valid state when $0\le \varphi <1$, it is on or above the production frontier when $\varphi \ge 1$, and the larger the value of $\varphi$, the higher the green innovation efficiency.

3.2.2. Theil Index

On the basis of entropy theory, the Theil index was initially introduced by Theil (1967) to study the problem of income inequality. Subsequently, owning to the good decomposition and applicability, it has now been introduced to measure the regional differences and sources of green innovation efficiency [25]. The Theil index takes a non-negative value of no more than 1, and the larger the value is, the greater the regional differences. Moreover, it can be further decomposed into inter-regional and intra-regional differences. Specific equations and decompositions are shown in Equation (2).

$$\begin{array}{l}T=T_{BR}+T_{WR}\\ T_{BR}={\displaystyle \sum _{i=1}^m\frac{n_i}{n}}\left(\frac{\overline{E_i}}{\overline{E}}\right)\ln \left(\frac{\overline{E_i}}{\overline{E}}\right)\\ T_{WR}={\displaystyle \sum _{i=1}^m\left(\frac{n_i}{n}\frac{\overline{E_i}}{\overline{E}}\right)}{T}_i\\ T_i=\frac{1}{n_i}{\displaystyle \sum _{j=1}^{n_i}\frac{E_{ij}}{\overline{E_i}}}\ln \frac{E_{ij}}{\overline{E_i}}\end{array}$$

(2)

where $T \mathrm{and} T_i$ denote the Theil index of green innovation efficiency in UAs of China and UA $i$, respectively. $T_{WR}$ and $T_{BR}$ denote the difference of green innovation efficiency within and between UAs. $m$, $n_i/n$, ${\overline{E}}_i/\overline{E}$, and $E_{ij}$ represent the number of UAs, the proportion of prefecture-level cities of UA $i$, ratio of the mean green innovation efficiency of UA $i$ to the national mean, and the green innovation efficiency of city $j$ in UA $i$.

3.2.3. Moran’s I Test

Characteristics such as interdependence, mutual restriction, and mutual influence are prominent in the geospatial space. As an important method of testing as to whether green innovation efficiency is spatially dependent or not, the equation of Moran’s I test is essential and can be expressed in Equation (3).

$$I=\frac{n{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}\left(E_j-\overline{E}\right)\left(y_j-\overline{E}\right)}}}{{\displaystyle \sum _{i=1}^n{\displaystyle \sum _{j=1}^n{W}_{ij}{\left(E_j-\overline{E}\right)}^2}}}$$

(3)

where $n, E_j$ and $\overline{E}$ donate the number of studies, the green innovation efficiency of city $j$ and the mean value of all areas studied. $W_{ij}$ represents the economic geographical spatial weight matrix. When $I\in [$−1, 0), it indicates that the spatial correlation is negative, while $I\in \left(0, 1\right]$ denotes that a positive correlation exists. The greater the absolute value of I, the stronger the spatial correlation is.

3.2.4. Spatial Econometric Model

There are three types of spatial econometric models, which include the spatial lag model (SLM), spatial error model (SEM) and spatial Dubin model (SDM). Among these types, SDM is considered as a more general spatial metrology model due to its comprehensiveness. The formula is represented as follows:

$$Y=\gamma WY+X\beta +WX\phi +\epsilon$$

(4)

where $Y$, $X$, and $W$ denote the interpreted variable, efficiency influential factors matrix, and spatial weight matrix, respectively. Considering the fact that regions are connected with economic levels and geographical distance, we selected the economic geographical distance to construct the spatial weight matrix. The economic geography nested matrices $WY$ and $WX$ represent the explained and explanatory variables that are incorporated with spatially correlated factors, respectively. $\gamma$ and $\epsilon$ are the spatial autocorrelation coefficient and independently but identically distributed random error terms. $\beta \mathrm{and} \phi$ are parameters to be estimated.

While $\phi =0$, the SDM degenerates into the SLM, which means that the spatial dependence is reflected in the form of the lagged term of the explained variables, and Equation (4) can be expressed as shown in Equation (5).

$$Y=\gamma WY+X\beta +\epsilon$$

(5)

While $\phi +\gamma \beta =0$, the SDM degenerates into the SEM, which means that the spatial dependence is embodied in the form of the error term, and Equation (4) can be expressed as Equation (6).

$$Y=X\beta +\mu ,\mu =\rho W\mu +\epsilon$$

(6)

3.3. Variable Design

3.3.1. Input–Output Indicator System

The green innovation efficiency reflects the effective utilization of elements linked to both production and technological innovation, which are placed under resource and environmental constraints. Under certain conditions, the fewer the undesired outputs, the higher the green innovation efficiency. On the basis of clarifying the concepts above, as well as the representation and availability of data, the input–output indicator system of green innovation efficiency in the UAs of China was constructed and shown in

Table 1. Input–output indicator system of green innovation efficiency in the UAs of China.

Index	Variables	Indicator Explanation
Input	Capital	Capital stock of fixed assets
		Science and technology expenditure/fiscal expenditure
	Labor	Number of employees in science and technology service industry
	Energy	Comprehensive industrial energy consumption
Output	Society	Green areas in municipal districts
	Environment	Industrial SO2 emissions per GDP
		Industrial soot and sulfur emissions per GDP
		Industrial waste water discharge per GDP
	Innovation	Number of papers published
		Number of patent applications accepted

.

Input: Capital and labor are always the basic elements of the production function [26]. According to the work of Yan et al. [27], the innovation input factors are represented by the proportion of science and technology expenditure in terms of overall fiscal expenditure, indices of the capital stock of fixed assets, and the number of science and technology service labor forces. Considering the objective reality that modern economic development is closely connected with energy consumption, we selected the comprehensive energy consumption function to represent this, so as to eliminate the inconsistencies in the types and dimensions selected.

Output: The desired economic growth, society’s welfare, and undesired environmental pollution were considered. Desired outputs are represented by green areas in municipal districts and innovation achievements. In contrast, industrial SO₂ emissions per GDP, industrial soot and sulfur emissions per GDP, and industrial waste water discharge per GDP were used to examine the impact on the environment.

3.3.2. Influential Factors

Given their complexity, the green innovation efficiency of UAs is affected by various internal and external factors. Referring to the existing research results of Lv et al. (2020), and taking the characteristics of China’s UAs into consideration, as well as the availability and accuracy of data, a model including the following eight influential factors was constructed.

Economic scale ($GDP$) and structure ($STR)$. Economic development has the dual effects of innovative financial support and green human resources agglomeration [28], which is an important indicator when measuring comprehensive strength, and is represented by the historical GDP of relevant cities. Compared with the energy-intensive secondary industry, such as manufacturing and mining, the proportion of the tertiary industry can better reflect the level of industrial upgrading of a region, and the ratio of the output value of the tertiary industry to the secondary industry is selected.

Urbanization (URB) and informatization level ($INF$). The improvement of urbanization positively stimulates the concentration of talent and capital, and hence promotes the generation and diffusion of innovation achievements. Therefore, the ratio of urban population to total population at the end of the year is used. Furthermore, the Internet is the infrastructure on which green innovation activities depend, as convenient and efficient transmission approaches can both reduce communication costs and improve information accessibility. Therefore, it is represented by the number of Internet users per 10,000 people.

Government guidance ($GOV$) and financial support ($SUP$). Government support is the premise and foundation for regional green innovation activities, and plays an important role in resolving issues such as the lack of innovation funds and promoting the breakthrough of key technologies. Therefore, the ratio of science and technology expenditure to the general budget of local finance is reviewed. Furthermore, the financial industry can enhance the vitality of urban innovation by diversifying risks, easing financing constraints, and facilitating the effective docking of technology and capital. Thus, we used the proportion of outstanding loans in GDP that year to explore this factor.

Openness ($OPE$) and competitive environment ($IND)$. An open circulation environment helps to establish direct links between international factors and domestic industries and technologies. However, the importation of foreign capital indiscriminately may also lead to a “paradise of pollution”. We used the proportion of actually utilized foreign capital in GDP to measure this factor. Moreover, since enterprises are important actors in innovation activities, and compared with other enterprises, enterprises above a designated size have more weighting and are able to pay more attention to technology and product innovation, while hoping to maintain and expand their own advantages. Therefore, the number of industrial enterprises above a certain scale was used to represent it.

3.4. Research Area and Data Source

3.4.1. Research Area

According to Fang’s (2021) research results [29], combined with the approval documents of the State Council, the National Development and Reform Commission, and the provinces and autonomous regions, we took the “5 + 8 + 6” UA model as the research object, and divided the research area into three parts, which are the East, Middle and West (see Figure 2 below). Considering the objective fact that Haidong City and Bijie City adjusted their administrative divisions during the study period, 213 prefecture-level cities were finally selected (

Table 2. The classification of UAs.

Regions	Urban Agglomerations (Abbr.)	Cities
Eastern	Beijing–Tianjin–Hebei (BTH)	Beijing, Tianjin, Shijiazhuang, Tangshan, Baoding, Handan, Zhangjiakou, Chengde, Langfang, Qinhuangdao, Cangzhou, Xingtai, Hengshui
	Shandong Peninsula (SP)	Jinan, Qingdao, Zibo, Zaozhuang, Dongying, Yantai, Weifang, Jining, Tai’an, Weihai, Rizhao, Linyi, Dezhou, Liaocheng, Binzhou, Heze, Laiwu
	Yangtze River Delta (YRD)	Shanghai, Nanjing, Suzhou, Wuxi, Changzhou, Nantong, Yangzhou, Taizhou, Zhenjiang, Yancheng, Hangzhou, Jiaxing, Shaoxing, Huzhou, Jinhua, Zhoushan, Ningbo, Wenzhou, Taizhou, Hefei, Ma’anshan, Tongling, Anqing, Xuancheng, Chuzhou, Chizhou, Wuhu
	Western Strait (WS)	Fuzhou, Xiamen, Quanzhou, Putian, Zhangzhou, Sanming, Nanping, Ningde, Longyan, Lishui, Quzhou, Shangrao, Yingtan, Fuzhou, Ganzhou, Shantou, Chaozhou, Jieyang, Meizhou
	Pearl River Delta (PRD)	Guangzhou, Shenzhen, Foshan, Dongguan, Zhongshan, Huizhou, Zhuhai, Jiangmen, Zhaoqing
	South-central Liaoning (SCL)	Shenyang, Dalian, Anshan, Fushun, Benxi, Yingkou, Liaoyang, Tieling, Panjin
Central	Harbin–Changchun (HC)	Harbin, Daqing, Qiqihar, Suihua, Mudanjiang, Changchun, Jilin, Siping, Liaoyuan, Songyuan
	Central Henan (CH)	Zhengzhou, Kaifeng, Luoyang, Nanyang, Shangqiu, Anyang, Xinxiang, Pingdingshan, Xuchang, Jiaozuo, Zhoukou, Xinyang, Zhumadian, Hebi, Puyang, Luohe, Sanmenxia, Changzhi, Jincheng, Yuncheng, Huaibei, Bengbu, Suzhou, Fuyang, Bozhou
	Central Shaanxi (CSX)	Xi’an, Baoji, Xianyang, Tongchuan, Weinan, Tianshui, Pingliang, Qingyang
	Middle Reaches of Yangtze River (MYR)	Wuhan, Huangshi, Ezhou, Huanggang, Xiaogan, Xianning, Xiangyang, Yichang, Jingzhou, Jingmen, Changsha, Zhuzhou, Xiangtan, Yueyang, Yiyang, Changde, Hengyang, Loudi, Nanchang, Jiujiang, Jingdezhen, Xinyu, Yichun, Pingxiang, Ji’an
	Ningxia Yanhuang (NY)	Yinchuan, Shizuishan, Wuzhong, Zhongwei
	Huhhot–Baotou–Erdos–Yulin (HBEY)	Hohhot, Baotou, Ordos, Yulin
	Central Shanxi (CS)	Taiyuan, Jinzhong, Xinzhou, Yangquan, Luliang
Western	Chengdu–Chongqing (CC)	Chongqing, Chengdu, Zigong, Luzhou, Deyang, Mianyang, Suining, Neijiang, Leshan, Nanchong, Meishan, Yibin, Guang’an, Dazhou, Ya’an, Ziyang
	Central Yunnan (CY)	Kunming, Qujing, Yuxi
	Central Guizhou (CG)	Guiyang, Zunyi and Anshun
	Lanzhou–Xining (LX)	Lanzhou, Xining, Baiyin, Dingxi
	Beibu Gulf (BG)	Nanning, Beihai, Qinzhou, Fangchenggang, Yulin, Chongzuo, Zhanjiang, Maoming, Yangjiang and Haikou
	Northern Slope of Tianshan Mountain (NTM)	Urumqi, Karamay

). In addition, the Fifth Plenary Session of the 16th Central Committee of the Communist Party of China officially raised the construction of a resource-saving and environment-friendly society as a strategic task in 2005, so the year 2006 was chosen as the starting point of the study, and the study period was 2006–2018.

3.4.2. Data Source

The capital stock of fixed assets was measured according to the sustainable inventory method of Zhang et al. [30]. The comprehensive industrial energy consumption was converted by referring to the General Principles for the Calculation of Comprehensive Energy Consumption (GB/T 2589-2020), and the coefficients for gas, liquefied petroleum gas and electricity consumption were 1.330, 1.714 and 0.123, respectively. The numbers of the invention patent, utility model patent and design patent were obtained from the CNRDS database and then converted by using the method of Bai et al. [31], with assigned weights of 0.5, 0.3 and 0.2, respectively, to overcome the inconsistency of the innovation degree. Referring to the method of Lv et al. (2018), China National Knowledge Infrastructure (CNKI) and Web of Science (WoS) were used to retrieve the number of papers both in Chinese and English from 2006 to 2018 in 213 target cities, and the search conditions were “the first unit” and “$\mathrm{AD}=\mathrm{address}$”, respectively.

The data used in our paper were collected from the China Statistical Yearbook and China Urban Statistical Yearbook, which are published by the Statistic Bureaus of the Chinese central government and local government. All the price variables were adjusted with the year 2006 as the base period, and some missing data were supplemented by a linear interpolation method. Moreover, the pair numbers of $GDP$, INT and $IND$ were taken to reduce heteroscedasticity problems.

4. Empirical Results

4.1. Spatio-Temporal Evolution Analysis of Green Innovation Efficiency

4.1.1. Temporal Evolution Characteristics

Based on the input–output indicator system of green innovation efficiency, the Max DEA 8.0 software was used to calculate the Super-SBM model containing unexpected outputs, and the green innovation efficiency of 213 cities in 19 UAs was obtained from 2006 to 2018.

Figure 3 shows the mean value of green innovation efficiency of UAs in different regions during 2006–2018. On the whole, the mean value during the study period is 0.42, which means there is still a 58% improvement space from the production frontier. Specifically, after a slow increase from 2006 to 2011, the green innovation efficiency dropped to 0.41 in 2012, and then increased steadily and peaked at 0.4829 in 2015, when a zigzag fluctuation of a “V” type appeared. Since the two inflection points are in the transitional period of the 11th Five-Year Plan to the 12th Five-Year Plan, and the 12th Five-Year Plan to the 13th Five-Year Plan, respectively, we deduced that the industrial structure adjustment and policy transformation, including energy conservation and emission reductions, may be the main reasons for this [32]. Furthermore, the green innovation efficiency of UAs generally presents a hierarchical pattern of east > central > west, with mean values of 0.595, 0.359 and 0.318, respectively. Among them, the east has always been in the lead, while the central and western UAs are below the national average.

Table 3. Green innovation efficiency values of UAs in China between 2006 and 2018.

Regions	UAs	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018	Mean
Eastern	BTH	0.41	0.45	0.46	0.51	0.49	0.59	0.54	0.50	0.52	0.62	0.50	0.52	0.54	0.55
	SP	0.61	0.57	0.56	0.55	0.54	0.62	0.57	0.59	0.61	0.66	0.55	0.57	0.64	0.62
	YRD	0.61	0.59	0.62	0.64	0.66	0.68	0.67	0.66	0.72	0.72	0.66	0.67	0.72	0.61
	WS	0.50	0.48	0.46	0.50	0.52	0.64	0.57	0.70	0.65	0.64	0.58	0.48	0.53	0.71
	PRD	0.85	0.78	0.77	0.79	0.83	0.86	0.86	0.90	0.99	0.93	0.86	0.94	0.99	0.62
	SCL	0.43	0.44	0.43	0.42	0.43	0.34	0.30	0.31	0.30	0.30	0.31	0.44	0.48	0.46
Central	HC	0.47	0.48	0.52	0.50	0.51	0.55	0.54	0.57	0.56	0.62	0.60	0.56	0.58	0.44
	CH	0.29	0.29	0.29	0.30	0.30	0.37	0.34	0.42	0.40	0.45	0.39	0.32	0.38	0.38
	CSX	0.34	0.31	0.36	0.33	0.37	0.44	0.40	0.57	0.56	0.48	0.51	0.33	0.34	0.40
	MYR	0.37	0.40	0.40	0.40	0.38	0.44	0.34	0.41	0.41	0.38	0.38	0.30	0.33	0.29
	NY	0.20	0.19	0.17	0.24	0.25	0.34	0.11	0.24	0.16	0.16	0.16	0.17	0.22	0.28
	HBEY	0.21	0.23	0.20	0.20	0.14	0.39	0.58	0.49	0.61	0.52	0.47	0.25	0.27	0.32
	CS	0.39	0.25	0.25	0.25	0.27	0.27	0.24	0.32	0.30	0.31	0.30	0.19	0.24	0.33
Western	CC	0.27	0.30	0.32	0.34	0.36	0.34	0.33	0.42	0.49	0.50	0.39	0.42	0.43	0.36
	CY	0.30	0.34	0.33	0.26	0.28	0.33	0.26	0.38	0.51	0.40	0.36	0.31	0.37	0.34
	CG	0.31	0.31	0.37	0.27	0.28	0.28	0.29	0.38	0.44	0.59	0.33	0.29	0.38	0.33
	LX	0.31	0.32	0.33	0.34	0.38	0.39	0.37	0.27	0.24	0.25	0.27	0.23	0.27	0.30
	BG	0.26	0.25	0.25	0.30	0.31	0.31	0.25	0.31	0.34	0.35	0.38	0.28	0.30	0.27
	NTM	0.25	0.24	0.24	0.25	0.28	0.25	0.20	0.24	0.24	0.27	0.22	0.19	0.23	0.41

further reveals the green innovation efficiency of 19 UAs in detail during 2006–2018 and an upward trend was observed for most UAs. Specifically, the Pearl River Delta and Yangtze River Delta UAs has experienced rapid growth, from 0.85 to 0.99, and 0.61 to 0.72, respectively, while the UAs of the Lanzhou–Xining, Central Shanxi and Northern Slope of Tianshan Mountain regions show a downward trend. Among them, the Central Shanxi UA decreased significantly, from 0.39 to 0.24. The possible reason may lie in the fact that as it is an area whose industry involves mainly highly polluted coal mining and processing, green processes present a greater challenge, leading to the large downward trend of green innovation efficiency here. From the perspective of different UAs, during the study period, the Western Strait Pearl River Delta, Shandong Peninsula and Yangtze River Delta UAs, which belong to eastern region, have an ideal green innovation efficiency with the average values of 0.71, 0.62, 0.62 and 0.61, respectively. The Beibu Gulf, Ningxia Yanhuang, Middle Reaches of Yangtze River and Lanzhou–Xining UAs, which are located in western and central regions, have a low green innovation efficiency, with mean values of 0.27, 0.28, 0.29 and 0.3, respectively.

4.1.2. Spatial Evolution Characteristics

According to the green innovation efficiency of UAs in China, as calculated above, the natural breakpoint method of ArcGIS10.2 software was used to provide spatial visualization mapping of the green innovation efficiency and divide them into four levels (Figure 4).

As shown in Figure 4, it is obvious that UAs have a significant spatial difference in terms of green innovation efficiency. The level presents a spatial distribution pattern with the southeast coastal UAs at the core, and this gradually decreases from east to west. From the perspective of spatial pattern evolution, the UAs at the high value level of green innovation efficiency are relatively stable, while the others present a certain degree of fluctuation. Specifically, the Pearl River Delta and Yangtze River Delta UAs were always in the “High” value region, while Chengdu–Chongqing, Zhongyuan and Beibu Gulf UAs showed an increasing trend, and Chengdu–Chongqing UA presented a leapfrog development from the low-efficiency area in 2006 to the medium-high efficiency area in 2018. Huhhot–Baotou–Erdos–Yulin and Ningxia–Yanhuang UAs increased to medium-low efficiency areas in 2011 and then decreased to low efficiency areas again in 2018. In contrast, Lanzhou–Xining and Jinzhong UAs transformed from medium–low-efficiency areas to low-efficiency areas during the study period.

4.1.3. Analysis of Regional Differences

To identify the regional differences of green innovation efficiency in UAs, the Theil index in Equation (2) was measured, as shown in

Table 4. The Theil index of UAs in China from 2006 to 2018.

Year	2006	2007	2008	2009	2010	2011	2012	2013	2014	2015	2016	2017	2018
Theil index	0.4	0.358	0.354	0.349	0.363	0.34	0.4	0.307	0.35	0.304	0.311	0.405	0.361

and Figure 5. During the study period, the Theil index decreased from 0.400 to 0.361, with a decline of 9.74%, indicating that there were regional differences in the green innovation efficiency of UAs in China; however, the gap tended to narrow. As for its components, the difference between UAs is significant, whose contribution are over 55%, while the difference within tends to narrow.

Figure 6 shows that the average value of the Theil index within UAs differs greatly during the study period. Specifically, the Theil index within the UAs of Lanzhou–Xining, Guanzhong, Harbin–Changchun and Huhhot–Baotou–Erdos–Yulin are greater than 0.2, with Lanzhou–Xining UA ranking the first, at 0.359. Meanwhile, the Theil index within UAs of Pearl River Delta, Yangtze River Delta, Shandong Peninsula, central Guizhou, and Tianshan are relatively low. It is interesting that these five UAs are polarized in terms of green innovation efficiency. The UAs of Pearl River Delta, for which the green innovation efficiency is the highest, have the smallest internal difference, and the Theil index is 0.022, followed by the Tianshan UA, for which the green innovation efficiency is quite low, and the Theil index is 0.025.

As shown in Figure 7, the within-group difference of the Theil index in most UAs showed a decreasing trend. Specifically, from 2006 to 2011, the difference between UAs, especially those located in central and western regions, increased. From 2011 to 2016, the difference of most UAs narrowed. From 2016 to 2018, the difference in the central and western regions showed a decreasing trend, while that in the eastern region increased.

4.2. Driving Factor Analysis

4.2.1. Spatial Correlation Test

Based on Equation (3),

Table 5. Moran’s I and standardized Z values.

Year	Moran’s I	Z Value	Year	Moran’s I	Z Value
2006	0.202 ***	5.366	2013	0.124 ***	3.348
2007	0.220 ***	5.841	2014	0.109 **	2.944
2008	0.199 ***	5.291	2015	0.152 ***	4.077
2009	0.222 ***	5.894	2016	0.178 ***	4.765
2010	0.230 ***	6.109	2017	0.265 ***	7.012
2011	0.115 ***	3.096	2018	0.226 ***	6.025
2012	0.182 ***	4.861

Note. *** and ** denote rejection of the null hypothesis at the 1% and 5% levels, respectively.

identifies the Moran’s I test results. It can be seen that the Moran’s I index is positive and significant at more than a 5% level, accompanied by a fluctuating upward trend. This indicates that the overall spatial agglomeration and dependence of UAs’ green innovation efficiency in China are obvious.

4.2.2. Empirical Results

The variance inflation factor (VIF) test results showed that the mean value for each variable was 3.18 (<10), which means the collinearity of the index is not substantial enough to affect the research results. The Wald test, LR test and Hausman test indicate that the hypothesis of SDM degrades to SEM, or SLM can be rejected, and the SDM with a time-fixed effect is the best choice. The regression results are shown in

Table 6. Spatial panel regression results.

Variable	Nested Matrix of Economic Geography				Matrix of Geographic Distance
	Whole	East	Central	West
$GOV$	0.029 ***	0.034 ***	0.004	−0.009	0.027 ***
$STR$	0.001 ***	0.001 ***	0.001 ***	0.001 ***	0.001 ***
$LN\left(GDP\right)$	0.062 ***	0.027	0.089 ***	0.041	0.066 ***
$OPE$	0.008 ***	−0.001	0.026 ***	0.013 **	0.007 ***
$LN\left(INF\right)$	0.050 ***	0.029	0.055 ***	0.025	0.053 ***
$SUP$	−0.000	0.000	0.000	−0.000	−0.000
$LN\left(IND\right)$	0.065 ***	0.057 ***	0.038 **	0.117 ***	0.044 ***
$URB$	0.003 ***	0.008 ***	0.000	0.004 ***	0.001 ***
W * $GOV$	−0.012	−0.016	0.051 **	−0.056 *	−0.049
W * $STR$	0.002 ***	0.000	0.006 ***	0.000	0.002
W * $LN\left(GDP\right)$	−0.073 **	−0.154 ***	−0.052	−0.093	−0.180 *
W * $OPE$	0.001	0.011	0.018 **	−0.006	−0.080 ***
W * $LN\left(INF\right)$	−0.048 *	0.083	0.047	0.105 ***	−0.257
W * $SUP$	0.000	−0.002 ***	0.000	−0.000	−0.001
W * $LN\left(IND\right)$	−0.002	0.068 **	−0.164 ***	−0.060	0.271 ***
W * $URB$	0.001	−0.005 ***	−0.004 **	−0.003	0.014 ***
$\rho$	0.135 ***	0.204 ***	0.263 ***	−0.113 *	0.383 ***
$sigma2\_e$	0.045 ***	0.043 ***	0.036 ***	0.025 ***	0.046 ***
R_squared	0.455	0.519	0.153	0.506	0.324

Note. ***, ** and * denote rejection of the null hypothesis at the 1%, 5% and 10% levels, respectively.

.

It can be seen from Table 6 that the spatial regression coefficient $\rho$ of green innovation efficiency is 0.135 and significant at the 1% level, meaning a positive spatial spillover of green innovation efficiency does exist. However, the spillover effect is not consistent in the eastern, central and western UAs. The possible reasons are that the scales of western UAs are relatively small, accompanied by the low level of economic development, resulting in strong siphon effects in capital and project investment, as well as talent introduction.

It is interesting that the regression results of $SUP$ are not significant in all areas, which means that this factor has no significant influence on the efficiency of green innovation in UAs of China. The possible reason is that uncertainties in innovation activities are huge, such as high risks, large investments and long cycles; these issues mean that decisions made in terms of investments in these are more cautious. Furthermore, financial behaviors are obviously guided by the government. For certain innovation subjects, the leading role of the government makes it possible for innovation subjects to “bypass” the financial market to obtain financing, thus causing the financial market to “short-circuit” [33]. On the contrary, variables of $STR$ and $LN\left(IND\right)$ are positively significant for all areas. A probable explanation is that, compared with the manufacturing industry, the green attribute of the service industry is stronger. For example, a knowledge-intensive service industry directly promotes the improvement of green innovation efficiency. Furthermore, driven by the motivation of chasing profits from internal markets and breaking the competitive squeeze caused by external competition, industrial enterprises above a designated size attach great importance to technological innovation. The effect in the west is better than that on the east and center, which is largely related to the situation of high redundancy and low output traditions in the old industrial bases gathered in the central and eastern UAs.

Considering the rest of the factors, their performances are different. Specifically, $GOV$ and $URB$ are positively significant for the eastern UAs. This may be because governments and enterprises have different preferences for technology. The government tends to support cutting-edge and long-term technologies, which means limited benefits in the short-term, and this may be unfavorable to the current R&D efficiency [34]. For the eastern UAs, advanced research facilities and abundant research talents enable the government and enterprises to take corresponding risks. Meanwhile, owing to the high urbanization level here, huge market size and the low transaction cost of innovative resources, the spawning of high-level innovation platforms and a high-quality knowledge spillover space are observed. However, a strong economic foundation built on the basis of supporting facilities and industrial chains has numerous advantages in industrial agglomerations and pollution control technologies. Accompanying the early start in Internet construction, the strict quality requirements and controls on foreign investment make variables of $GDP$,$LN\left(INF\right)$ and $OPE$ insignificant.

For central UAs, $LN\left(GDP\right)$, $OPE$ and $LN\left(INF\right)$ are positively significant. The reason is that the expansion of the economic scale under the new development concept means changing the traditional mode, which includes high energy consumption and pollution, raising the threshold of industrial efforts, optimizing the macro market environment, and forming a positive connection between economic strength and green innovation capabilities. Furthermore, the information technology represented by the Internet is conducive to the production and diffusion of innovative elements in the central UAs and constantly leads to innovation. The advanced technology and management experience brought by foreign investment are meaningful in improving the level of green technology in central UAs. Meanwhile, as an important carrier of industrialism from the east, central UAs are faced with the dual pressure of economic development and resource regulation, which is accompanied by the economic and technological benefits derived from urbanization; therefore, how to give full play to the positive role of urbanization in green innovation efficiency should be further investigated.

For western UAs,$URB$ and $OPE$ are positively significant. A probable explanation is that the improvement of urbanization in the western region will both increase the supply of urban labor forces and infrastructures, which are central to the attraction of talent and industries. Furthermore, with the support from policies of the Great Western Development Strategy and poverty alleviation, the effects of the growth pole of the western UAs, in terms of population and industry concentration, as well as the attraction of production factors, are strengthened. Particularly, thanks to the Belt and Road Initiative and the Chongqing–Xinjiang–Europe Corridor, the western regions have become a bridgehead in opening-up inland China and are a latecomer in technological innovation. However, due to the poor risk tolerance of enterprises and the information asymmetry between government and enterprises, the actual effect of government funds are often affected, which is also the case for central UAs.

4.2.3. Robustness Test

Furthermore, in this study, the matrix of geographical distance was adopted to test the accuracy of the above results, as shown in Table 6. Economic scale, industrial structure, the level of informatization and urbanization, government orientation, opening up to the outside world, and competition have significant positive effects on the green innovation efficiency of UAs, while the influence of financial support is not significant, which is consistent with the test results of the economic geography nested matrix, indicating that the constructed spatial econometric model has good robustness.

5. Discussion

5.1. Revisiting Green Innovation Efficiency in the UAs of China

Improving the green innovation efficiency is of great significance to China’s high-quality economic development. A number of studies discuss the spatio-temporal evolution and influencing factors at the national, provincial, industry and basin levels [35,36,37,38]. Some studies consider the existence of undesirable outputs and introduce them into an envelope analysis framework suitable for measuring green innovation efficiency. As an important and complex economic system, the framework of green innovation efficiency and influential factors in cities of a certain and representative region remain challenging. Thus, this paper broadens applicability and builds an input–output indicator system for green innovation efficiency measurement. Furthermore, 213 cities in 19 UAs were selected as research areas to measure green innovation efficiency during 2006–2018, to analyze the differences in spatial pattern and evolution and explore the influential factors using SDM. This work improves the applicability of green innovation efficiency and the SDM framework, and enriches the existing research scale. The results can provide theoretical support for decision makers.

5.2. Limitations and Potential Solutions

In this study, we measured the green innovation efficiency in the UAs of China based on the Super-SBM model, analyzed the spatial evolution law of green innovation efficiency by combining the dimensions of economics and geography, and used SDM to test the influential factors. However, due to the lack of official data in some cities, the undesired outputs covered in this paper are only the most important ones, resulting in small environmental damage as a whole. We will improve the integrated environmental damage accounting model, focus on the emission of CO₂ with the use of night light data and pay attention to the latest official datasets in future studies. In addition, this paper focuses on the UAs’ green innovation efficiency in China. Comparative studies on the green innovation efficiency of international UAs such as in America, Japan and Europe can be added in the future.

6. Conclusions

By constructing the input–output indicator system involving undesired outputs, this paper conducted an empirical study on influential factors of green innovation efficiency using SDM based on large-dimensional data for the period of 2006–2018 in the UAs of China. The main conclusions of the study are as follows:

From the perspective of temporal variation, the green innovation efficiency of most UAs in China presents a fluctuated increase during the study period; however, the efficiency is still relatively low. Very few UAs, such as the Central Shanxi UA, show a downward trend. Considering the mean value, UAs located in the east are much more ideal, while UAs such as the Beibu Gulf and Ningxia Yanhuang located in the central and western regions possess a low green innovation level.

As for spatial variation and regional differences, there exists a core and a stepped pattern of “east > center > west” for the southeast coastal UAs. The number of UAs of a high value level is relatively stable, while others presented a certain degree of fluctuation. The gap in the green innovation efficiency for UAs of China is narrowing, and the difference between UAs is greater than within, indicating that an imbalance is mainly reflected among UAs.

Regarding driving factors, a significant positive spatial spillover effect of green innovation efficiency does exist in UAs of China. Optimizing the industrial structure and increasing the industry competition have a direct positive impact on the green innovation efficiency, while the impact of financial support is not significant. Furthermore, the influences of other factors, such as economic scale, urbanization, informatization, government orientation and opening up, on the green innovation efficiency of UAs vary across the eastern, central and western UAs.