Evaluation of Green Innovation Efficiency in Chinese Provincial Regions under High-Quality Development and Its Influencing Factors: An Empirical Study Based on Hybrid Data Envelopment Analysis and Multilevel Mixed-Effects Tobit Models

Abstract

In the context of China’s high-quality economic development, in-depth research on green innovation efficiency and its influencing factors is crucial for promoting economic transformation and energy conservation. This study employs the Hybrid Data Envelopment Analysis (DEA) method to measure the green innovation efficiency of 30 provinces in China from 2013 to 2019. Subsequently, based on the Multilevel Mixed-Effects (MME) Tobit model and a spatial econometric model, the study investigates the factors influencing green innovation efficiency under the backdrop of high-quality development, and conducts various robustness tests from different perspectives. The results indicate the following: Firstly, the overall level of green innovation efficiency in China is relatively low, but it shows a steady growth trend, with significant differences in green innovation efficiency among provinces in different stages of high-quality development. Secondly, the level of digital economic development, optimization of industrial structure, scale of knowledge dissemination, and degree of openness to the outside world have significant positive effects on green innovation efficiency. On the other hand, the scale of technological innovation, degree of environmental regulation, and guarantee of green innovation have significant negative effects, and the low quality of technological innovation hinders the improvement of green innovation efficiency. Thirdly, the new factors emerging under the backdrop of high-quality development exhibit certain spillover effects on green innovation efficiency. The green innovation efficiency of a province may be influenced by relevant factors in neighboring provinces. This provides new insights for provinces to enhance their green innovation efficiency. The contribution of this study lies in the incorporation of newly emerged factors in the context of high-quality development into the evaluation framework of green innovation efficiency. It accurately measures the green innovation efficiency of each province in China and, based on the analysis of influencing factors, provides novel insights for enhancing green innovation efficiency across provinces.

1. Introduction

In recent years, the issue of climate change caused by the high consumption of fossil energy has attracted significant attention from the international community. To address global climate change and achieve sustainable economic development, many countries have reached important agreements such as the Kyoto Protocol (1997) and the Paris Agreement in 2015, aiming to reduce environmental impact through low-carbon transformation of economic development. For developing countries that heavily rely on fossil energy for economic growth, the transformation and upgrading of their economic development models are particularly crucial for global climate governance.

As the world’s largest developing country, China has achieved rapid economic growth in the past under a factor-driven and investment-driven development model. However, this predominantly extensive economic development, characterized by scale expansion, has led to significant consumption of fossil fuels and natural resources, resulting in a surge of industrial solid waste and emissions of exhaust gases, leading to numerous environmental pollution issues. Currently, China faces prominent downward risks in its macroeconomic situation, along with further degradation of ecosystems and increased “environmental pressure” caused by climate change, which further constrains economic development [1]. Consequently, the previous economic development model in China has become unsustainable. To address the mounting economic pressures and environmental pollution issues, the 19th National Congress of the Communist Party of China (CPC) introduced the concept of “high-quality development”, signifying the shift from a phase of high-speed growth to one of high-quality development [2]. In response to this, the Chinese government has embarked on efforts to transform the economic development model through technological innovation, the development of a green economy, and optimization of industrial structures. The focus of economic development has gradually shifted towards a high-quality development model characterized by “innovation”, “coordination”, “greenness”, “openness”, and “shared benefits”. Within the context of China’s high-quality economic development, several new factors have emerged to drive the transformation of economic development. For instance, the rapid growth of China’s digital economy continuously facilitates industrial restructuring, while the promotion of a low-carbon economy has significantly reduced China’s carbon emissions. Particularly, green innovation, characterized by resource conservation and environmental improvement, has become a vital driving force for establishing a sound, green, low-carbon, and circular economic development system in China, advancing innovation-driven development strategies, achieving the “peak carbon emissions” before 2030, and realizing “carbon neutrality” before 2060. The level of green innovation efficiency will determine the speed, quality, and sustainability of China’s economic development. Therefore, the key to decoupling economic development from environmental pollution, completing the transformation of the economic development model, and reducing the consumption of fossil energy and resource wastage lies in improving green innovation efficiency.

Given the aforementioned background, it becomes imperative to contemplate the following questions: What is the current status of green innovation efficiency in China? How do provincial regions differ in terms of green innovation efficiency across different stages of high-quality development? What impacts do new factors under high-quality development have on green innovation efficiency? How can we effectively promote further improvements in green innovation efficiency? Such issues demand thorough exploration and examination.

Therefore, this paper conducts a measurement and analysis of the provincial green innovation efficiency in China under high-quality development and explores the differences in green innovation efficiency among provinces at different stages of high-quality development. This provides targeted policy guidance for enhancing green innovation efficiency in different provinces. The innovative contributions of this study lie in: (i) constructing a green innovation evaluation system that aligns with the current context of China’s high-quality economic development and analyzing the differences in green innovation efficiency among provinces at different stages of high-quality development; (ii) considering both the radial and non-radial relationships between green innovation inputs and outputs when measuring the green innovation efficiency of each province, resulting in greater measurement accuracy; (iii) exploring the impact of new factors emerging under China’s high-quality development context on green innovation efficiency, thereby providing effective references for enhancing green innovation efficiency in the context of high-quality development in China; and (iv) further considering the impact of spatial spillover effects, thereby providing effective references for collaborative improvement of green innovation efficiency among provinces.

The following is the organizational structure of the subsequent content: Section 2 provides a literature review on the concept of green innovation, measurement of green innovation efficiency, and its influencing factors. Section 3 introduces the evaluation system for green innovation efficiency, the evaluation model for green innovation efficiency in the context of high-quality development, and regression models for the influencing factors of green innovation efficiency. Section 4 conducts a measurement and analysis of provincial green innovation efficiency in China. Section 5 validates the different influencing factors of provincial green innovation efficiency in the context of high-quality development in China. Section 6 discusses the results of this study. Finally, Section 7 summarizes the conclusions, policy implications, and research limitations of the paper.

2. Literature Review

2.1. Research Related to Green Innovation and Its Measurement

The concept of green innovation was first proposed by James in 1996, who referred to new processes or products that provide value while reducing environmental pollution as green innovation [3]. Currently, research on green innovation is still in its early stages, and the academic community has not yet reached a unified definition of green innovation [4]. From a micro perspective, Kemp et al. defined ecological innovation as “new products, services, and management methods that reduce environmental risks, pollution, and other negative impacts through resource use” [5]. Furthermore, Ying et al. consider green innovation to be any innovative activity that promotes the coordinated development of energy, the economy, and the environment and contributes to energy conservation and emission reduction [6]. From a broader perspective, Ma believes that green economic development aims to achieve the harmonious integration of economic development, social progress, and ecological construction while considering global climate change and ecological crises [7]. Li categorizes green innovation as falling within the realm of “weak sustainability” and states that any innovation activity that reduces environmental pollution and resource consumption can be regarded as green innovation [8]. Compared with traditional innovation, green innovation should generate certain environmental benefits [9]. Overall, there are generally three main definitions of green innovation in the academic community, namely, reducing environmental pollution, introducing environmental performance, and the innovation or improvement of environmental performance [10].

The core of green innovation lies in improving resource allocation and reducing environmental pollution, with the key focus on enhancing green innovation efficiency [11]. Green innovation efficiency refers to the efficiency of transforming green innovation inputs, which contribute to reducing environmental pollution, into green innovation outputs. It represents the input–output ratio of green innovation and consists of scale efficiency and pure technical efficiency. The measurement of green innovation efficiency mainly involves non-parametric methods such as Data Envelopment Analysis (DEA) and parametric methods such as Stochastic Frontier Analysis (SFA) [12]. The SFA method was initially proposed by Aigner et al. and Broeck et al. [13,14]. This method requires the establishment of specific production forms in advance and can only be used for efficiency measurement of single-output systems. In contrast, the DEA method can handle efficiency measurement problems with multiple inputs and outputs, without the need to specify production forms and input weights in advance. It provides a relatively fair evaluation of decision unit efficiency and is not influenced by subjective factors. Moreover, researchers have developed various models such as two-stage DEA, three-stage DEA, super-efficiency DEA, network DEA, and Hybrid DEA to overcome the shortcomings of traditional DEA methods, leading to significant development and application of the DEA method.

Given the superior performance of the DEA method, researchers often use different DEA models to measure green innovation efficiency. For example, from an enterprise perspective, Feng used the DEA-SBM method to measure the green innovation efficiency of industrial enterprises in 30 provinces in China and found that green innovation efficiency is related to the level of economic development [15]. He Feng et al. used the network SBM-DEA model to calculate the green technology efficiency of Chinese steel enterprises and found that the green technology efficiency of steel enterprises in the eastern, central, and western regions of China decreases sequentially [16]. In terms of measuring green innovation efficiency at the provincial level in China, Tan et al. used the DEA-SBM model to measure the ecological efficiency level of Jiangsu Province and found that ecological efficiency in Jiangsu Province shows a decreasing trend from south to north [17]. Tao Xue ping et al. used the non-separable input–output SBM method to measure the provincial green economic efficiency in China from 1995 to 2012 and found significant regional differences in green economic efficiency [18]. Wang et al. in the context of carbon neutrality, used the super-efficiency DEA model to measure the energy efficiency level of Chinese regions and found that the energy efficiency level in China is relatively low, with a gradual reduction in regional differences [19].

Considering the application of two-stage and three-stage DEA models, Yuan et al. evaluated the innovation efficiency of large-scale manufacturing enterprises in 30 provinces in China from the perspective of green growth using a three-stage DEA model, and found significant differences in the average innovation efficiency of large-scale manufacturing enterprises between the eastern and western regions of China, with scale efficiency being the main factor hindering the improvement of innovation efficiency [20]. Li et al. explored the current status of industrial green innovation efficiency in Chinese provinces using the SBM-DEA three-stage method, and found that industrial structure and economic development level are important factors hindering the improvement of industrial green innovation efficiency [21]. Zhang et al. used an improved three-stage DEA model to calculate the real green technological innovation efficiency of Chinese industrial enterprises and found that enterprises passively engaging in green technological innovation under external pressures is a significant reason for the low green technological innovation efficiency [22]. Wen et al. measured the industrial green technology efficiency of Chinese provinces using a two-stage DEA model and found that there is significant room for improvement in the green research and development and transformation efficiency of provincial industries in China [23]. Wang et al. used a three-stage DEA model to calculate the innovation factor allocation efficiency of provinces in China, and the results showed significant regional disparities in innovation factor allocation in China [24]. In light of this, the widespread use of the DEA method fully demonstrates the reliability and effectiveness of using the DEA method to analyze green innovation efficiency.

2.2. Research on Factors Influencing Green Innovation Efficiency

For the study of factors affecting green innovation efficiency, scholars often combine the DEA model with the Tobit regression model as green innovation efficiency is a constrained variable. From the perspective of higher education institutions, Guo et al. employed the DEA-Tobit model to analyze the input–output efficiency of scientific and technological research in various provinces of China’s higher education institutions and found that the short-term output of research activities was relatively low. Additionally, the lack of management in technological resources and research funding were identified as the three main factors leading to the lower research efficiency in higher education institutions [25]. Wang et al. used the DEA-Tobit model to analyze the innovation efficiency of 17 universities in Shanghai, revealing an inverted “U”-shaped influence of physical capital on the research and innovation efficiency of these universities [26].

From the perspective of enterprises, You et al. analyzed the factors influencing the green efficiency of 24 listed photovoltaic enterprises in China and found that the proportion of research and development personnel and the attitude of enterprise managers towards the environment had a positive impact on the green development of photovoltaic enterprises [27]. Jiang et al. utilized the DEA-Tobit model and discovered that the level of human capital and the investment in human capital had a significant negative effect on the green innovation performance of China’s manufacturing industry [28]. Cheng et al., employing the DEA-Tobit model, revealed an upward trend in China’s industrial green technology innovation efficiency. They identified factors such as the degree of openness, marketization, and environmental regulation intensity as important factors promoting the improvement of green technology efficiency [29]. Based on the DEA-Tobit model, Luigi et al. found that the land spillover efficiency had a negative impact on the efficiency indicators of American enterprises [30].

From the perspectives of cities and regions, Tian et al. using the SBM-Malmquist-Tobit model, analyzed the spatiotemporal dynamic characteristics and influencing factors of green development efficiency in 22 coal resource-based cities in China from 2007 to 2019. They found that the total factor productivity of coal resource-based cities was relatively low but showed a small fluctuating and stable upward trend [31]. Li et al. utilized the Tobit model to study the factors influencing the green innovation efficiency of 17 cities in Shandong Province, China. They discovered that the intensity of environmental regulation, the degree of economic openness, the optimization of industrial structure, and the level of government funding had significant positive effects on green innovation efficiency [32]. Based on the DEA-Malmquist and Tobit models, Wu et al. analyzed the trend of agricultural carbon emission efficiency in 31 provinces of China, revealing provincial disparities in the main contributing factors to agricultural carbon emission efficiency and its index [33]. Chen et al. utilized the super-efficiency DEA method and Tobit model to analyze the environmental efficiency and influencing factors in the Beijing–Tianjin–Hebei region, finding that the overall environmental efficiency in the region was poor, with technological progress being the main obstacle to environmental efficiency growth [34]. Liu et al., using the three-stage DEA-Tobit model, found that the urbanization rate had a significant positive effect on the green growth efficiency of the Yangtze River Economic Belt and the Yellow River Basin in China. However, external openness had a restraining effect on the green growth efficiency of the Yangtze River Economic Belt, and the improvement of financial development and human capital level had a negative impact on the green growth efficiency of the Yellow River Basin [35]. Sun et al. used the SBM-Malmquist-Tobit model to evaluate and analyze the green economic efficiency of key provinces in the “Belt and Road” initiative, and found that the development gap in green economic development between regions in China was narrowing [36]. In summary, these studies confirm the strong reliability and applicability of the DEA-Tobit method.

In addition, many scholars have investigated the factors influencing green innovation efficiency using non-Tobit methods. For example, Kong et al. employed a spatial error model to analyze the factors influencing green innovation efficiency in various provinces of China and found that China needs to further strengthen environmental regulations and increase investment in innovation efficiency [37]. Li et al. studied the factors influencing the green innovation efficiency of 12 prefecture-level cities in Hubei Province using a stochastic frontier model, and found an inverted “U”-shaped relationship between the intensity of environmental regulation and green innovation efficiency [38]. Shen et al. investigated the mechanism of action of green innovation efficiency and key factors using a structural equation model in China. They found a positive correlation between green innovation foundation and green innovation and technological innovation capabilities [39]. Li et al., employing the SBM-DEA model and generalized least squares model, explored the regional innovation-based green development efficiency and its influencing factors in China. The results showed a positive correlation between regional economic development level and regional green innovation efficiency. The level of urbanization and environmental regulation played positive and negative driving roles, respectively, in the national innovation-based green development efficiency [40]. Lv et al. analyzed the factors influencing China’s green innovation efficiency using the difference GMM model and system GMM model. They found that research funding, financial development efficiency, and per capita GDP were positively correlated with green innovation efficiency, while the proportion of the tertiary industry restricted the improvement of China’s green innovation efficiency [41].

2.3. Literature Evaluation

In the literature, we have found evident spatiotemporal characteristics in the green innovation efficiency across various provinces in China. It is in a steady growth state and influenced by multiple factors such as research personnel and funding input, industrial structure, environmental regulation level, and economic development level. However, there are several shortcomings identified in the literature. Firstly, as mentioned by Ma, “green innovation should consider the current global climate change situation” [7]. Green innovation should be closely integrated with the current national situation and global climate conditions in China. However, few scholars have constructed an evaluation system for green innovation efficiency that aligns with China’s current national context, including high-quality development, achieving peak carbon by 2030, and carbon neutrality by 2060. Secondly, the economic development level is an important influencing factor for green innovation efficiency, but the existing literature lacks research on the differences and spatial spillover effects of green innovation efficiency among provinces in different stages of high-quality economic development. Thirdly, in constructing the indicator system for factors affecting green innovation efficiency, the existing literature pays little attention to various new factors emerging in the context of China’s high-quality development, leading to limited policy implications from the obtained results. Fourthly, in terms of models, the aforementioned literature does not simultaneously consider the radial and non-radial relationships between green innovation input and output when evaluating green innovation efficiency. There is also less consideration of the mixed effects of fixed and random effects on green innovation efficiency, which may lead to significant errors in efficiency evaluation and impact factor analysis.

Based on this, the present study aims to construct a green innovation evaluation system in the context of China’s high-quality development. Furthermore, it takes into account both the radial and non-radial relationships between green innovation inputs and outputs. The Hybrid Data Envelopment Analysis (DEA) method is employed to accurately measure the green innovation efficiency of each province. Additionally, the study analyzes the differences in green innovation efficiency among provinces at different stages of economic high-quality development. Based on the results of the analysis of factors influencing green innovation efficiency, targeted policy recommendations are proposed. This research not only addresses the research needs in line with the high-quality development context but also fills gaps in the existing literature.

3. Methodology

Due to the involvement of multiple models and the complexity of the analytical process in this study, this paper visually presents the flow of green innovation efficiency evaluation, analysis of factors influencing green innovation efficiency, and analysis of green innovation efficiency spillover effects in a process diagram. The specific flowchart of the research design is shown in Figure 1, which illustrates the research methodology employed throughout the entire study.

3.1. Green Innovation Efficiency Measurement Methodology Design

3.1.1. Principles of the Hybrid-DEA Model

The DEA model was initially proposed by Charnes et al. [42]. It can handle multiple input and output variables simultaneously and is known as the Constant Returns to Scale (CRS) CCR-DEA model as it assumes constant returns to scale for each decision-making unit. To address the limitations of the CCR-DEA model, Banker et al. introduced the BCC-DEA model [43], which allows measurement of the efficiency of decision-making units under conditions of constant returns to scale, increasing returns to scale, and decreasing returns to scale. Since the introduction of the DEA model, numerous scholars have proposed different DEA models to address efficiency evaluation problems, leading to rapid development in the research and application of DEA models. In general, all DEA models can be classified into radial efficiency models, as represented by the CCR and BCC models, and non-radial efficiency models, as represented by the SBM model [44]. Radial efficiency models assume that the input and output variables can be proportionally adjusted to bring decision-making units to the efficiency frontier, but they ignore the non-radial relationships between input and output variables, i.e., non-radial input–output deviations. Non-radial efficiency models address non-radial input–output deviations but neglect the radial relationships between inputs and outputs [45].

Thus, Tone proposed a mixed model called the Hybrid DEA model [46], which can simultaneously handle the radial and non-radial relationships between input and output variables to address the aforementioned issues. The Hybrid model defines the input and output variable matrices as $X^R∊R_{+}^{m\times n}$ and $Y^R∊R_{+}^{s\times n}$, respectively. The input matrix X and output matrix Y are decomposed into radial component matrices and non-radial component matrices for analysis, enabling the determination of efficiency values for both radial input and output variables, as well as non-radial input and output variables. The formulas used in the Hybrid model proposed by Tone are as follows:

$$\begin{gathered} X^R∊R_{+}^{m1\times n},X^{NR}∊R_{+}^{m2\times n},X=\left(\genfrac{}{}{0pt}{}{X^R}{X^{NR}}\right) \\ {Y}^R∊R_{+}^{\mathrm{s}1\times n},Y^{NR}∊R_{+}^{s2\times n},X=\left(\genfrac{}{}{0pt}{}{X^R}{X^{NR}}\right) \\ m=m1+m2,s=s1+s2 \\ p=\left\{\right(x,y)\mid x\ge X\lambda ,y\le Y\lambda ,\lambda \ge 0\} \end{gathered}$$

(1)

For a specific DMU:

$$\begin{gathered} DMU\left(x_0,y_0\right)=(x_0^R,x_o^{NR},y_0^R,y_0^{NR})∊P \\ \theta x_0^R=X^R\lambda +s^{R-} \\ {x}_0^{NR}=X^{NR}\lambda +s^{NR-} \\ \varphi y_0^R=Y^R\lambda -s^{R+} \\ {y}_0^{NR}=Y^{NR}\lambda -s^{NR+} \end{gathered}$$

(2)

Among them, θ ≤ 1, ϕ ≥ 1, λ ≥ 0, $s^{R-}$ ≥ 0, $s^{NR-}$ ≥ 0, $s^{R+}$ ≥ 0, $s^{NR+}$ ≥ 0, $s^{R-}∊R^{m1}$ and $s^{NR-}∊R^{m2}$ represent the excessive radial inputs and non-radial inputs, respectively, while $s^{R+}∊R^{s1}$ and $s^{NR+}∊R^{s2}$ represent the insufficient radial and non-radial outputs.

We analyze the efficiency value of a specific $DMU\left(x_0,y_0\right)$ as follows:

$$\begin{gathered} {\rho }^{\mathrm{*}}=\mathrm{m}\mathrm{i}\mathrm{n}\frac{1-\frac{m1}{m}\left(1-\theta \right)-\frac{1}{m}\sum _{i=1}^{m2}{s}_i^{NR-}/x_{io}^{NR}}{1+\frac{s1}{s}\left(\varphi -1\right)+\frac{1}{s}\sum _{r=1}^{s2}{s}_r^{NR+}/y_{ro}^{NR}} \\ st.\theta x_0^R\ge X^R\lambda \\ {x}_0^{NR}=X^{NR}\lambda +s^{NR-} \\ \varphi y_0^R\le Y^R\lambda \\ {y}_0^{NR}=Y^{NR}\lambda -s^{NR+} \\ \theta \le 1,\varphi \ge 1,\lambda \ge 0,s^{NR-}\ge 0,s^{NR+}\ge 0 \end{gathered}$$

(3)

Based on Equation (3), we obtain (${\theta }^{\mathrm{*}},{\varphi }^{\mathrm{*}},{\lambda }^{\mathrm{*}},s^{NR-\mathrm{*}},s^{NR+\mathrm{*}}$). The efficiency indicator ${\rho }^{\mathrm{*}}$ can be decomposed as follows:

Inefficiency of radial inputs:

$${\alpha }_1=\frac{m_1}{m}\left(1-{\theta }^{\mathrm{*}}\right)$$

(4)

Inefficiency of non-radial inputs:

$${\alpha }_2=\frac{1}{m}{\sum }_{i=1}^{m_2}{s}_i^{NR-\mathrm{*}}/x_{io}^{NR}$$

(5)

Inefficiency of radial outputs:

$${\beta }_1=\frac{s_1}{s}({\varphi }^{\mathrm{*}}-1)$$

(6)

Inefficiency of non-radial outputs:

$${\beta }_2=\frac{1}{s}{\sum }_{r=1}^{s_2}{s}_r^{NR+\mathrm{*}}/y_{ro}^{NR}$$

(7)

Input inefficiency:

$$\alpha ={\alpha }_1+{\alpha }_2$$

(8)

Output inefficiency:

$$\beta ={\beta }_1+{\beta }_2$$

(9)

where n, m, and s represent the number of decision-making units, the number of input variables, and the number of output variables, respectively. $m_1$ and $m_2$ represent the number of radial input variables and non-radial output variables, while $s_1$ and $s_2$ represent the number of radial output variables and non-radial output variables. λ is a non-negative vector.

3.1.2. Selection of Input and Output Indicators

The selection of input and output indicators in calculating the green innovation efficiency at the provincial level in China using the Hybrid-DEA model is crucial for accurately measuring green innovation efficiency. These indicators should effectively represent the input and output aspects of the green innovation process. Therefore, to select appropriate indicators for green innovation input and output, this study follows the principles of comprehensiveness, scientific rigor, operational feasibility, and data availability in constructing the indicator system. Drawing upon the relevant literature [38,39,47,48,49], we have chosen a set of representative green innovation input and output indicators.

In terms of inputs, this study considers full-time equivalent of R&D personnel, internal expenditure on R&D, and electricity consumption as indicators of green innovation inputs. The full-time equivalent of R&D personnel reflects the level of human resource investment in green innovation in a region. A higher full-time equivalent indicates a greater human resource investment in green innovation in that region. Internal expenditure on R&D represents the actual level of financial investment in green innovation and is an important indicator for measuring the green innovation capacity. Higher internal expenditure on R&D indicates a greater consumption of financial resources for green innovation. Since the green innovation process requires energy consumption, electricity consumption is used as an indicator of energy input in green innovation.

In terms of outputs, this study considers three types of patent grants, per capita regional GDP, and carbon dioxide emissions as indicators of green innovation outputs. The three types of patent grants are important indicators of knowledge output in green innovation, reflecting the quantity of achievements in green innovation activities. They serve as indicators of knowledge output in green innovation. Innovation is the core driving force behind economic development, and per capita regional GDP reflects the level of economic development and the capacity for green innovation output in a region. It serves as an indicator of economic output in green innovation. Green innovation can also lead to non-desired outputs, such as environmental pollution. Shen et al. [44] used the total emissions of “three wastes” as the indicator of non-desired outputs in green innovation. Considering the missing data for “three wastes” in certain years and the fact that achieving peak carbon by 2030 and carbon neutrality by 2060 has become a major national strategic goal for China, this study replaces the indicator of total emissions of “three wastes” with carbon dioxide emissions as the indicator of environmental pollution output in green innovation.

Table 1. Green innovation efficiency inputs, outputs and impact factor variables.

Type of Variable	Name of Variable	Description of Variable	Unit	Variable Symbols
Input variables	Energy input	Electricity consumption	100 million kW·h	EC
	Human resource input	Full-time equivalent of R&D personnel	man-year	HI
	Capital investment	Internal R&D expenditure	10,000 CNY	CI
Output variables	Knowledge output	Number of three types of patent grants	piece	KO
	Economic output	Per capita GDP	CNY	EO
	Environmental pollution	Total carbon dioxide emissions	Mt	EP
Impact factor variables	Level of digital economic development	Digital economic development index	-	LDED
	Optimization of industrial structure	Proportion of the tertiary industry in GDP	%	ISO
	Scale of technological innovation	Number of R&D projects in higher education institutions	piece	STI
	Degree of environmental regulation	Investment in industrial pollution control projects completed in the current year	10,000 CNY	DER
	Scale of knowledge dissemination	Technology market turnover	10,000 CNY	SKD
	Degree of openness to the outside world	Total import and export of goods by foreign-invested enterprises	10,000 USD	EOW
	Green innovation guarantee	Social security and employment expenditure	100 million CNY	GIG

Source: The Authors.

presents the selected input and output variables for green innovation efficiency in this study.

3.2. Regression Model Establishment

3.2.1. Selection and Definition of Factors Affecting Green Innovation Efficiency

The efficiency of green innovation is not only determined by the input and output of green innovation but also influenced by various external factors. Exploring the impact of these external factors on green innovation efficiency can guide the government in formulating rational green innovation policies, and thereby create a favorable environment for green innovation and promote the improvement of green innovation efficiency in different provinces. This study incorporates the research of multiple scholars [37,50,51,52,53,54], while considering new factors emerging under high-quality development. The explanatory variables were selected as follows: (1) Digital economic development level: The digital economy, including digital industrialization and industrial digitalization, serves as a new driving force for the structural transformation of the Chinese economy under the background of high-quality development. It also provides a new pathway for improving green innovation efficiency. In this study, the Digital Economic Development Index obtained from the “China Provincial Digital Economy Index Estimation” in the Mark Data Network was selected to represent the level of digital economic development. (2) Optimization of industrial structure: Improving the industrial structure is an important indicator of the transformation of the economic development model under the background of high-quality development. Optimizing and adjusting the industrial structure can reduce the environmental impact caused by economic development. In this study, the proportion of the tertiary industry of GDP was selected to represent the industrial structure. (3) Scale of technological innovation: Under the background of high-quality development, technological innovation is the primary driving force for economic development. The larger the scale of technological innovation, the stronger the driving force for high-quality economic transformation, and the more conducive it is to improving green innovation capabilities. In this study, the number of R&D projects in higher education institutions was selected to represent the scale of technological innovation. (4) Environmental regulation level: Under the high-quality development in China, the government pays special attention to environmental governance during the process of economic development. The government aims to promote energy conservation, emission reduction, and environmental pollution control through environmental regulations, thereby promoting green development. The level of environmental regulation represents the degree to which the government values green innovation and is an important factor determining the external environment for green innovation. In this study, the annual investment in industrial pollution control projects was selected to represent the environmental regulation level. (5) Scale of knowledge dissemination: The dissemination of green innovation knowledge contributes to the transformation of such knowledge into actual economic output. The scale of knowledge dissemination in green innovation can reflect to some extent the application scale of green innovation products. In this study, the transaction volume in the technology market was selected to measure the scale of knowledge dissemination for green innovation products. (6) Degree of openness to the outside world: Promoting high-level opening-up is an important aspect of China’s high-quality development. Actively advancing high-level opening-up can facilitate the introduction of advanced green innovation technologies, talents, and the promotion of more advanced business models from abroad. It can also expand the sales channels for green innovative products. In this study, the total value of goods imported and exported by foreign-invested enterprises is selected as an indicator to represent the degree of openness to the outside world. (7) Green innovation guarantee: The implementation of green innovation activities relies on a solid foundation of full employment of scientific research personnel and social stability. Therefore, in this study, social security and employment expenditures are chosen as indicators to measure the level of green innovation guarantee. These factors are shown in Table 1.

3.2.2. Multilevel Mixed-Effects Tobit

Due to the numerous factors affecting the actual green innovation efficiency of each province, conducting regression analysis with all these factors as independent variables would not only require a tremendous amount of work but also increase the risk of multicollinearity, thereby compromising the reliability of the regression results. Therefore, this study primarily selects representative factors in the context of China’s high-quality development for regression analysis. However, this method of indicator selection may also introduce the issue of omitted variables to some extent. To address these concerns, this paper proposes a multilevel mixed-effects Tobit (MME-Tobit) model, which incorporates both individual fixed effects capturing the inter-provincial differences in green innovation efficiency and random time effects to account for temporal variations. By doing so, it aims to minimize endogeneity concerns and accurately analyze the influencing factors of green innovation efficiency across provinces. The formulation of the MME-Tobit model is as follows:

$$\begin{gathered} Y_{ij}^{\mathrm{*}}={\beta }_0+{\mu }_j+({\beta }_1+v_j)x_{ij}+{\epsilon }_{ij} \\ {Y}_{ij}=\left\{\begin{array}{c}0 , Y_{ij}^{\mathrm{*}}<0 \\ Y_{ij}^{\mathrm{*}}, 0<Y_{ij}^{\mathrm{*}}<1 \\ 1 , Y_{ij}^{\mathrm{*}}<1\end{array}\right. \\ \epsilon ~N\left(0,{\sigma }_{\epsilon }^2\right),\mu ~N\left(0,{\sigma }_{\mu }^2\right),v~N\left(0,{\sigma }_v^2\right) \end{gathered}$$

(10)

In the equation, $Y$represents the green innovation efficiency of each province, $x$represents the influencing factors of green innovation, ${\beta }_0$ and ${\beta }_1$ are the constant terms,$i$ denotes the individual, $j$ represents the stratification variable, $\mu$ represents the random intercepts, $v$ represents the random slopes, and $\epsilon$ represents the random error term.

3.2.3. Spatial Durbin Model

To further investigate the spatial effects of regional green innovation efficiency in China, this study employs spatial error, spatial lag, and spatial Durbin models for analysis. Through subsequent tests, we find that the fixed-effects spatial Durbin model is the optimal model; thus, this study primarily utilizes the spatial Durbin model for exploration. The equation for the spatial Durbin model is presented below:

$$Y_{it}=\rho {\sum }_{j=1, j\ne i}^N{w}_{ij}{Y}_{it}+{\sum }_{k=1}^M{\gamma }_k{x}_{kit}+{\sum }_{k=1}^M{\theta }_k{\sum }_{j=1}^N{w}_{ij}{x}_{kit}+{\mu }_t+{\epsilon }_{it}$$

(11)

In the equation, $Y$ represents green innovation efficiency,$X$ represents the influencing factors of green innovation efficiency, $\gamma$ represents the weight of the influencing variables from the same province,$\theta$represents the weight of the influencing variables from other provinces, $w$represents spatial weights, $\rho$represents the spatial autoregressive coefficient, $i$and $j$ represent different provinces, $N$ represents the total number of provinces, $M$ represents the total number of influencing factor variables, $t$ represents the year, $k$ represents the $k$-th independent variable, $\mu$ represents time fixed effects, and $\epsilon$ represents the random error term.

In this study, we selected the economic distance matrix and economic geography distance matrix as the spatial weight matrices for conducting spatial econometric analysis. The formulas for the economic distance matrix and economic geography distance matrix are as follows:

$$\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{i}\mathrm{c} \mathrm{d}\mathrm{i}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e} \mathrm{m}\mathrm{a}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{x}: w_{ij}=\left\{\begin{array}{c} \frac{1}{{GD}_{ij}}, i\ne j\\ 0 , i\ne j\end{array}\right.$$

(12)

$$\mathrm{E}\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{o}\mathrm{m}\mathrm{i}\mathrm{c} \mathrm{g}\mathrm{e}\mathrm{o}\mathrm{g}\mathrm{r}\mathrm{a}\mathrm{p}\mathrm{h}\mathrm{y} \mathrm{d}\mathrm{i}\mathrm{s}\mathrm{t}\mathrm{a}\mathrm{n}\mathrm{c}\mathrm{e} \mathrm{m}\mathrm{a}\mathrm{t}\mathrm{r}\mathrm{i}\mathrm{x}: w_{ij}=\left\{\begin{array}{c} \frac{1}{{\mid GPC}_i-{GPC}_j\mid {GD}_{ij}}, i\ne j\\ 0, i\ne j\end{array}\right.$$

(13)

where ${GD}_{ij}$ represents the geographical distance between region i and region j, and ${GPC}_i-{GPC}_j$ represents the difference in per capita GDP between region i and region j.

3.3. Sample, Data Source, and Data Preprocessing

Due to significant data gaps in the Hong Kong Special Administrative Region, Macau Special Administrative Region, Tibet Autonomous Region, and Taiwan Province, as well as the suspension of data collection by relevant departments in China during the period of 2020–2022 due to the impact of the COVID-19 pandemic, the data collection process for this study has been greatly constrained. Thus, this study takes 30 provinces (including municipalities directly under the central government) in China as samples and constructs panel data on green innovation input–output and the influencing factors of green innovation efficiency at the provincial level from 2013 to 2019. Based on the classification of high-quality development levels of each province in China by Tang et al. (as shown in

Table 2. Distribution of high-quality economic development levels in Chinese provinces.

Type	Province
High Level	Beijing, Shanghai, Tianjin, Zhejiang, Guangdong, Jiangsu, Fujian
Intermediate–High Level	Shandong, Hainan, Fujian, Guizhou, Hubei, Chongqing, Inner Mongolia, Ningxia
Intermediate–Low Level	Jilin, Jiangxi, Heilongjiang, Anhui, Sichuan, Hunan, Henan
Low Level	Hebei, Liaoning, Qinghai, Guangxi, Shanxi, Yunnan, Gansu, Xinjiang

Source: Reference [55].

) [55], a comparative analysis of green innovation efficiency is conducted among provinces with different levels of high-quality development.

The data for regional electricity consumption, three types of granted patents, technology market turnover, scientific and technological expenditure, the proportion of the tertiary industry to regional GDP, total import and export value of foreign-invested enterprises, social security and employment expenditure, regional GDP, and year-end population are all derived from the China Statistical Yearbook for the years 2014–2020. Per capita GDP is calculated by dividing regional GDP by the year-end population. The digital economy development index is obtained from the Mark Data Network. Full-time equivalent R&D personnel, internal R&D expenditure, and the number of R&D projects in higher education institutions are sourced from the China Science and Technology Statistical Yearbook for the years 2014–2020. The completed investment in industrial pollution control projects is sourced from the China Environmental Statistical Yearbook for the years 2014–2020. Carbon dioxide emissions are sourced from the CEADS China Carbon Emission Database.

To account for the variations in data comparability across different years, the year 2013 is chosen as the base year, and the GDP deflator index is used to ensure the data are comparable for each province (municipality). Considering the significant numerical differences among the indicators of green innovation efficiency factors, this study applies a logarithmic transformation to nine indicators: technology market turnover, total investment in urban environmental infrastructure construction, completed investment in industrial pollution control projects, the number of research and development projects in research institutions, the number of research and development projects in higher education institutions, energy conservation and environmental protection expenditure, total foreign investment, total import and export value of foreign-invested enterprises, and social security and employment expenditure. As carbon dioxide emissions represent an undesirable output of green innovation, this study takes the reciprocal of carbon dioxide emissions.

3.4. Descriptive Statistical Analysis of Variables

Based on the preprocessed data of green innovation input, output, and influencing factor variables in various provinces of China from 2013 to 2019, descriptive statistics were generated using Stata 17 software for the relevant variables, as shown in

Table 3. Descriptive statistics for variables.

Type of Variable	Variable Symbols	Max	Min	Median	Mean	Std. Dev.
Input variables	EC	5235.230	232.020	1445.345	1816.591	1235.825
	HI	506,862.000	4731.000	94,189.500	120,682.267	127,776.972
	CI	3054.481	11.944	369.238	539.546	602.210
Output variables	KO	250,290.000	502.000	24,236.500	44,217.889	60,951.766
	EO	16.307	2.201	4.799	5.806	2.662
	EP	0.025	0.001	0.004	0.005	0.005
Impact factor variables	DLDE	0.701	0.073	0.184	0.206	0.112
	ISO	0.835	0.320	0.467	0.481	0.092
	STI	11.658	6.898	10.145	9.978	0.967
	DER	4.964	−1.068	2.884	2.825	1.008
	SKD	10.294	−0.442	4.957	4.852	1.781
	EOW	11.797	5.207	8.508	8.582	1.348
	GIG	7.464	4.630	6.501	6.388	0.590

Source: The Authors.

.

From Table 3, it can be observed that there are significant differences in the scale of green innovation inputs and outputs among Chinese provinces, indicating regional imbalances in green economic development. The high standard deviation of HI suggests significant disparities in green innovation human resource inputs among provinces in China. The average value of ISO is 0.481, indicating that there is still considerable room for optimizing the industrial structure in China.

4. Measurement Analysis of Provincial Green Innovation Efficiency in China

4.1. Results of Provincial Green Innovation Efficiency Measurement in China

Based on the Hybrid DEA model, the measurement of green innovation efficiency in Chinese provinces under variable scale conditions was conducted using DEA-SOLVER Pro8 software. The results are presented in

Table 4. Measurements of green innovation efficiency at the provincial level in China.

Type	DMU	2013	2014	2015	2016	2017	2018	2019	Average
High Level	Beijing	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
	Shanghai	0.445	0.540	0.564	0.514	0.645	0.664	0.670	0.577
	Tianjin	0.742	0.773	0.822	0.803	0.878	0.820	0.815	0.808
	Zhejiang	0.310	0.323	0.335	0.319	0.329	0.326	0.339	0.326
	Guangdong	0.140	0.152	0.210	0.225	0.227	0.223	0.221	0.200
	Jiangsu	0.168	0.173	0.176	0.161	0.167	0.166	0.131	0.163
	Fujian	0.368	0.371	0.485	0.530	0.508	0.484	0.465	0.459
	Average	0.453	0.476	0.513	0.507	0.536	0.526	0.520	0.505
Intermediate–High Level	Shandong	0.068	0.073	0.079	0.077	0.089	0.095	0.111	0.085
	Hainan	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
	Shaanxi	0.204	0.220	0.272	0.361	0.300	0.317	0.262	0.277
	Guizhou	0.456	0.524	0.540	0.389	0.440	0.462	0.495	0.472
	Hubei	0.151	0.162	0.193	0.190	0.233	0.254	0.236	0.203
	Chongqing	0.696	0.668	0.773	0.764	0.679	0.654	0.593	0.689
	Inner Mongolia	0.201	0.203	0.211	0.197	0.193	0.175	0.163	0.192
	Ningxia	0.822	0.845	0.763	0.857	0.708	0.661	0.511	0.738
	Average	0.450	0.462	0.479	0.479	0.455	0.452	0.421	0.457
Intermediate–Low Level	Jilin	0.391	0.403	0.431	0.462	0.470	0.446	0.406	0.430
	Jiangxi	0.268	0.389	0.491	0.546	0.524	0.526	0.482	0.461
	Heilongjiang	0.275	0.249	0.274	0.273	0.354	0.436	0.381	0.320
	Anhui	0.219	0.235	0.243	0.238	0.241	0.272	0.229	0.240
	Sichuan	0.227	0.254	0.332	0.307	0.304	0.368	0.301	0.299
	Hunan	0.177	0.212	0.216	0.200	0.215	0.226	0.225	0.210
	Henan	0.085	0.094	0.115	0.118	0.156	0.200	0.193	0.137
	Average	0.236	0.262	0.300	0.306	0.323	0.353	0.317	0.300
Low Level	Hebei	0.051	0.060	0.072	0.071	0.088	0.109	0.111	0.080
	Liaoning	0.143	0.146	0.176	0.148	0.139	0.141	0.151	0.149
	Qinghai	1.000	1.000	1.000	1.000	1.000	1.000	1.000	1.000
	Guangxi	0.225	0.296	0.379	0.386	0.430	0.441	0.422	0.368
	Shanxi	0.082	0.094	0.113	0.107	0.139	0.124	0.125	0.112
	Yunnan	0.283	0.362	0.364	0.349	0.388	0.392	0.425	0.366
	Gansu	0.343	0.362	0.412	0.467	0.680	0.767	0.700	0.533
	Xinjiang	0.293	0.304	0.337	0.254	0.310	0.300	0.284	0.297
	Average	0.303	0.328	0.357	0.348	0.397	0.409	0.402	0.363
Average		0.361	0.383	0.413	0.410	0.428	0.435	0.415	0.406

.

4.2. Results Analysis

From the perspective of economic quality development, as shown in Table 4, provinces in the “High” stage of quality economic development exhibit the best performance in green innovation efficiency. The annual overall mean values of green innovation efficiency for these provinces from 2013 to 2019 are 0.453, 0.476, 0.513, 0.507, 0.536, 0.526, and 0.520. Among the provinces in different stages of economic quality development, those in the “High” stage consistently demonstrate the highest annual overall mean values of green innovation efficiency, indicating that provinces in this stage are better at avoiding waste in green innovation inputs and possess strong green innovation capabilities. On the other hand, the provinces at the “Medium–Low” level of quality economic development exhibit the poorest performance in green innovation efficiency. The annual overall mean values of green innovation efficiency for these provinces from 2013 to 2019 are 0.236, 0.262, 0.300, 0.306, 0.323, 0.353, and 0.317. Among the provinces in different stages of economic quality development, those at “Medium–Low” level consistently have the lowest annual overall mean values of green innovation efficiency, indicating a relatively higher reliance on resource-intensive and environmentally polluting economic development models. These provinces require focused efforts to enhance their green innovation capabilities and optimize their economic development models in order to promote the realization of China’s “carbon peak” and “carbon neutrality” strategies.

From a spatial perspective, as shown in Figure 2, Beijing, Hainan, and Qinghai have the highest average green innovation efficiency from 2013 to 2019, and their green innovation efficiency is consistently DEA-efficient. This indicates that these provinces are at the forefront in effectively and reasonably utilizing green innovation inputs, with minimal environmental impact from their economic development models. On the other hand, provinces with average green innovation efficiency below 0.3 from 2013 to 2019 include Guangdong, Jiangsu, Shandong, Hubei, Inner Mongolia, Anhui, Sichuan, Hunan, Henan, Hebei, Liaoning, Shanxi, and Xinjiang. These provinces exhibit poor green innovation efficiency, with Shandong and Hebei having extremely low levels of 0.085 and 0.080, respectively. This suggests that the economic development models in these provinces have a significant environmental impact and that their innovation-driven capabilities are insufficient. It is necessary to vigorously promote energy conservation and emissions reduction efforts, reduce carbon emissions, enhance the research capabilities of R&D personnel, and improve the efficient utilization of R&D funds to enhance green innovation efficiency.

From a temporal perspective, as shown in Figure 3, the average green innovation efficiency of the provincial regions in China from 2013 to 2019 is 0.361, 0.383, 0.413, 0.410, 0.428, 0.435, and 0.415, respectively. Although there were slight declines in average green innovation efficiency in 2016 and 2019, the overall trend is upward. This indicates that the level of green innovation efficiency in China is relatively low, but through concerted efforts to promote the transformation of the economic development model and reduce carbon emissions and environmental pollution caused by economic development, there has been a gradual improvement in overall green innovation efficiency.

In summary, from 2013 to 2019, only three provinces in China achieved DEA-effective green innovation efficiency, while many provinces had green innovation efficiency below 0.3. The overall level of green innovation efficiency was relatively low, indicating that the energy consumption and environmental pollution issues caused by the current economic development model in China remain severe and that the transformation of the economic development model has not been fully realized. However, comparing the changes in green innovation efficiency among provinces over the years, it can be observed that China’s overall green innovation efficiency is on an upward trend. This indicates that in the context of high-quality development, China is gradually transitioning from a high-pollution, high-emission extensive economic development model to an innovative-driven, high-quality, and green economic development model. The impact of economic development on the environment is gradually diminishing, and the capability for green innovation is gradually improving.

5. Analysis of Factors Influencing Provincial Green Innovation Efficiency in China

5.1. Analysis of Benchmark Regression Results

Using panel data of green innovation and its influencing factors in Chinese provinces from 2013 to 2019, the MME-Tobit model was applied using Stata 17 software. The green innovation efficiency of each province was taken as the dependent variable, with time as the equation-level variable. A stepwise regression strategy was employed for the regression analysis, and the presentation follows a top-down approach based on the importance of each influencing factor in the context of high-quality development. Model (1)(2)(3) successively reduces the explanatory variables, while the benchmark regression model includes all explanatory variables. The regression results are presented in

Table 5. Benchmark regression results of the MME-Tobit model.

Variable	Benchmark Model	Model (1)	Model (2)	Model (3)
${DLDE}_i$	0.374 *** (2.97)	0.699 *** (3.67)	0.866 *** (4.61)	0.854 *** (4.56)
${ISO}_i$	0.969 *** (6.51)	1.126 *** (5.91)	1.129 *** (5.77)	1.230 *** (6.53)
${lnSTI}_i$	−0.099 *** (−4.04)	−0.244 *** (−8.30)	−0.183 *** (−7.61)	−0.153 *** (−9.15)
${lnDER}_i$	−0.129 *** (−11.33)	−0.175 *** (−11.56)	−0.169 *** (−11.17)	−0.169 *** (−11.16)
${lnSKD}_i$	0.044 *** (4.91)	0.026 ** (2.21)	0.022 * (1.84)
${lnEOW}_i$	0.017 *** (2.72)	0.029 *** (3.55)
${lnGIG}_i$	−0.302 *** (−12.80)
C	2.832 *** (14.74)	2.359 *** (9.27)	1.899 *** (8.58)	1.657 *** (9.76)
Year FE	Yes	Yes	Yes	Yes
Log likelihood	125.105 ***	71.367 ***	65.054 ***	63.351 ***
Wald Chi2	851.530	431.520	425.790	422.300
N	210	210	210	210

Note: *, **, and *** represent the significance levels of 1%, 5%, and 10%, respectively; the values in parentheses are the Z-statistics.

.

According to Table 5, in general, the log-likelihood value of the baseline regression model is 125.105, which is significant at the 1% level. The Wald test statistic is 851.53, indicating that the regression results of the baseline model are consistent with models (1)–(3). The log-likelihood values for all models are greater than 60 and significant at the 1% level. The Wald test statistics are all greater than 400, indicating that the regression results of the baseline model are reliable. The regression coefficients of the influencing factors are reliable, and the model fits well, effectively reflecting the impact of the influencing factors on green innovation efficiency.

For each explanatory variable, this study interprets the results in order of their respective importance. As shown in Table 5, the regression results of the benchmark model are as follows:

(1) The coefficient of digital economic development level is significantly positive at the 1% level, with a coefficient of 0.374, indicating a significant positive impact of digital economic development level on green innovation efficiency. The rise of digital economic development level facilitates industrial transformation and upgrading, as well as the improvement of green innovation efficiency. Promoting the construction of digital economic infrastructure and deep integration of the digital economy and real economy are important ways to enhance green innovation efficiency and transform China’s economic development model.

(2) The optimization of industrial structure has a significant positive impact on green innovation efficiency, with a regression coefficient of 0.969 and passing the significance test at the 1% level. This indicates that the higher the proportion of the tertiary industry in GDP, the smaller the environmental impact per unit of economic output, and the higher the green innovation efficiency. Therefore, vigorously developing the tertiary industry will help accelerate the pace of green economic development and enhance China’s green innovation efficiency.

(3) The coefficient of technological innovation scale is −0.099 and passes the significance test at the 1% level, indicating that the conversion of research results from higher education institutions to economic growth efficiency is insufficient. Although there are a large number of research outcomes, their quality is low, leading to substantial inputs of green innovation manpower and resources. Therefore, it is necessary to continuously improve the quality of research outcomes in higher education institutions, establish mechanisms that facilitate the conversion of research outcomes into economic growth, and balance research quality and quantity to enhance research efficiency.

(4) The degree of environmental regulation has a significant negative impact on green innovation efficiency, and it is significant at the 1% level. This indicates that while the government invests in industrial pollution control projects to address environmental pollution, it reduces the funds available for green innovation investment, which hinders the improvement of green innovation efficiency. Therefore, the government needs to control industrial pollution control project investment reasonably and reduce the impact of economic development on the environment by enhancing green innovation efficiency.

(5) The scale of knowledge dissemination has a significant positive impact on green innovation efficiency, with a regression coefficient of 0.044 and passing the significance test at the 1% level. This indicates that the volume of technology market transactions is an important influencing factor for green innovation efficiency. The larger the scale of knowledge dissemination, the greater the practical application of green innovation research outcomes and the stronger the knowledge spillover effect, thereby contributing to the improvement of green innovation efficiency.

(6) The coefficient of openness level is 0.017, which passes the significance test at the 1% level. This indicates that foreign trade helps adjust the imbalance in national economic development and promotes the sale of green innovation products, thereby enhancing green innovation efficiency.

(7) Green innovation guarantee has a significant negative impact on green innovation efficiency, with a regression coefficient of −0.302 and passing the significance test at the 1% level. This indicates that the more the government spends on social security and employment, and the more government financial expenditure is allocated, the less government investment is available for green innovation, which hampers the improvement of green innovation efficiency. Therefore, the government should enhance green innovation efficiency, establish an innovation-driven high-quality development model, and thereby improve social security and expand employment.

5.2. Robustness Test

To validate whether the benchmark regression results significantly change with the alteration of parameter settings and explore the reliability and stability of the benchmark regression results, it is necessary to conduct robustness tests on the benchmark regression model. This paper primarily conducts robustness tests on the benchmark model from four aspects. Firstly, from the perspective of econometric methods, different econometric models are employed to investigate the relationship between independent variables and the dependent variable. Secondly, from the perspective of the dependent variable, the dependent variable is replaced with green innovation efficiency under constant returns to scale. Thirdly, from the perspective of independent variables, the temporal lag of the endogenous factors with the dependent variable is considered, and the relevant influencing factors are subjected to one-period lag processing. Fourthly, from the perspective of data, specific independent variable data are randomized and the benchmark regression model is repeatedly utilized for regression analysis to examine the robustness of the influence of specific independent variables on the dependent variable.

5.2.1. Replacing Econometric Models

Using appropriate and different econometric methods, an analysis of the factors influencing green innovation efficiency is conducted. If the obtained results are consistent with the benchmark regression results, it can be considered that the robustness of the benchmark regression results in this study is high. Both the generalized linear mixed model and the multilevel mixed-effects Tobit model can simultaneously consider the effects of fixed and random effects on the differences in green innovation efficiency across provinces and over time, addressing the potential issue of omitted variables. Therefore, in this study, the generalized linear mixed model is employed to regressively analyze the factors influencing green innovation efficiency. The regression results are presented in

Table 6. Regression results of generalized linear mixed model.

Variable	Model (1)	Model (2)	Model (3)	Model (4)
${DLDE}_i$	0.267 ** (2.33)	0.439 ** (2.52)	0.573 *** (3.41)	0.578 *** (3.41)
${ISO}_{\mathrm{i}}$	0.906 *** (6.72)	1.097 *** (6.36)	1.092 *** (6.25)	1.155 *** (6.73)
${lnSTI}_{\mathrm{i}}$	−0.066 *** (−3.15)	−0.192 *** (−7.91)	−0.148 *** (−7.73)	−0.128 *** (−8.87)
${lnDER}_{\mathrm{i}}$	−0.106 *** (−10.99)	−0.141 *** (−10.99)	−0.140 *** (−10.79)	−0.140 *** (−10.73)
${lnSKD}_{\mathrm{i}}$	0.043 *** (5.35)	0.022 ** (2.15)	0.016 (1.61)
${lnEOW}_{\mathrm{i}}$	0.009 * (1.72)	−0.020 *** (2.85)
${lnGIG}_{\mathrm{i}}$	−0.296 *** (−12.97)
C	2.506 *** (15.66)	1.867 *** (9.42)	1.561 *** (9.19)	1.402 *** (10.07)
Year FE	Yes	Yes	Yes	Yes
Log likelihood	162.152 ***	102.046 ***	98.008 ***	96.720 ***
Wald chi2	1088.920	536.190	514.720	504.190
N	210	210	210	210

Note: *, **, and *** represent the significance levels of 1%, 5%, and 10%, respectively; the values in parentheses are the Z-statistics.

.

From Table 6, it can be observed that the results obtained from the analysis of the factors influencing green innovation efficiency using the generalized linear mixed model are generally consistent with the benchmark regression results. This indicates that the model used in the benchmark regression is appropriate and that the results obtained are robust.

5.2.2. Replacing the Dependent Variable

Considering the potential impact of the scale of returns on the significance of the factors influencing green innovation efficiency across provinces, this study utilizes the Hybrid DEA model to measure the green innovation efficiency of each province under the assumption of constant returns to scale. This measured efficiency is then used as the dependent variable in the multilevel mixed-effects Tobit model, with time as the hierarchical variable. A stepwise regression strategy is employed to examine the extent of influence of various factors. The regression results are presented in

Table 7. Regression results with replaced dependent variable.

Variable	Model (1)	Model (2)	Model (3)	Model (4)
${DLDE}_i$	0.331 *** (2.60)	0.615 *** (3.26)	0.781 *** (4.24)	0.772 *** (4.17)
${ISO}_i$	0.933 *** (6.08)	1.071 *** (5.72)	1.076 *** (5.60)	1.197 *** (6.41)
${lnSTI}_i$	−0.094 *** (−3.80)	−0.226 *** (−7.98)	−0.169 *** (−7.33)	−0.132 *** (−11.80)
${lnDER}_i$	−0.140 *** (−12.07)	−0.181 *** (−12.19)	−0.177 *** (−11.88)	−0.178 *** (−11.80)
${lnSKD}_i$	0.049 *** (5.29)	0.031 ** (2.77)	0.027 ** (2.37)
${lnEOW}_i$	0.016 *** (2.54)	0.027 *** (3.40)
${lnGIG}_i$	−0.278 *** (−11.64)
C	2.656 *** (13.75)	2.205 *** (9.27)	1.778 *** (8.43)	1.485 *** (9.05)
Year FE	Yes	Yes	Yes	Yes
Log likelihood	117.512 ***	71.486 ***	65.700 ***	62.873 ***
Wald Chi2	799.180	445.930	438.680	426.340
N	210	210	210	210

Note: ** and *** represent the significance levels of 5% and 10%, respectively; the values in parentheses are the Z-statistics.

.

Based on Table 7, when the green innovation efficiency under variable returns to scale is used as the dependent variable, the results of the factor analysis remain largely consistent with the benchmark regression results. This indicates that the variability of scale of returns to green innovation efficiency has minimal impact on the analysis of influencing factors, highlighting the robustness of the benchmark regression results.

5.2.3. One-Period Lag of the Influencing Factors

Considering the time-varying nature of the effects of various factors on green innovation efficiency and the potential endogeneity issue, where the current green innovation efficiency may be primarily influenced by factors such as the previous period’s technology market turnover and the investment completion of industrial pollution control projects in the current year, this study employs the MME-Tobit model to analyze the effects of the lagged factors on green innovation efficiency. This analysis is performed to test the robustness of the benchmark regression results. The results are presented in

Table 8. Regression results with lagged factors on green innovation efficiency.

Variable	Model (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)	Model (7)
${DLDE}_{\mathrm{i}}$		0.259 ** (1.96)	0.276 ** (2.17)	0.327 ** (2.39)	0.290 ** (2.27)	0.266 ** (2.08)	0.313 ** (2.48)
${ISO}_{\mathrm{i}}$	1.044 *** (6.50)		1.039 *** (6.59)	0.972 *** (6.13)	0.994 *** (6.24)	1.036 *** (6.57)	1.016 *** (6.52)
${lnSTI}_{\mathrm{i}}$	0.044 *** (4.52)	0.047 *** (4.90)		−0.075 *** (−2.83)	−0.081 *** (−3.07)	−0.087 *** (−3.31)	−0.087 *** (−3.29)
${lnDER}_{\mathrm{i}}$	−0.124 *** (−10.27)	−0.125 *** (−10.30)	−0.124 *** (−10.34)		−0.125 *** (−10.54)	−0.126 *** (−10.42)	−0.129 *** (−10.95)
${lnSKD}_{\mathrm{i}}$	0.083 *** (−3.09)	−0.095 *** (−3.48)	0.045 *** (4.58)	0.042 *** (4.37)		0.043 *** (4.45)	0.045 *** (4.66)
${lnEOW}_{\mathrm{i}}$	0.018 *** (2.64)	0.192 *** (2.82)	0.018 *** (2.68)	0.014 ** (2.10)	0.017 *** (2.59)		0.016 ** (2.34)
${lnGIG}_{\mathrm{i}}$	−0.324 *** (−12.63)	−0.319 *** (−12.34)	−0.322 *** (−12.62)	−0.317 *** (−11.83)	−0.324 *** (−12.71)	−0.321 *** (−12.55)
${DLDE}_{\mathrm{i}-1}$	0.283 * (1.95)
${ISO}_{\mathrm{i}-1}$		0.971 *** (6.24)
${lnSTI}_{\mathrm{i}-1}$			−0.089 *** (−3.26)
${lnDER}_{\mathrm{i}-1}$				−0.134 *** (−10.43)
${lnSKD}_{\mathrm{i}-1}$					0.045 *** (4.77)
${lnEOW}_{\mathrm{i}-1}$						0.020 *** (2.92)
${lnGIG}_{\mathrm{i}-1}$							−0.317 *** (13.12)
C	2.830 *** (13.33)	2.830 *** (13.33)	2.830 *** (13.33)	2.770 *** (13.18)	2.80 *** (13.47)	2.815 *** (13.57)	2.782 *** (13.24)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Log likelihood	106.807 ***	105.812 ***	107.697 ***	107.754 ***	108.370 ***	108.110 ***	110.520 ***
Wald chi	716.760	709.460	722.630	771.260	727.520	722.860	743.320
N	210	210	210	210	210	210	210

Note: *, **, and *** represent the significance levels of 1%, 5%, and 10%, respectively; the values in parentheses are the Z-statistics.

.

As shown in Table 8, after conducting the lagged processing of various factors on green innovation efficiency, the regression results remain largely consistent with the benchmark regression results. This indicates that the benchmark regression results in this study exhibit a high level of reliability.

5.2.4. Placebo Test

To test whether the benchmark regression results of green innovation efficiency are influenced by random factors, this study employs a placebo test using randomly shuffled green innovation efficiency impact factors. Firstly, the data for a specific impact factor are shuffled and replaced with the original data. Then the regression is rerun in the setting of the benchmark regression model. This process is repeated 1000 times. The coefficients, T-values, and p-values of the randomized impact factors in the regression results are recorded, and the average coefficients, T-values, and p-values of the randomized impact factors are calculated. Finally, kernel density estimation plots of the coefficients, scatter plots of p-values against coefficients, and kernel density estimation plots of T-values are generated. By observing the deviation between the mean values of the coefficients and T-values and the true values, as well as the distribution of p-values, it can be determined whether the impact of specific factors on green innovation efficiency is due to unobservable factors or omitted variables. If the mean deviation of the coefficients and T-values from the true values is low and the p-values are generally greater than 0.1, it can be concluded that the benchmark regression results of this study are robust and reliable. The placebo test results for all impact factors are shown in

Table 9. Results of placebo tests for each influence factor.

Variable	Placebo Test of Coefficients		Placebo Test of Z-Values		Placebo Test of p-Value Distribution
	Mean	True Value	Mean	True Value	Mean	True Value
${DLDE}_{\mathrm{i}}$	0.034	0.374	0.339	2.97	0.492	0.003
${ISO}_{\mathrm{i}}$	0.013	0.969	0.244	6.51	0.448	0.000
${lnSTI}_{\mathrm{i}}$	0.000	−0.099	−0.072	−4.04	0.513	0.000
${lnDER}_{\mathrm{i}}$	−0.004	−0.129	−0.399	−11.33	0.539	0.000
${lnSKD}_{\mathrm{i}}$	0.002	0.442	0.428	4.91	0.484	0.000
${lnEOW}_{\mathrm{i}}$	0.001	0.017	0.215	2.72	0.499	0.006
${lnGIG}_{\mathrm{i}}$	−0.001	−0.301	−0.166	−12.80	0.490	0.000

. Due to the large number of variables involved in the study, the kernel density estimation plots of the placebo test coefficients for the number of R&D projects in higher education institutions (Figure A1), the kernel density estimation of Z-values (Figure A2), and the scatter plot of p-values and coefficients (Figure A3) are presented in the Appendix A.

From Table 9, it can be observed that the average values of coefficients, average values of T-statistics, and deviations from the true values in the placebo test are relatively large for the level of digital economy development, optimization of industrial structure, scale of technological innovation, degree of environmental regulation, degree of openness, and green innovation guarantee. The average values of p-values are generally around 0.5. Moreover, the coefficient density estimation plots, scatter plots of p-values and coefficients, and density estimation plots of T-values for each influencing factor indicate that the absolute values of coefficients obtained from the placebo test are smaller than the true values. The majority of p-values are greater than 0.1, and the absolute values of the majority of T-values are smaller than the true values. These findings suggest that the baseline regression results of this study are not influenced by chance factors or omitted variables, demonstrating a high level of robustness and credibility.

5.3. Taking Spatial Spillover Effects into Further Consideration

5.3.1. Spatial Autocorrelation Analysis

The interdependence of green innovation efficiency data among provinces may exhibit spatial correlation. Therefore, this study employs commonly used spatial statistics in spatial econometrics, namely, Moran’s I index and Geary’s c index, to test for spatial correlation. Moran’s I index ranges from −1 to 1, while Geary’s c index ranges from 0 to 2. A positive Moran’s I index or a Geary’s c index between 0 and 1 indicates positive spatial correlation among provinces, suggesting spatial clustering. Conversely, a negative Moran’s I index or a Geary’s c index between 1 and 2 suggests negative spatial correlation among provinces, indicating spatial dispersion. When the Moran’s I index is zero or the Geary’s c index is one, it implies that the green innovation efficiency among regions is independent and not influenced by spatial distribution. The scatter plots of Moran’s I index for green innovation efficiency in each province of China in 2013, 2016, and 2019, based on the calculation of the economic geographical distance matrix, are presented in Appendix A as Figure A4, Figure A5 and Figure A6.

Based on the calculation of the economic–geographical distance matrix,

Table 10. Statistics for Moran’s I and Geary’s c indices.

Year	Moran’s I	Geary’s c
2013	0.178 ** (1.750)	0.833 * (−1.277)
2014	0.192 ** (1.855)	0.818 * (−1.397)
2015	0.181 ** (1.751)	0.835 * (−1.277)
2016	0.137 * (1.392)	0.867 (−1.035)
2017	0.228 ** (2.131)	0.778 ** (−1.0.35)
2018	0.224 ** (2.097)	0.765 ** (−1.830)
2019	0.249 *** (2.313)	0.754 ** (−1.897)

Note: *, **, and *** represent the significance levels of 1%, 5%, and 10%, respectively; the values in parentheses are the Z-statistics.

presents the Moran’s I and Geary’s c indices for the green innovation efficiency of Chinese provinces from 2013 to 2019.

As shown in Table 10, the Moran’s I indices for green innovation efficiency in various provinces of China during the period 2013–2019 are all greater than 0. The Geary’s c indices fall within the range of (0,1). With the exception of 2016, the Moran’s I indices are statistically significant at the 5% level of significance for all other years, and the Geary’s c indices pass the significance test at the 10% level for all years except 2016. These results indicate a significant positive spatial correlation in green innovation efficiency among the 30 provinces in China, suggesting a clear economic–geographical clustering phenomenon in the development of the green economy.

5.3.2. Selection of Spatial Econometric Models

The causes and interpretations of spatial effects vary among different spatial econometric models, necessitating the use of specific tests to identify the appropriate model. In the selection of spatial econometric models, the Lagrange Multiplier (LM) test is commonly employed to choose between spatial error models and spatial lag models. The results of the LM tests based on the geographic distance matrix and the economic geographic distance matrix are presented in

Table 11. Lagrange multiplier test.

Matrix Category	Test	Statistic	Df	p-Value
Geographical Distance Matrix	Spatial error: Robust Lagrange multiplier	3.885	1	0.049
	Spatial lag: Robust Lagrange multiplier	2.792	1	0.095
Economic Geography Distance Matrix	Spatial error: Robust Lagrange multiplier	3.740	1	0.053
	Spatial lag: Robust Lagrange multiplier	8.311	1	0.004

.

According to the results of the Lagrange Multiplier (LM) test, the spatial error model and spatial lag model are significant at the 10% and 1% significance levels, respectively, for the geographical distance and economic geography distance matrices. This suggests that the spatial Durbin model is suitable for regression analysis using both the geographical distance and economic geography distance matrices. To further select an appropriate spatial econometric model, the likelihood ratio test is employed for final confirmation. The results of the likelihood ratio test are presented in

Table 12. Likelihood ratio test.

Matrix Category	Assumption	Df	p-Value
Geographical Distance Matrix	SEM nested within SDM	7	0.016
	SLM nested within SDM	7	0.008
Economic Geography Distance Matrix	SEM nested within SDM	7	0.000
	SLM nested within SDM	7	0.000

.

Based on the results of the likelihood ratio test, the spatial Durbin model is found to be the most appropriate model compared with the spatial error model and spatial lag model.

5.3.3. Analysis of Spatial Regression Results for Green Innovation Efficiency

In this study, we investigated the impact of various influencing factors on green innovation efficiency using spatial error, spatial lag, and spatial Durbin models with fixed time effects with both a geographical distance matrix and an economic geographical distance matrix. The spatial Durbin models based on the geographical distance matrix and economic geographical distance matrix are referred to as Model (1) and Model (2), respectively. The spatial lag models based on the geographical distance matrix and economic geographical distance matrix are referred to as Model (3) and Model (4), respectively. The spatial error models based on the geographical distance matrix and economic geographical distance matrix are referred to as Model (5) and Model (6), respectively. The regression analyses were conducted using Stata 17 software, and the results are presented in

Table 13. Regression results of spatial econometric models for green innovation efficiency.

Variable	SDM		SLM		SEM
	Mode l (1)	Model (2)	Model (3)	Model (4)	Model (5)	Model (6)
${DLDE}_i$	0.034 (0.21)	0.581 *** (3.65)	0.207 (1.58)	0.305 ** (2.36)	0.213 (1.60)	0.232 * (1.84)
${ISO}_i$	1.050 *** (6.72)	0.378 ** (2.26)	0.892 *** (6.80)	0.687 *** (4.53)	0.902 *** (6.53)	0.924 *** (6.98)
${lnSTI}_i$	−0.095 *** (−3.76)	−0.078 *** (−3.56)	−0.064 *** (−3.07)	−0.068 *** (−3.26)	−0.066 *** (−3.06)	−0.061 *** (−2.88)
${lnDER}_i$	−0.100 *** (−6.88)	−0.116 *** (−11.25)	−0.101 *** (−9.51)	−0.108 *** (−10.79)	−0.104 *** (−9.55)	−0.107 *** (−10.60)
${lnSKD}_i$	0.051 *** (4.94)	0.033 *** (3.46)	0.047 *** (5.95)	0.044 *** (5.58)	0.046 *** (5.66)	0.047 *** (5.82)
${lnEOW}_i$	0.029 *** (4.12)	0.002 (0.38)	0.012 ** (2.16)	0.011 ** (2.15)	0.012 ** (2.09)	0.009 * (1.69)
${lnGIG}_i$	−0.319 *** (−10.08)	−0.255 *** (−8.66)	−0.309 *** (−12.17)	−0.302 *** (−11.81)	−0.310 *** (−11.96)	−0.313 *** (−12.00)
${W\mathrm{*}DLDE}_i$	−6.719 ** (−2.16)	−0.455 * (−1.91)
${W\mathrm{*}ISO}_i$	1.103 (0.38)	0.612 * (1.88)
${W\mathrm{*}lnSTI}_i$	−2.122 *** (−2.59)	0.001 (0.02)
${W\mathrm{*}lnDER}_i$	−0.126 (−0.61)	−0.008 (−0.27)
${W\mathrm{*}lnSKD}_i$	0.361 * (1.65)	0.739 *** (2.93)
${W\mathrm{*}lnEOW}_i$	0.785 *** (3.48)	−0.004 (−0.30)
${W\mathrm{*}lnGIG}_i$	−0.712 (−0.92)	−0.134 *** (−1.70)
$\rho$	−0.367 (−0.84)	−0.145 (−1.35)	−0.586 (−1.50)	0.136 ** (2.23)
$\lambda$					−0.995 ** (−2.09)	−0.163 (−1.39)
sigma2_e	0.011 *** (10.18)	0.010 *** (10.21)	0.012 *** (10.18)	0.012 *** (10.24)	0.011 *** (4.62)	0.012 *** (10.20)
$R^2$	0.923	0.940	0.916	0.924	0.915	0.916
Log-pseudo likelihood	178.245	182.582	168.750	169.570	169.586	168.068
N	210	210	210	210	210	210

Note: *, **, and *** represent the significance levels of 1%, 5%, and 10%, respectively; the values in parentheses are the Z-statistics.

.

As shown in Table 13, overall, the values of sigma2_e, R^2, and log-pseudo likelihood are relatively good in the regression results of each model. This indicates that the regression results of spatial econometric models are reliable. Although the significance levels of the rho and lambda values are relatively weak, the regression results of the influencing factors are statistically significant, suggesting that the factors influencing the green innovation efficiency of neighboring provinces have an impact on the green innovation efficiency of the focal province.

Regarding the spatial spillover effects of the influencing factors, firstly, the regression coefficients of the level of digital economy development in models (1) and (2) are −6.719 and −0.455, respectively, and they pass the significance tests at the 5% and 10% levels. This implies that the development of the digital economy in neighboring provinces leads to the outflow of green innovation resources and negatively affects the green innovation efficiency of the focal province. Secondly, the improvement of industrial structure in model (2) has a regression coefficient of 0.612, which passes the significance test at the 10% level. It can be inferred that the smaller the economic development gap and geographic distance with other provinces, the greater the positive impact of the proportion of the tertiary industry in GDP of other provinces on the green innovation efficiency of the focal province. The optimization of industrial structure in other provinces leads to spatial clustering of green innovation resources, reduces environmental pollution, and promotes the improvement of green innovation efficiency in the focal province. Thirdly, the regression coefficient of the scale of technological innovation in model (1) is −2.122, and it passes the significance test at the 1% level. This indicates that the larger the scale of technological innovation in neighboring provinces, the more severe the brain drain of green innovation R&D personnel in the focal province, which is detrimental to the improvement of green innovation efficiency. Fourthly, the regression coefficients of the knowledge dissemination level in models (1) and (2) are 0.361 and 0.739, respectively, and they pass the significance tests at the 10% and 1% levels. This suggests that the greater the similarity in economic development level and geographic distance between provinces, the larger the scale of green innovation knowledge dissemination among neighboring provinces. Consequently, there is a higher utilization of green innovation research achievements from neighboring provinces in the focal province, which contributes to the promotion of green innovation efficiency. Fifthly, the regression coefficient of the degree of openness to the outside world in model (1) is 0.785, and it passes the significance test at the 1% level. This indicates that the greater the degree of openness to the outside world in neighboring provinces, the greater the number of sales channels for green innovation products, which facilitates the sales of green innovation products in the focal province and thereby promotes the improvement of green innovation efficiency. Lastly, the regression coefficient of green innovation protection in model (2) is −0.134, and it passes the significance test at the 1% level. This implies that the improvement of social security and employment expenditure levels in neighboring provinces leads to an increase in social security and employment expenditure in the focal province, resulting in the crowding out of government investment in green innovation. This is unfavorable for the development of green innovation efficiency in the focal province.

Regarding the regression coefficients of the influencing factors, the level of digital economy development, the optimization of industrial structure, the scale of knowledge dissemination, and the degree of openness to the outside world have significant positive effects on green innovation efficiency. On the other hand, the scale of technological innovation, the degree of environmental regulation, and green innovation protection have significant negative effects on green innovation efficiency. This indicates that China’s relatively low level of technological innovation hinders the improvement of green innovation efficiency, and therefore, there is a need for further improvement in the government’s investment structure in green innovation. Furthermore, the regression results of the influencing factors are consistent with the baseline regression results, indicating the credibility of the baseline regression results in this study.

6. Discussion

In the context of high-quality development, the improvement of China’s green innovation efficiency is of paramount importance for economic transformation and global climate governance. This research aims to comprehensively evaluate China’s green innovation efficiency, study the variations in green innovation efficiency among provinces at different stages of high-quality development, explore the influence of various new factors emerging from high-quality development on green innovation efficiency, and subsequently propose targeted policy recommendations for promoting green innovation.

Regarding the measurement of green innovation efficiency, this study focuses on 30 provinces in China from 2013 to 2019. Taking the context of high-quality development into account and considering China’s strategies for “carbon neutrality” and “peak carbon emissions”, the evaluation system incorporates carbon emissions into the assessment of green innovation efficiency. A panel data approach is employed to analyze the radial and non-radial relationships between green innovation input and output. Empirical analysis is conducted using the Hybrid DEA-Tobit model and multiple spatial econometric models. The findings reveal that China’s green innovation efficiency level is relatively low but shows a gradual upward trend over time. Although this study’s green innovation evaluation system and measurement methods differ from those of Liu et al., Zhang et al. and Li et al. they all indicate a preliminary increase in green economic development across China’s provinces [40,56,57]. Compared with other research, the inclusion of carbon emissions in the evaluation system leads to a significant decline in green innovation efficiency for each province, implying that China’s economic development still heavily relies on energy consumption, posing significant challenges for establishing a robust and low-carbon circular economic system. Additionally, the study finds that provinces classified as “High” in regard to high-quality development exhibit the best performance in green innovation efficiency, indicating their effective utilization of green innovation investment and strong green innovation capacity. Conversely, provinces classified as “Medium–Low” in regard to high-quality development perform the worst in green innovation efficiency, relying more on resource-intensive and environmentally polluting economic development models. Therefore, enhancing green innovation capacity and transitioning to a more sustainable economic development model in these provinces becomes crucial to realizing China’s “peak carbon emissions” and “carbon neutrality” goals.

Regarding the analysis of factors influencing green innovation efficiency and considering the significant role of the digital economy as a new factor under high-quality development in China and its important contribution to the national economy, this study incorporates the level of digital economic development into the indicator system of factors affecting green innovation efficiency. The results reveal a significantly positive driving effect of digital economic development on green innovation efficiency, aligning with the conclusions of Zhou et al., Liu et al. and He et al., and thus affirming the reliability and robustness of our findings [58,59,60]. Beyond the digital economic development level, this study also investigates the influence of optimized industrial structure, scale of technological innovation, environmental regulation, knowledge dissemination, degree of openness, and green innovation guarantee on green innovation efficiency. Robustness tests are conducted on all the influencing factors based on benchmark regression results, indicating the reliability and credibility of our findings. Particularly, this research highlights that China’s higher education institutions generate a large number of research outcomes but with relatively low quality, which hinders the improvement of green innovation efficiency. Furthermore, the government’s inadequate investment structure in green innovation is identified as another crucial factor impeding the enhancement of green innovation efficiency.

Regarding the analysis of spatial spillover effects, this study employs spatial econometric regression analysis to examine the impact of factors influencing green innovation efficiency from neighboring provinces. The regression results are generally consistent with the baseline regression results of green innovation efficiency, enhancing the credibility of the baseline regression results in this study.

7. Conclusions, Policy Implications and Limitations

In recent years, as the Chinese economy has entered a “new normal” phase, green innovation has gradually attracted the attention of scholars as an important force driving China’s economic transformation. It is of great value to study the green innovation efficiency of various provinces in China and its influencing factors from the perspective of high-quality development. However, previous studies mainly used traditional DEA models to measure green innovation efficiency, which failed to consider both the radial and non-radial relationships between green innovation inputs and outputs. Moreover, few studies have been able to construct green innovation evaluation systems that incorporate China’s high-quality development background, leading to limited reference value for the measured green innovation efficiency in the current economic and social development of China.

This study combines multiple techniques, including the Hybrid DEA model, GMM-Tobit model, and various spatial econometric models. Based on China’s high-quality development background, a green innovation efficiency evaluation system suitable for the current national conditions is reconstructed. To a certain extent, this study accurately and reasonably explores green innovation efficiency and its influencing factors, minimizing the impact of subjective and error factors. The main conclusions of this study are as follows: (1) Provinces classified as “High” in regard to high-quality development exhibit the best performance in green innovation efficiency, while those classified as “Medium–Low” show the poorest performance. This indicates that these provinces excessively rely on an extensive economic development model and that the transformation of the economic development model is urgently needed. Only three provinces achieved effective green innovation efficiency in the period from 2013 to 2019, and the overall level of green innovation efficiency is relatively low. Although there is a gradual upward trend over time, the task of transforming the economic development model remains challenging. (2) The level of digital economic development, optimization of industrial structure, scale of knowledge dissemination, and degree of openness to foreign trade all have significant positive effects on green innovation efficiency. On the other hand, the degree of environmental regulation, scale of technological innovation, and green innovation security have significant negative effects on green innovation efficiency. The low quality of scientific research and innovation in China hinders the improvement of green innovation efficiency. (3) Under the background of high-quality development, new factors show certain spatial spillover effects on green innovation efficiency, and the green innovation efficiency of a province may be influenced by relevant factors in neighboring provinces.

Based on the above results, this study proposes several effective policy recommendations: (1) China needs to prioritize improving the green innovation efficiency of provinces classified as “Medium–Low” in regard to high-quality development, while maintaining a high-speed development of green innovation efficiency in provinces classified as “Low Level”. (2) In the context of high-quality development, China should vigorously develop the digital economy, improve the mechanisms for technological market transactions, expand foreign trade, and enhance the country’s industrial structure. It is also important to exercise prudent control over government fiscal expenditures in social welfare and environmental governance while enhancing the quality of scientific research in higher education institutions and the efficiency of research outcome commercialization. (3) China needs to promote inter-regional collaboration in green innovation, optimize channels for the circulation of green innovation resources, foster complementary advantages in green innovation among different regions, and achieve coordinated development in green innovation.

This study inevitably has certain limitations. First, evaluating green innovation efficiency requires consideration of numerous indicators related to green innovation inputs and outputs. However, in utilizing the Hybrid DEA model to evaluate green innovation efficiency, it is worth conducting further research on how to select representative indicators for green innovation inputs and outputs and enable the Hybrid DEA model to handle multiple input–output indicators simultaneously. Second, green innovation efficiency is often influenced by various factors. It is of great significance to select more representative factors influencing green innovation efficiency and comprehensively explore their impact. Finally, due to data availability constraints, this study only analyzes green innovation efficiency at the provincial level. Analysis at the micro level, such as at city and enterprise levels, would contribute to improving China’s green innovation efficiency and also have important research value. In the future, we will undertake new research to address the limitations that exist in this research.