Can R&D Intensity Reduce Carbon Emissions Intensity? Evidence from China

Abstract

Among the ways to reduce carbon emission intensity (CEI), increasing the intensity of research and development intensity (RDI) plays an important role in the process. In China, how RDI reduces CEI has attracted widespread attention. Most scholars have not considered spatial effects in the study of the correlation between RDI and CEI; therefore, this paper uses panel data of 30 Chinese provinces from 2007–2019 as a research sample to explore the spatial effects of RDI on CEI using spatial measures, analyzes the regulatory effects of the market and government in the process using the interaction effect model, and explores the role and mediating effects in the process of industrial upgrading, technological innovation and human capital effects using the mediating effect model. The empirical results illustrate that: (1) RDI and CEI have significant positive spatial autocorrelation. The spatial clustering characteristics of CEI have obvious regional differences. (2) RDI reduces the CEI of the local area while it has the same reducing effect on the CEI of the surrounding areas. The conclusion is robust. (3) The market and government play a facilitating role in RDI that affects CEI, but there are regional differences. (4) RDI can indirectly reduce CEI by promoting industrial upgrading, improving technological innovation, and increasing human capital. Finally, according to the research conclusions, the paper put forward policy suggestions: strengthen regional cooperation, guide funds into the research and development field, improve the business environment, promote technological innovation and train relevant talents. The research content and findings of this paper enrich the theories related to the influence of RDI on CEI, and have certain implications for future research on CEI based on spatial perspective.

1. Introduction

Natural disasters caused by climate change have caused great damage to human beings, and excessive carbon dioxide emissions are the main cause of climate change. China produces a large amount of CO₂ emissions and has become the world’s largest CO₂ emitter. Scholars believe that the rapid economic growth for more than 40 years is the main reason. As the world’s largest developing country, China not only needs to reduce emissions, but also to develop its economy, and it is of great significance how to do a good job to achieve a win-win situation for both the economy and emissions reduction. And CEI is the indicator that puts the relationship between the economy and carbon emission. Therefore, how to reduce CEI and achieve a win-win situation between economy and emission reduction has become one of the main tasks to achieve green development in China [1]. To reduce CEI, China has taken many actions. For example, improving the energy consumption structure, increasing the use of new energy [2], establishing a carbon trading market [3], implementing low-carbon city pilot policies [4], increasing government subsidies in the field of scientific research [5], promoting industrial upgrading [1], establishing special green funds, strengthen green credit issuance [6], guide funds to green investment and other policies and actions to reduce carbon emission intensity and achieve low-carbon transformation and development [7]. In the “2020 Global Carbon Budget Report” released by the Global Carbon Project, it is pointed out that after years of efforts, China’s carbon emissions are expected to drop by 1.7% by 2020. From 2011 to 2020, China’s CEI has been on a decreasing trend year by year, and there are regional differences. The CEI in the western region is higher than that in the eastern and central regions, and the eastern region has the lowest carbon emission intensity (see Figure 1). In recent years, in the context of China’s “double carbon” target, it is worthwhile to explore the reasons for the year-on-year decline in CEI.

Technological innovation can change the structure of energy consumption, while funding can promote technological innovation. Therefore, increased capital investment plays a critical role in reducing CEI [8]. In China, research and development investments (RDIs) are one of the main sources of funding to promote technological innovation. In 2011, to further deepen the deep integration of science and technology and finance, my country issued the “Opinions on Promoting the Integration of Science and Technology and Finance and Accelerating the Implementation of the Independent Innovation Strategy”, proposing that technological innovation requires financial support. After the release of this document, China started to focus on RDI, which is the proportion of RDI per unit of GDP and can better reflect the importance of R&D innovation in Chinese provinces. From 2011 to 2020, China’s RDI is on the rise (see Figure 2). As can be seen from Figure 2, there are significant differences in RDI in the east, middle and west of China. The RDI in the eastern region is higher than the national level, and the RDI in the central and western regions is lower than the national level. It can be seen that there are significant differences in RDI intensities in different regions. RDI can effectively promote technological progress, industrial upgrading, economic growth and reduce carbon emission levels [9,10,11,12,13]. It can be seen that the importance of RDI in China’s economic activities is self-evident, and there are differences in RDI in various regions.

From Figure 1 and Figure 2, it can be seen that there may be a connection between RDI and CEI. In recent years, there have been many studies on the influencing factors of CEI, financial geographical density and spatial structure, energy consumption structure, economic agglomeration, green credit policies, etc., [14,15,16]. And the relationship between financial development and CEI has been the hot spot of scholars’ research. Increasing financial geographical density can reduce CEI [14]. The government’s green financial policy can be distributed to each enterprise more precisely and reduce CEI from a microscopic perspective [16]. Some scholars also explore the impact of R&D funding on CEI and argue that RDIs can reduce CEI [8]. Finance and CEI have obvious spatial effects [14,17], and although some scholars have studied the relationship between finance and CEI from a spatial perspective [17], they have not explored the relationship between RDI and CEI from a spatial perspective. Few scholars have explored what role government and market play between RDI influencing CEI based on spatial perspective? Are there differences between regions and what are the intermediate mechanisms by which RDI affects CEI? To address these shortcomings, this paper will systematically analyze the relationship between RDI and CEI, the spatial and temporal heterogeneity, the moderating role and the mediating role.

The main contributions of this paper are as follows. First, this paper enriches the theory that RDI affects CEI. Times are changing, and new conclusions are obtained in the context of the new era to lay some foundation for future research. This paper explores the spatial effect of RDI on CEI from a new perspective in the new context of China’s goal of achieving “double carbon” and in the new era of giving full play to innovation-led development. Second, this paper not only complements the research on the spatial effect of RDI on CEI, but also helps Chinese provinces recognize the spillover or siphoning effect between themselves and their neighboring regions and take effective measures to reduce CEI. this paper adopts a spatial measurement approach to study in depth how RDI reduces CEI and analyzes its regional and temporal heterogeneity. Third, this paper fills the theoretical gap of regional heterogeneity in the path of RDI’s impact on CEI, and provides a theoretical basis for RDI to reduce CEI in each region. This paper explores what roles do markets and governments play in the process of RDI’s impact on CEI? Do they play the same role in different regions? The paper also delves into the mechanisms by which RDI affects CEI.

The research structure of this paper is as below. In Section 2, we mainly sort out the literature review on RDI and CEI. Methods and data are the content of Section 3, which introduces primarily and constructs the spatial econometric model, the variable selections, and descriptive statistics. Section 4 specifically discusses the results and the robustness test. The conclusions and policy recommendations of this paper are presented in Section 5.

2. Literature Review

There is evidence that technological innovation, industrial upgrading [1] and human capital can be effective in reducing CEI [18]. However, RDI is associated with technological innovation, industrial upgrading, and human capital [10,13,14,18,19]. Thus, RDI can be effective in reducing CO2 emissions [13]. We have organized and synthesized our review of the available research results in the following four areas.

2.1. RDI and Carbon Emission Intensity

RDI plays an important role in reducing CEI. Some scholars believe that R&D activities play a leading role in reducing CEI, and R&D activities can create new technologies, thereby reducing carbon emissions [13,20,21,22,23]. First, companies can reduce carbon emissions by increasing RDIs [8] and gain a competitive advantage without changing performance. To achieve the dual carbon goal, China has formulated various policies to restrain the carbon emission behavior of enterprises. Carbon reduction policies have an intuitive impact on the RDI of enterprises [24,25,26,27,28,29,30,31,32,33,34,35]. Because of the existence of cost effects, carbon reduction policies will reduce the willingness of enterprises to invest. To obtain factor resources and reduce costs, enterprises strengthen RDI and increase the level of R&D activities [24]. At the same time, companies demonstrate to consumers and investors the level of R&D activity that not only generates more profits and investment, but also reduces carbon emissions to gain a competitive advantage in the market, maintain sustainable development, and achieve a win-win situation. Second, fossil energy consumption is the main source of carbon emissions. Improving the energy structure, reducing fossil energy consumption, developing new energy, and reducing costs are the main ways to reduce CEI [26]. Enhancing RDI affects the technological absorptive capacity of energy technologies, thereby increasing energy efficiency and reducing carbon emissions [27,28], thereby reducing CEI. Third, in terms of clean energy, the government improves the RDI of nuclear energy technology and energy efficiency, strengthens RDI, promotes nuclear energy technology innovation, improves energy efficiency, reduces carbon emissions, and produces higher social benefits [29], thereby reducing CEI. In China, RDI is one of the most important ways to reduce CEI [28]. RDI can reduce CEI by increasing experimental and developmental R&D activities [30].

2.2. Mediation of RDI’s Influence on CEI

2.2.1. Industrial Upgrading

Industrial upgrading has a positive effect on reducing CEI, while RDI can promote industrial upgrading [1]. First, firms increase RDI to enhance innovation capacity and vitality, and promote the concentration of innovation factors to firms, thus promoting their transformation and upgrading. Increasing RDI can make up for the lack of relevant knowledge and capabilities and promote the emergence of new industrial specialization, which in turn changes the industrial structure and achieves industrial upgrading, for example. There is evidence that enhancing RDI can promote product diversification of firms, which in turn leads to industrial upgrading [31]. The emergence of emerging products mainly follows the trend of the times. At this stage, the world is concerned about the low-carbon transition, therefore, low-carbon products are the future trend. Second, with the development of economy and technology, people’s quality of life is gradually improving and they have new demands for products. For a better living environment, low-carbon products are the primary choice. In order to meet people’s needs in the new period, enterprises need to improve the RDI value of high technology and new products, enhance their innovation strength, achieve industrial upgrading, and have the ability and technology to produce low-carbon products, so as to maintain sustainable development of [32]. Third, scholars have pointed out that the allocation function of capital can promote changes in industrial structure, maintain the long-term development of industries, and achieve an advanced industrial structure [33,34]. In recent years, the country has strengthened RDI to achieve low-carbon transition by first realizing a part of low-carbon industries.

2.2.2. Technology Innovation

Firstly, strengthening RDI not only improves the existing energy technology and obtains energy technology innovation, but also prevents the value of energy technology from being stolen [35]. RDI is the main driving force of energy technology innovation, and by strengthening RDI, the number of energy patents can be increased, thus helping the country to achieve energy technology breakthroughs. Enhancing RDI to reduce the cost of new energy technologies and improve energy efficiency are the main ways to reduce CEI. Second, firms pay attention to strengthen RDI at an early stage and increase RDI in basic research to improve technological innovation, capture the market, and obtain future monopoly profits [36]. In China, RDI misallocation is the main obstacle to improve technological innovation [37], and strengthening RDI can solve the problem of RDI misallocation. Emerging firms have strong financing constraints, but strong innovation dynamics. And emerging companies are established to meet the trend of the times, and at this stage, most emerging companies are aiming at reducing CEI as a profitability goal. Strengthening RDI is visible to ease the financing constraint of emerging companies and have more funds to promote low-carbon technology innovation.

2.2.3. Human Capital

The improvement of human capital level is mainly reflected in the technical staff: on the one hand, the increase of quantity; on the other hand, the improvement of level. A good R&D center can create a good R&D environment and atmosphere by increasing RDI, attracting more excellent R&D personnel, expanding the number of R&D personnel, and improving R&D capabilities. Companies compensate for the low development trap of R&D by increasing R&D investment and accumulating human capital through training general workers [38]. In countries with a limited skilled workforce, bringing in international senior talent requires significant financial resources, and an increase in RDI can give just such financial support [39]. RDI has a significant impact on human capital. Enhanced RDI can achieve a favorable research environment and the nutritional supplements needed for human capital; therefore, the contribution of RDI to human capital is immeasurable [40]. Reducing CEI is not done behind closed doors; it requires exchange with international advanced levels and the introduction of international high-level talents for technical guidance and management. Increasing RDI can attract high-level R&D talents from different backgrounds around the world, forming an international R&D team and improving the overall level of the R&D team [11]. Having an advanced team is an important way to reduce CEI.

In summary, RDI has an effect on CEI, and RDI affects industrial upgrading, technological innovation, and human capital. And industrial upgrading, technological progress and human capital have effects on CEI. Based on this, the research themes of this paper are proposed: (1) RDI can reduce CEI and whether it has a spatial effect. (2) Whether geographical differences and policy effectiveness in China lead to geographical and temporal heterogeneity in the impact of RDI on CEI. (3) Whether RDI can affect CEI through industrial upgrading, technological innovation and human capital.

The research framework of this paper is shown in Figure 3.

3. Methodology and Data

3.1. Methodology

3.1.1. Spatial Correlation Test

To ensure the reasonableness of the spatial regression results, spatial autocorrelation tests need to be performed on the variables of interest before spatial regression is performed. Spatial autocorrelation is divided into two categories, global spatial autocorrelation and local spatial autocorrelation. The spatial weights are mainly divided into geographic distance weights, economic distance weights, and 0–1 distance weights. 0–1 distance is selected as the spatial weight in this study, and the Moran index I is used to measure the spatial autocorrelation.

The formula for Global Moran’s I is written as follows,

$$I=\frac{{\displaystyle {\sum }_{i=1}^n{\displaystyle {\sum }_{j=1}^n{\omega }_i{}_j(x_i-\overline{x})(x_j-\overline{x})}}}{{\displaystyle {\sum }_{i=1}^n{(x_i-\overline{x})}^2}}$$

(1)

The formula for the local Moran’s I is as follows,

$$I_i=\frac{(x_i-\overline{x})}{S^2}{\displaystyle {\sum }_{j=1}^n{\omega }_i{}_j(x_j-\overline{x})}$$

(2)

where n is the number of Chinese provinces, $x$ is the observed variable, $\omega$ is the spatial weight matrix, and $S^2$ is the sample variance. The Moran’s I takes a value between −1 and 1. A positive Moran’s I indicates positive spatial autocorrelation, meaning that high values cluster with high values, and low values cluster with low values. If Moran’s I is negative, then the spatial autocorrelation is negative, i.e., high and low values cluster together, and low and high values cluster together.

3.1.2. Spatial Econometric Model

In this paper, we focus on the spatial effect of RDI to reduce CEI, therefore, spatial econometric models should be considered. The benchmark model established in this study is as follows.

Spatial Lag Model,

$$\begin{array}{l}{\mathrm{lnCEI}}_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+\rho {\displaystyle \sum _{j\ne i}^N{\omega }_i{}_j\ln {\mathrm{CEI}}_{i,t}}+{\alpha }_2{\omega }_i{}_j\ln contro{l}_{i,t}\\ +{\mu }_i+\lambda t+{\epsilon }_{i,t},{\epsilon }_{i,t}~N(0,{\sigma }^2{I}_n)\end{array}$$

(3)

Spatial Errors Model,

$$\begin{array}{l}{\mathrm{lnCEI}}_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2{\omega }_i{}_j\ln contro{l}_{i,t}+{\mu }_i+\lambda t+{\epsilon }_{i,t},\\ {\epsilon }_{i,t}=\nu {\displaystyle \sum _{j\ne i}^N{\omega }_i{}_j{\epsilon }_{j,t}}+{\varphi }_{i,t},{\varphi }_{i,t}~N(0,{\sigma }^2{I}_n)\end{array}$$

(4)

Spatial Durbin Model,

$$\begin{array}{l}{\mathrm{lnCEI}}_{i,t}={\alpha }_0+\rho {\displaystyle \sum _{j\ne i}^N{\omega }_i{}_j\ln {\mathrm{CEI}}_{i,t}}+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2{\displaystyle \sum _{j\ne i}^N{\omega }_i{}_j\ln {\mathrm{RDI}}_{i,t}}\\ +{\alpha }_3{\omega }_i{}_j\ln contro{l}_{i,t}+{\alpha }_3{\displaystyle \sum _{j\ne i}^N{\omega }_i{}_j\ln contro{l}_{i,t}}+{\mu }_i+\lambda t+{\epsilon }_{i,t},{\epsilon }_{i,t}~N(0,{\sigma }^2{I}_n)\end{array}$$

(5)

where i is a province, t is time, CEI is carbon emission intensity, RDI is R&D intensity, and control is the control variables, including education level (EDU) and employment level (EL). $\rho$ is the spatial spillover coefficient of CEI, If $\rho$ is positive, it is a spatial spillover effect, and if $\rho$ is significantly negative, it is a spatial siphon effect. ${\alpha }_0$, ${\alpha }_1$, ${\alpha }_2$, ${\alpha }_3$, ${\alpha }_4$ are the coefficients, ${\mu }_t$ is an individual fixed effect, ${\lambda }_i$ is a time-fixed effect, ${\epsilon }_{i,t}$ is a random perturbation term, $\nu$ is the spatial residual autocorrelation coefficient. To make the data smoother and attenuate the covariance and heteroskedasticity of the model, all variables are taken as logarithms. ${\omega }_{ij}$ is the space weight matrix of order N × N. In this study, the 0–1 spatial weight matrix is used to measure the spatial spillover effect, and a geographic weight matrix is used for the robustness test.

3.1.3. Interaction Effect Models

The above analysis reflects the spatial effect of the RDI on CEI. To study the interaction role of marketization level and government support in RDI affecting CEI. The formula for the interaction effect is as follows,

$$\begin{array}{l}{\mathrm{lnCEI}}_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2\ln \left({\mathrm{RDI}}_{i,t}\times R\right)+{\alpha }_3{\omega }_i{}_j\ln contro{l}_{i,t}\\ +{\mu }_i+\lambda t+{\epsilon }_{i,t}\end{array}$$

(6)

where R represents the interaction variables, including two variables (marketization level and government support). If the α₁ and α₂ regression coefficients are significant, it indicates substantial interaction variables. There is a specific interaction between the market and the government in the process of RDI affecting CEI.

3.1.4. Mediation Effect Models

This paper will verify the mediation effect of RDI on CEI. The mediation effect model is as follows,

$${\mathrm{lnCEI}}_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2{\omega }_i{}_j\ln contro{l}_{i,t}+{\mu }_i+{\lambda }_t+{\epsilon }_{i,t}$$

(7)

$$\ln M_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2{\omega }_i{}_j\ln contro{l}_{i,t}+{\mu }_i+{\lambda }_t+{\epsilon }_{i,t}$$

(8)

$${\mathrm{lnCEI}}_{i,t}={\alpha }_0+{\alpha }_1\ln {\mathrm{RDI}}_{i,t}+{\alpha }_2\ln M_{i,t}+{\alpha }_3{\omega }_i{}_j\ln contro{l}_{i,t}+{\mu }_i+{\lambda }_t+{\epsilon }_{i,t}$$

(9)

where M_i,t represents the meditation variables, including three variables, industrial upgrading, technology innovation, and human capital. Equation (7) indicates the direct effect of RDI on the CEI, If α₁ is significant, it means that Equation (7) holds. Equation (8) shows the impact of RDI on the intermediary variable, If α₁ is significant, it means that Equation (8) holds. and Equation (9) adds the mediation variables to Equation (7) to show the impact of the RDI and the mediation variables on the CEI. If α₂ is significant, it means that Equation (9) holds. And if the three models are established simultaneously, it means that the RDI can reduce the CEI through three channels, industrial upgrading, technological innovation, and human capital.

3.2. Data and Variable Selection

In this paper, we study the impact of RDI on CEI, taking the data of 30 provinces from 2007 to 2019 as the research model. Due to the availability of the data, including Tibet, Hong Kong, Macao, and Taiwan. In 2006, energy conservation and emission reduction targets were first proposed for the first time in the 11th Five-Year Plan. There was a specific process in the policy implementation, so we chose to lag one year behind. The data used in this research were from 2007 to 2019. Data of the study were gathered from the China Statistical Yearbook, China High-tech Industry Statistical Yearbook, China Science and Technology Statistical Yearbook. Some of the data came from the wind database, the carbon dioxide emissions came from the CEADs database, and the market index came from the China sub-province market index report compiled by Fang et al.

3.2.1. Explained Variable

Carbon emission intensity (CEI). An objective and scientific measure of the carbon emission efficiency of a region can be measured by choosing carbon emissions per unit. The explanatory variable of this study, CEI, is calculated using the ratio of carbon emissions to local GDP [1].

3.2.2. Core Explanatory Variable

R&D intensity (RDI). RDI is an important element that reflects the independent innovation capability of each region and the process of building an innovative region. This study selects the ratio of R&D expenditure to GDP as the core explanatory variable, which represents the R&D intensity [13].

3.2.3. Interaction and Mediation Variables

Interaction variables. Market mechanisms and government support are the two main tools to promote economic growth. Government support can work where market mechanisms cannot reach, and market mechanisms can allocate resources efficiently. In this study, the market mechanism (MAR) and government support (GOV) serve as regulatory variables where RDC affects CEI. More of government support is indicated by more local government expenditure in the area. China is a vast country with geographic differences among regions, varying degrees of economic development, and government revenues. Quality indicators can better reflect the degree of government support, and the proportion of local government general public budget expenditure to GDP is used to represent the degree of government support. A scientific marketization index can genuinely reflect the degree and level of marketization development of a region. In this study, the China Marketization Index prepared by Fan et al. is chosen as the degree of the market mechanism. A higher level of market mechanism is indicated by a higher the index.

Mediation variables. In this study, industrial upgrading (IU), technology innovation (IN), and human capital (HC) are chosen as mediation variables. The industrial structure can affect green development, and industrial structure upgrades can promote green development. IU is the share of total tertiary sector output in GDP. The number of patents is a good indicator to measure the degree of technological innovation in a region. The degree of IN is indicated by three kinds of domestic patents granted in this study. The number of R&D full-time equivalents gives an indication of how many people a region has in R&D investment activities. The level of HC is indicated by the number of R&D full-time equivalents.

3.2.4. Control Variables

According to this study, the following indicators were selected as control variables.

Education level (EDU). The educational level of the region is represented by the mean years of education as a proxy variable. The longer the average years of schooling, the higher the level of education in the region.

Employment level (EL). The employment level of the region is measured by the urban registered unemployment rate at the year-end. The higher the level of unemployment, the lower the level of employment.

Descriptive statistics for each variable are shown in

Table 1. The statistical description of variables.

Variable	Obs.	Mean	Std. Dev.	Min.	Max.
CEI	390	0.020	0.014	0.002	0.106
RDI	390	1.531	1.090	0.210	6.310
MAR	390	6.562	1.969	2.330	11.710
GOV	390	3889.294	2674.831	213.800	17,297.850
IU	390	44.774	9.988	27.606	83.521
IN	390	40,762.100	66,932.250	222.000	527,390.000
HC	390	109,544.000	126,680.800	1262.000	803,208.000
EDU	390	8.941	0.977	6.763	12.681
EL	390	3.690	3.009	1.200	33.000

.

4. Results and Discussion

4.1. Spatial Correlation Test

4.1.1. Global Spatial Correlation Test

Using 0–1 distance as the spatial weight, the Moran’s I was used to test the global spatial correlation between RDI and CEI of each province in China from 2007 to 2019. The test results are shown in

Table 2. The global spatial autocorrelation test results.

Time	RDI				CEI
	Moran’s I		Geary’s C		Moran’s I		Geary’s C
	I	p-Value	C	p-Value	I	p-Value	C	p-Value
2007	0.178	0.014	0.664	0.037	0.307	0.002	0.678	0.013
2008	0.209	0.008	0.633	0.022	0.352	0.001	0.688	0.016
2009	0.159	0.026	0.686	0.044	0.342	0.001	0.679	0.012
2010	0.169	0.018	0.670	0.039	0.337	0.001	0.583	0.002
2011	0.203	0.010	0.636	0.022	0.240	0.006	0.742	0.062
2012	0.230	0.005	0.612	0.014	0.332	0.001	0.569	0.002
2013	0.245	0.004	0.601	0.010	0.176	0.024	0.741	0.066
2014	0.261	0.003	0.585	0.007	0.326	0.001	0.552	0.001
2015	0.274	0.002	0.571	0.005	0.355	0.000	0.540	0.001
2016	0.264	0.003	0.584	0.006	0.363	0.000	0.529	0.000
2017	0.233	0.009	0.614	0.008	0.366	0.000	0.564	0.002
2018	0.241	0.006	0.607	0.008	0.356	0.000	0.601	0.004
2019	0.284	0.002	0.563	0.004	0.400	0.000	0.568	0.002

. As can be seen from Table 2, Moran’s I and Geary’s C of RDI and CEI in China from 2007 to 2019 are significantly positive at the significant level of 10%, indicating that there is a positive spatial autocorrelation between RDI and CEI in each province of China [34,41]. RDI and CEI in each province illustrate high-high agglomeration and low-low agglomeration, and the spatial measurement method can be used in this study.

4.1.2. Local Spatial Correlation Test

To further analyze the spatial agglomeration characteristics of CEI in each province, the local spatial autocorrelation was drawn using the local Moran index. From Figure 4, we can see that most of the provinces are located in one or three quadrants, indicating that there is a clear spatial clustering characteristic among the provinces. The fact that different provinces are located in different quadrants indicates regional heterogeneity in spatial agglomeration characteristics. To dynamically analyze the spatial agglomeration characteristics of each province, we summarized this, as shown in

Table 3. Distribution of Moran scatter plots in various provinces and regions.

	2007	2011	2015	2019
Ⅰ	Ningxia, Inner Mongolia, Shanxi, Hebei, Gansu, Xinjiang, Qinghai, Liaoning, Jilin, Yunnan	Ningxia, Jilin, Inner Mongolia, Shanxi, Gansu, Xinjiang, Hebei	Ningxia, Shanxi, Xinjiang, Inner Mongolia, Gansu, Qinghai, Hebei, Heilongjiang, Liaoning	Ningxia, Inner Mongolia, Shanxi, Xinjiang, Hebei, Heilongjiang, Liaoning, Gansu, Jilin, Qinghai
Ⅱ	Heilongjiang, Shannxi, Henan, Guangxi, Sichuan, Chongqing, Beijing	Heilongjiang, Liaoning, Shannxi	Shannxi, Jilin, Henan, Shandong, Sichuan, Beijing	Shannxi, Henan, Beijing
Ⅲ	Shandong, Hunan, Hubei, Anhui, Tianjin, Jiangxi, Jiangsu, Zhejiang, Fujian, Shanghai, Guangdong, Hainan	Henan, Yunnan, Sichuan, Beijing, Tianjin, Shandong, Hunan, Hubei, Anhui, Guangdong, Guangxi, Jiangxi, Fujian, Shanghai, Hainan, Jiangsu, Zhejiang, Chongqing	Yunnan, Tianjin, Chongqing, Hubei, Hunan, Jiangsu, Zhejiang, Guangxi, Guangdong, Shanghai, Fujian, Jiangxi, Hainan, Anhui	Tianjin, Shandong, Sichuan, Yunnan, Chongqing, Hunan, Guangxi, Guangdong, Jiangsu, Anhui, Fujian, Zhejiang, Shanghai, Hainan, Jiangxi, Guizhou, Hubei
Ⅳ	Guizhou	Guizhou, Qinghai	Guizhou

.

From 2007 to 2019, more than 50% of the provinces in the western and northeastern regions are located in the first quadrant, indicating a high—high concentration of carbon emissions intensity in these provinces [42]. The main reasons are as follows. First, Inputs and outputs asymmetry. Carbon dioxide is mainly produced by the combustion of fossil energy. In contrast, western energy consumption is primarily used for power generation and heating, and carbon dioxide emissions do not match the output of the regional economy. Second, unreasonable industrial structure. In the early days, the northeast region made China’s industrial base, mainly heavy chemical industries, which had the relative advantage of abundant resources and relied primarily on resources to form its industrial structure, carbon emissions were relatively large.

From 2007 to 2019, more than 50% of the provinces located in the third quadrant were from the eastern and central regions. They are more than 50% of the provinces with low CEI showing low-low agglomeration characteristics are from the eastern and central regions. There are two main reasons for the phenomenon. First, High economic output. The economic effect can drive the environmental impact. Because of the reform and development, the eastern region introduced advanced international technology and has high economic output, so the CEI is low [43]. Second, the industry structure has lower carbon emissions. The central provinces are mainly agricultural products, with soft development of heavy industry and weak carbon displacement. With the development of the economy, which drives local technological innovation and industrial upgrading, industrial migration occurs. High-emission and high-polluting enterprises in the eastern region relocate to other regions, reducing CEI.

The provinces in the second quadrant are mainly the western provinces, the central provinces, and Beijing, indicating the CEI of these provinces in high-low agglomeration from 2007 to 2019. The main reason is the CEI of these provinces is high or low, but the CEI of neighboring provinces is the opposite. For example, Beijing’s CEI is very low, but the neighboring Hebei’s CEI is always at a high level.

From 2007 to 2015, only Guizhou Province was in the fourth quadrant, indicating that Guizhou showed low and high clustering characteristics of CEI [42]. In 2019, Guizhou Province moved to the third quadrant, and there were no provinces in the fourth quadrant [44]. Although the contribution of the three major industries to GDP in Guizhou has changed in recent years, and carbon emissions have decreased significantly, CEI is still higher than that of neighboring provinces. Until 2019 CEI development decreases sharply and shows the same clustering characteristics as the surrounding provinces [42,44].

4.1.3. Analysis of Spatial Regression Results

To better reflect that RDI can affect CEI, this paper considers non-spatial effects to obtain more robust findings, reporting the regression results of an OLS, FGLS, SYS-GMM and DIFF-GMM. The R² of OLS regression is 0.626, indicating that the regression results have a reasonable degree of fit [29]. From the OLS regression results, in China, the RDI has a significant inhibitory effect on CEI [41]. To address the problem of autocorrelation and heterovariance within the model, FGLS regression was performed and found that the coefficient of RDI was significantly negative. The IV model or the GMM model is used, and the GMM model has the following advantages compared to the IV model. One is that the GMM model is more effective than the simple model with heteroscedasticity in IV. Second, when there is no heteroscedasticity, the asymptotic of the GMM model is not worse than the IV model [40]. The GMM model is divided into the SYS-GMM model and the DIFF-GMM model. Using the GMM model based on the rationality of tool variables, according to Anderson-Hsiao’s method, introducing a longer lag period as the gmm tool variable to improve efficiency, this study choose 2 to 4 as the tool variables [45]. In this paper, we report the results of AR(2) and Hansen’s test, based on which we conclude that the instrumental variables of the SYS-GMM and DIFF-GMM models are reasonably chosen and the SYS-GMM and DIFF-GMM regression results are reliable [42,43,45,46,47]. FGLS, SYS-GMM and DIFF-GMM are consistent with the OLS regression results, and the coefficient of RDI is substantially negative. The significant inhibitory effect of RDI on CEI is further verified.

Before conducting spatial regression. First, LM tests are conducted on the variables, and the test results are shown in

Table 4. Results of the LM-test.

Test	0–1
	Statistic	p-Value
Spatial error
Moran’s I	4.030	0.000
Lagrange multiplier	270.594	0.000
Robust Lagrange multiplier	259.343	0.000
Spatial lag
Lagrange multiplier	11.611	0.001
Robust Lagrange multiplier	0.360	0.549

. From Table 4, it can be seen that the LM tests of spatial errors and spatial lags are remarkable at the level of significance of 1%, indicating that the relationship between RDI and CEI in China can be analyzed empirically using spatial econometric methods. Second, the spatial econometric model is selected based on the AIC and BIC principles. The spatial model with the smallest AIC and BIC values is selected for regression. The AIC of the SAR model is −394.783 and BIC is −367.580. The AIC of the SEM model is −344.658 and BIC is −317.455. The SDM model AIC is −401.968 and BIC is −355.335. The SDM model or SAR model is more suitable to be selected according to the AIC and BIC values. The wald test and LR test were conducted to judge further the reasonableness of selecting the SDM model. Both the wald test and LR test were significant at the significance level of 1%, indicating that the model would not degenerate into a spatial lag model and a spatial error model, so it was more logical to select the SDM model [14].

Based on the model selection above, next, the spatial relationship between RDI and CEI is examined. Under the 0–1 matrix, the coefficient of ρ is significantly positive at 0.218. The result shows that there is a significant spatial positive autocorrelation characteristic between CEI of Chinese provinces [26]. The coefficient of the core explanatory variable lnRDI is −0.129 and significant at the 10% level, indicating that RDI has a significant negative effect on CEI, and RDI can effectively reduce CEI [18]. This is similar to the findings of Churchill et al. who concluded that there is a significant relationship between RDI and CEI [19]. Strengthening RDI can attract more talent, purchase advanced equipment and machines, improve the level of technological innovation, and promote economic development while reducing CEI [8]. Furthermore, technology has spillover effects (

Table 5. The impact of RDI on CEI.

Variable					0–1
	(1)	(2)	(3)	(4)	(5)
	OLS	FGLS	SYS-GMM	DIFF-GMM	FE
lnRDI	−0.390 ***	−0.386 ***	−0.942 ***	−0.946 **	−0.129 *
	(0.0736)	(.025)	(0.213)	(0.063)	(0.0769)
lnEDU	−3.568 ***	−0.439 ***	−0.948	−1.940 ***	−1.460 ***
	(0.271)	(0.111)	(0.883)	(0.431)	(0.403)
lnEL	0.0709 *	0.014	0.635 **	0.071	0.0396
	(0.0406)	(0.012)	(0.308)	(0.103)	(0.0380)
WxlnRDI					−0.333 ***
					(0.117)
WxlnEDU					−1.199 **
					(0.508)
WxlnEL					0.0338
					(0.0691)
_cons	3.737 ***	−2.894 ***	−2.545
	(0.589)	(0.242)	(2.019)
$\rho$					0.218 ***
					(0.0529)
AR(2)			1.15	1.17
Hansen test			29.46	29.90
Wald test/R2	0.626	297.19 ***	2210.71 ***	-	0.227
N	390	390	390	390	390

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

).

4.1.4. Results of Spatial Effect Decomposition

To further analyze the spatial impact of RDI on CEI in China, this study draws on Lesage and Pace (2009) to explore the average direct effects, average indirect effects and total effects. The specific regression results are presented in

Table 6. Results of spatial effect decomposition.

Variable	Direct Effect	Indirect Effect	Total Effect
lnRDI	−0.147 *	−0.445 ***	−0.592 ***
	(0.0764)	(0.131)	(0.125)
lnEDU	−1.561 ***	−1.835 ***	−3.396 ***
	(0.376)	(0.502)	(0.428)
lnEL	0.0461	0.0543	0.100
	(0.0361)	(0.0790)	(0.0857)

Note: Standard errors in parentheses, *** p < 0.01, * p < 0.1.

. As can be seen in Table 6.

The coefficient of lnRDI is significantly negative at −0.147. It indicates that RDI has a significant inhibitory effect on CEI, and increasing RDI in the region can reduce CEI. The feedback effect can be directly reflected in the average direct effect. The strengthening of RDI in the region can reduce carbon emissions in neighboring regions, which feed back to the local area. The coefficient ratio of the average direct effect of RDI to WxlnRDI directly reflects this feedback effect, and 12.244% is the feedback effect here. This feedback effect makes the SDM model superior to the general model [26].

The average indirect effect of RDI on CEI is significantly negative, and the coefficient of 0.445 is greater than the average direct effect, indicating that the RDI of this province has a significant adverse impact on the CEI neighboring provinces. The strengthening of RDI in this province will reduce the CEI of neighboring provinces, and the degree of influence is greater than the direct effect. RDI has a strong spillover effect. This is similar to the findings of Yan et al. [13] who concluded that the effect of financial development on CEI has a significant spatial effect. First, increasing RDI can bring physical capital to local green technologies, which can improve energy efficiency and reduce CEI. Second, increasing RDI can attract more scientific and technological talents, bringing human capital to local green technologies and thus reducing CEI. Third, because of the spillover effect of economic circles and knowledge, the physical and human capital brought by increased RDC can deepen the linkages between regions. Therefore, strengthening RDI reduces the CEI of surrounding regions [6].

The total effect of RDI on CEI is significantly negative, and the coefficient of −0.592 is greater than the average direct and indirect effects, indicating that strengthening RDI can substantially reduce CEI based on the national level. Strengthening RDIs is one of the main ways to reduce CEI [40].

4.2. Analysis of Heterogeneous Results

4.2.1. Analysis of Regional Heterogeneity Results

According to the local Moran scatter plots, it is found that the CEI in China has significant regional characteristics. The eastern, central and western regions exhibit different clustering characteristics [43]. This study combines the characteristics of regional economic development. It conducts spatial regression analysis on the impact of RDI on CEI in the eastern, central and western regions, respectively. The specific regression results are presented in

Table 7. Results of SDM at the regional level.

Variable	Eastern	Central	Western
	(1)	(3)	(5)
	FE	FE	FE
lnRDI	0.0334	−0.476 ***	−0.182 *
	(0.0891)	(0.170)	(0.104)
lnEDU	−1.676 ***	−1.615	−2.043 ***
	(0.550)	(1.159)	(0.523)
lnEL	0.00232	0.0895	0.0114
	(0.0474)	(0.0732)	(0.0615)
WxlnRDI	−0.241 **	−0.284	−0.187
	(0.108)	(0.243)	(0.150)
WxlnEDU	−1.427 **	0.131	−0.174
	(0.620)	(1.275)	(0.625)
WxlnEL	−0.0539	0.0931	−0.0554
	(0.157)	(0.0905)	(0.0882)
$\rho$	0.328 ***	0.242 **	0.334 ***
	(0.0712)	(0.0997)	(0.0866)
R2	0.352	0.167	0.175
N	143	104	143

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

.

As shown in Table 7, it can be seen that ρ is significantly positive in the eastern, central and western regions, but the values are not the same. The largest value of ρ in the western, the second largest in the eastern and the smallest in the central region, indicating that the spillover effect of CEI in China has apparent differences. This has similarity with the findings of Hao et al. who concluded that regional heterogeneity is present [48]. With the most enormous spillover effect in the western, the second largest in the eastern and the smallest in the central region. Geographical features, economic development and energy distribution are the most important reasons. In China, coal resources in the western region account for 66% of the country. In contrast, the western region has a more backward economy, low RDI, immature technology for carbon reduction and emission reduction, unable to form technology and knowledge spillover, and sizeable regional carbon emissions [1]. The eastern region is located in the coastal zone, which is connected with the international, with fast economic development and developed economies, and a high level of technological innovation, forming a particular scale of the economic and technological circle. Although the CEI is lower than in other regions, the Yangtze River Delta and Pearl River Delta have developed industries and have specific carbon emissions. The central region is dominated by the agriculture and retail industry, and the development of the heavy industry is weak. The central region is close to the eastern region, and gets specific technology and knowledge spillover benefits, so the CEI shows a weaker spatial dependence.

To further analyze the spatial variability of RDI on CEI in the eastern, central and western regions. This study decomposes this spatial effect, as shown in

Table 8. Results of spatial effect decomposition.

	Variable	Direct Effect	Indirect Effect	Total Effect
Eastern	lnRDI	−0.000279	−0.303 **	−0.303 *
		(0.0919)	(0.137)	(0.177)
	lnEDU	−2.020 ***	−2.602 ***	−4.622 ***
		(0.501)	(0.628)	(0.704)
	lnEL	0.000439	−0.0573	−0.0569
		(0.0564)	(0.208)	(0.247)
Central	lnRDI	−0.525 ***	−0.479 *	−1.004 ***
		(0.177)	(0.267)	(0.342)
	lnEDU	−1.703	−0.230	−1.933 *
		(1.067)	(1.227)	(1.161)
	lnEL	0.114	0.143	0.258 *
		(0.0748)	(0.108)	(0.153)
Western	lnRDI	−0.205 *	−0.346 *	−0.551 ***
		(0.105)	(0.193)	(0.209)
	lnEDU	−2.158 ***	−1.158 *	−3.316 ***
		(0.484)	(0.609)	(0.559)
	lnEL	0.0125	−0.0643	−0.0518
		(0.0622)	(0.127)	(0.162)

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

.

The RDI in the central and western regions had a significant inhibitory effect on CEI, but had no significant effect on the eastern regions [48]. The suppression effect in the western region is lower than that in the central region. It shows that strengthening RDI in the central and western regions can significantly reduce the CEI of the whole province, but not in the eastern regions. Therefore, most regions in China can significantly reduce the CEI of their provinces by strengthening the RDI.

From the average direct effect test results in the eastern, central and western regions, it can be seen that RDI has a significant negative impact on CEI [41]. The absolute value of the coefficient is the largest in the central region, the second in the western region, and the smallest in the eastern region. The results show that in the eastern, central and western regions of China, the RDI has a significant spillover effect on the CEI of the neighboring regions, and reduces the CEI of the neighboring regions. The spillover effect of RDI to surrounding areas is the largest in the central region, followed by the western region, and the smallest in the eastern region. Increasing the RDI of Chinese provinces can boost economic development and cause technology spillovers, which can reduce not only the CEI of the region, but also the CEI of neighboring regions. Various regions in China can promote technological progress by increasing RDI, introduce international high-level soft or hard facilities, and realize industrial upgrading, to reduce CEI in the region and neighboring regions.

It can be seen from the results of the total effect test on the eastern, central and western regions that RDI has a significant negative impact on CEI [15]. The most obvious negative impact is the central region, followed by the western region, and finally the eastern region. The results showed that among the total effects of the eastern, central and western regions, the RDI in the central region had the largest negative impact on CEI, followed by the western region. Again, there are significant differences in strengthening RDI and reducing CEI in the eastern, central, and western regions of China, mainly due to the obvious differences in industrial structure, technological innovation, and economic development in the eastern, central, and western regions.

4.2.2. Analysis of Temporal Heterogeneity Results

Pilot policies have a significant impact on China [49], and RDIs are part of China’s science and technology finance. In this paper, we verify the impact of temporal heterogeneity of RDIs in China using the OPST&FIIS released in 2011 as a boundary. The specific spatial results are shown in

Table 9. Results of SDM at the temporal level.

Variable	y2007–y2011	y2012–y2014
	(1)	(2)
	RE	RE
lnRDI	0.0841 ***	−0.206 ***
	(0.0286)	(0.0293)
lnEDU	−1.828 ***	−1.536 ***
	(0.400)	(0.376)
lnEL	0.0400	0.0719 *
	(0.0381)	(0.0400)
WxlnRDI	−0.0901	−0.245 ***
	(0.0581)	(0.0578)
WxlnEDU	−1.746 ***	0.0986
	(0.480)	(0.387)
WxlnEL	0.0361	0.179 ***
	(0.0694)	(0.0693)
$\rho$	0.293 ***	0.298 ***
	(0.0500)	(0.0519)
R2	0.196	0.340
N	390	390

Note: Standard errors in parentheses, *** p < 0.01, * p < 0.1.

, and the results of the decomposition of spatial effects are shown in

Table 10. Results of spatial effect decomposition.

Time	Variable	Direct Effect	Indirect Effect	Total Effect
y2007–y2011	lnRDI	0.0800 ***	−0.0879	−0.00792
		(0.0304)	(0.0770)	(0.0909)
	lnEDU	−2.020 ***	−3.032 ***	−5.052 ***
		(0.363)	(0.446)	(0.290)
	lnEL	0.0479	0.0675	0.115
		(0.0364)	(0.0875)	(0.0968)
y2012–y2019	lnRDI	−0.229 ***	−0.413 ***	−0.642 ***
		(0.0292)	(0.0668)	(0.0737)
	lnEDU	−1.579 ***	−0.470	−2.050 ***
		(0.343)	(0.346)	(0.0786)
	lnEL	0.0919 **	0.271 ***	0.363 ***
		(0.0378)	(0.0849)	(0.0915)

Note: Standard errors in parentheses, *** p < 0.01.

.

According to Table 9, it can be seen that the coefficients of ρ are significantly positive in 2007–2011 and 2012–2019, indicating that the CEI in China shows a significant spillover effect in 2007–2011 and 2012–2019. The coefficient of lnRDI is significantly positive in 2007–2011, indicating that 2007–2011, strengthening RDI could increase CEI in China. The coefficient of lnRDI is significantly negative in 2012–2019, indicating that in 2012–2019, strengthening RDI can decrease CEI. The same phenomenon occurs for WxlnRDI, and therefore, this study further analyzes the spatial effect decomposition.

From the spatial decomposition effect, it can be seen that there is significant temporal heterogeneity in the average direct effect, average indirect effect and total effect of RDI affecting CEI. This is similar to the findings of Churchil et al. with the difference that Churchil et al.’s is that the relationship between R&D and CO2 emissions varies over time [13]. Average direct effect. From 2007 to 2011, RDI has a significant contribution to CEI, and from 2012 to 2019, RDI has a significant inhibitory effect on CEI. Average indirect effect. From 2007 to 2011, RDI has no significant impact on CEI, and from 2012 to 2019, RDI has a significant inhibitory effect on CEI. The total effect. From 2007 to 2011, RDI has no significant impact on CEI, and from 2012 to 2019, RDI has a significant inhibitory effect on CEI.

4.3. Estimated Results of Interaction Effects

This section looks at the role of the market and government in the process of RDI influencing CEI and analyzes whether the role of the market and government is the same in the eastern, central and western regions. The regression results are shown in

Table 11. Results of interaction effects.

Variable	China		Eastern		Central		Western
	MAR	GOV	MAR	GOV	MAR	GOV	MAR	GOV
lnRDI	−0.146 ***	−0.451 ***	−0.166 ***	−0.217 **	−0.135 *	−0.679 ***	−0.126 *	−0.444 ***
	(0.0396)	(0.0667)	(0.0508)	(0.104)	(0.0691)	(0.177)	(0.0742)	(0.101)
ln(RDI*MAR)	0.774 ***		0.790 ***		1.009 ***		0.650 ***
	(0.0281)		(0.0488)		(0.0415)		(0.0509)
ln(RDI*GOV)		0.172 ***		0.00813		0.193 **		0.279 ***
		(0.0394)		(0.0679)		(0.0855)		(0.0593)
lnEDU	−1.502 ***	−3.925 ***	−1.230 ***	−4.185 ***	−0.932 ***	−3.857 ***	−1.575 ***	−4.254 ***
	(0.167)	(0.292)	(0.304)	(0.440)	(0.267)	(0.789)	(0.264)	(0.376)
lnEL	0.0316	0.0855 **	0.00364	0.0380	0.0267	0.0935	0.0509	0.122 *
	(0.0246)	(0.0407)	(0.0433)	(0.0645)	(0.0356)	(0.0817)	(0.0510)	(0.0658)
_cons	0.939 ***	3.836 ***	0.297	5.049 ***	0.205	3.655 **	0.915 *	3.935 ***
	(0.347)	(0.585)	(0.619)	(0.941)	(0.574)	(1.580)	(0.548)	(0.738)
R2	0.463	0.651	0.880	0.764	0.911	0.501	0.835	0.721
N	390	390	143	143	104	104	143	143

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

.

As can be seen in Table 11, based on the country level, the coefficient of lnRDI is significantly negative and the coefficient of ln(RDI*MAR) is significantly positive, indicating that the market is having a significant positive impact in the process of RDI reducing CEI. The higher the degree of marketization, the greater the impact of increasing RDI on reducing CEI. Among the Eastern, Central and West regions, the coefficient of ln(RDI*MAR) is significantly positive, indicating that the market can promote RDI to reduce CEI in the Eastern, Central and Western regions.

The coefficient of lnRDI is significantly negative, and the coefficient of ln(RDI*GOV) is significantly positive, indicating that the government has a significant positive role in the process of RDI reducing CEI. The higher the level of government support, the greater the effect of increasing RDI on reducing CEI. In the central and western regions, the coefficient of ln(RDI*GOV) is significantly positive. It indicates that in the central and western regions, the government can promote RDI to reduce CEI. On the contrary, the coefficient of ln(RDI*GOV) in the eastern region is not significant, indicating that in the eastern region, the government has no significant influence in the process of RDI to reduce CEI.

From models China in Table 11, we can see that the degree of government support plays a greater role than the degree of marketization in reducing the CEI of China’s RDI. This is because China has a vast territory, and there are obvious differences in the economic development status of various regions. Government guidance plays an important role in the process of economic development, as does RDI. In the process of RDI reducing CEI, the market mechanism plays the most prominent role in the central region, followed by the eastern region and the western region. In contrast, the role of government support is in the western region, followed by the central region.

4.4. Estimation Results of Mediation Effects

What is the transmission mechanism of RDI to reduce CEI? Next, this study analyzes the three aspects of industrial upgrading (IU), technological innovation (IN) and human capital (HC), and the regression results are shown in

Table 12. Results of the mediation effects.

Variable	(1)	(2)	(3)	(4)	(5)	(6)
lnRDI	0.176 ***	−0.314 ***	1.234 ***	−0.129 *	2.491 ***	−0.059 **
	(0.0369)	(0.0675)	(0.132)	(0.0761)	(0.216)	(−2.22)
lnIU		−0.541 ***
		(0.102)
lnIN				−0.211 ***
				(0.0273)
lnHC						−0.0646 ***
						(0.0173)
lnEDU	1.140 ***	−2.773 ***	9.727 ***	−1.515 ***	6.160 ***	−2.984 ***
	(0.136)	(0.281)	(0.487)	(0.366)	(0.794)	(0.281)
lnEL	−0.0437 **	0.0523	−0.157 **	0.0379	0.0676	0.0825 **
	(0.0203)	(0.0398)	(0.0729)	(0.0379)	(0.119)	(0.0405)
_cons	1.302 ***	4.049 ***	−11.75 ***	1.257 **	−2.945 *	3.136 ***
	(0.295)	(0.574)	(1.057)	(0.633)	(1.725)	(0.572)
R2	0.463	0.650	0.806	0.679	0.630	0.634
N	390	390	390	390	390	390

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

.

In Table 12, the coefficients of lnRDI in model (1) are all significantly positive, indicating that RDI can promote industrial upgrading. The coefficients of industrial upgrading in model (2) are significantly negative, indicating that RDI can promote industrial upgrading to indirectly reduce CEI. This has similarity with the findings of Yan et al. who concluded that IU has a significant mediating effect in reducing CEI [14]. RDI reduces CEI through industrial upgrading. Increasing RDI can allow more funds to introduce advanced technologies and increase investment funds to develop new products. R&D capital is the knowledge input to transform a labor-intensive enterprise into a knowledge-intensive enterprise. Improving the industrial structure is the main factor to reduce CEI [50].

The coefficients of lnRDI in model (3) are all significantly positive, indicating that RDI can improve the level of IN. The coefficients of technological innovation in model (4) are significantly negative, indicating that RDI can improve IN to indirectly reduce CEI. This has similarity with the findings of Yan et al. who concluded that IN has a significant mediating effect in reducing CEI [14]. RDI reduces CEI through technological innovation. By increasing RDI, the equipment can be upgraded and advanced technologies can be introduced for technological upgrades and inventions. The innovation of environmental technology, especially energy technology, will increase the consumption of clean energy. The invention of environmental technology, especially energy technology, will increase the proportion of clean energy in energy consumption, thereby reducing CEI [51]. In resource-based areas, reduce energy dependence, reduce energy intensity, improve energy utilization, develop new energy, and reduce the cost of new energy, thereby reducing CEI.

The coefficients of lnRDI in model (5) are all significantly positive, indicating that RDI can increase human capital. The coefficient of human capital in model (6) is significantly negative, indicating that RDI can increase human capital to indirectly reduce CEI. This has similarity with the findings of Huang et al. who concluded that IN has a significant effect in reducing CEI [52].RDI reduces CEI by increasing human capital. Human capital is an indispensable factor for innovation and development. Increasing RDI can improve the salary level of scientific researchers, thereby attracting more scientific research talents and improving the overall level of human capital. Scientific and technological talents with more professional knowledge and a high level of innovation can improve the level of clean production technology in the province. The knowledge spillover effect will promote the diffusion of low-carbon technologies and improve the conversion rate of advanced green production technologies, thereby reducing CEI [52].

4.5. Robustness Test

To ensure the robustness of the conclusion of this study, different weight matrices (geographical weights) are used to verify the spatial effect of RDI on CEI, and the test results are shown in

Table 13. Results of the robustness test.

Panel A Results of SDM
Variable		China	Eastern	Central	Western	y2007–y2011	y2012–y2019
		RE	FE	FE	RE	RE	RE
lnRDI		−0.161 **	0.180 **	−0.425 **	−0.198 *	0.0781 ***	−0.114 ***
		(0.0654)	(0.0819)	(0.172)	(0.109)	(0.0278)	(0.0282)
lnEDU		−1.368 ***	−0.993 *	−1.649	−1.856 ***	−1.577 ***	−1.486 ***
		(0.396)	(0.510)	(1.196)	(0.550)	(0.398)	(0.387)
lnEL		0.0398	−0.00866	0.0867	0.0287	0.0564	0.0439
		(0.0358)	(0.0415)	(0.0733)	(0.0651)	(0.0356)	(0.0353)
WxlnRDI		−0.181 ***	−0.938 ***	−0.354	−0.462 **	0.0642	−0.421 ***
		(0.0655)	(0.200)	(0.283)	(0.195)	(0.119)	(0.121)
WxlnEDU		−1.394 ***	0.271	0.313	0.633	0.958 **	0.644
		(0.377)	(0.741)	(1.431)	(0.591)	(0.420)	(0.426)
WxlnEL		0.0424	−0.0772	0.0676	0.00988	0.193 **	−0.0339
		(0.0348)	(0.108)	(0.114)	(0.107)	(0.0931)	(0.106)
$\rho$		0.544 ***	0.423 ***	0.280 **	0.344 ***	0.755 ***	0.528 ***
		(0.0856)	(0.113)	(0.114)	(0.111)	(0.0486)	(0.0873)
R2		0.424	0.197	0.149	0.173	0.262	0.370
N		390	143	104	143	390	390
Panel B Results of spatial effect decomposition
Direct effect	lnRDI	−0.181 ***	0.115	−0.470 ***	−0.234 **	0.0927 ***	−0.132 ***
		(0.0655)	(0.0818)	(0.177)	(0.106)	(0.0331)	(0.0283)
	lnEDU	−1.394 ***	−1.023 **	−1.718	−1.882 ***	−1.619 ***	−1.507 ***
		(0.377)	(0.489)	(1.110)	(0.509)	(0.374)	(0.366)
	lnEL	0.0424	−0.00969	0.106	0.0368	0.0856**	0.0469
		(0.0348)	(0.0428)	(0.0737)	(0.0644)	(0.0353)	(0.0342)
Indirect effect	lnRDI	−1.101 ***	−1.435 ***	−0.617*	−0.769 ***	0.503	−1.005 ***
		(0.231)	(0.292)	(0.340)	(0.237)	(0.494)	(0.185)
	lnEDU	−0.347	−0.205	−0.0979	0.0156	−0.919 **	−0.275
		(0.409)	(1.076)	(1.482)	(0.532)	(0.445)	(0.394)
	lnEL	−0.0486	−0.118	0.128	0.0317	0.951 ***	−0.0221
		(0.240)	(0.190)	(0.148)	(0.155)	(0.308)	(0.221)
Total effect	lnRDI	−1.282 ***	−1.320 ***	−1.087 ***	−1.003 ***	0.595	−1.137 ***
		(0.225)	(0.310)	(0.398)	(0.222)	(0.512)	(0.186)
	lnEDU	−1.741 ***	−1.228	−1.816	−1.867 ***	−2.538 ***	−1.782 ***
		(0.176)	(1.164)	(1.326)	(0.156)	(0.239)	(0.162)
	lnEL	−0.00621	−0.128	0.234	0.0684	1.036 ***	0.0248
		(0.249)	(0.213)	(0.185)	(0.184)	(0.319)	(0.228)

Note: Standard errors in parentheses, *** p < 0.01, ** p < 0.05, * p < 0.1.

. Panel A indicates the SDM regression results, and Panel B indicates the spatial effect decomposition results. According to the results of the robustness test, we get that the conclusion of this study is robust [26]. RDI can reduce CEI in China. RDI not only significantly reduces CEI in this region, but also has some influence on CEI in neighboring provinces.

5. Conclusions and Policy Implications

This paper uses the panel data of 30 provinces in China from 2007 to 2019 as the research sample, adopts the spatial measurement method to study the impact of RDI on CEI, empirically analyzes the regional and temporal heterogeneity of the impact of RDI on China’s CEI, and explores the market and government in this process. and verify whether RDI can influence CEI through three channels of industrial upgrading, technological innovation and human capital. The results show that:

(1)

CEI has obvious spatial autocorrelation and agglomeration characteristics, and the spatial agglomeration characteristics change over time. from 2007 to 2019, most western and northeastern provinces with high CEI show high-high agglomeration characteristics. Most central and eastern provinces with low CEI exhibit low-low agglomeration characteristics. The number of provinces with low-low agglomeration increases by 2019, and provinces with low-high agglomeration achieve zero.

(2)

Strengthening RDI can reduce CEI not only in this province, but also in neighboring provinces. Enhanced RDI can reduce CEI not only in the province but also in neighboring provinces. The strengthening of RDI in this province will reduce the CEI of neighboring provinces, and the degree of influence is greater than the direct effect. RDI has a strong spillover effect.

(3)

The spatial effect of RDI in reducing CEI has significant regional and temporal heterogeneity. The RDI in the central and western regions had a significant inhibitory effect on CEI, but had no significant effect on the eastern regions. The coefficient of lnRDI is significantly positive in 2007–2011, indicating that 2007–2011, strengthening RDI could increase CEI in China. The coefficient of lnRDI is significantly negative in 2012–2019, indicating that in 2012–2019, strengthening RDI can decrease CEI.

(4)

Market and government can significantly promote RDI to reduce CEI. the higher the level of marketization and government support, the more significant the effect of enhancing RDI on reducing CEI. There is significant regional heterogeneity between market and government in RDI to reduce CEI. In the process of RDI reducing CEI, the market mechanism plays the most prominent role in the central region, followed by the eastern region and the western region. In contrast, the role of government support is in the western region, followed by the central region.

(5)

RDI can reduce CEI through three channels, industrial upgrading, technological innovation and human capital. RDI can indirectly reduce CEI by promoting industrial upgrading, enhancing technological innovation and improving human capital levels.

Based on the findings of this paper, the following policy recommendations are proposed.

CEI has a significant spatial spillover effect, and there are large differences in CEI between regions, so when the government formulates policies to reduce CEI, it should fully consider regional differences and strengthen the linkages between regions. According to the results of the above study, the CEI is higher in the western region and the northeastern region, and lower in the central region and the eastern region. The western and northeastern regions should establish long-term cooperation mechanisms with the central and eastern regions from all aspects to jointly reduce the total value of CEI in China. First, establishing technology cooperation and sharing mechanism between enterprises, so that technologically advanced enterprises can drive technologically backward enterprises to achieve the goals of improving the utilization rate of fossil energy and reducing the cost of new energy, and so on, to reduce CEI. Second, establishing a cooperation mechanism between regions with high CEI and some universities in the east with strong scientific research capability in superior disciplines. Third, giving full play to the role of the market and the government, and in the eastern region, the government should devolve its power to the greatest extent and give full play to the role of the market. In the western region, the role of the market is smaller, so the government should provide a full play to its guiding role and increase the government support in introducing capital, talents and technology to reduce CEI.

(1)

The results of this paper show that strengthening RDI can significantly reduce CEI in China. The government should formulate relevant policies to guide funds into the R&D field and build a multi-level RDIs system to reduce CEI. Because the impact of RDI on CEI is different between regions, RDI should be reasonably allocated according to the characteristics of regional financial development and industrial structure. The economic development of the western region mainly relies on coal resources. It should focus on investments in improving coal utilization, promoting energy technology progress, and reducing the cost of new energy. The economic development of the central region mainly comes from crops. It should focus on investments in upgrading industrial institutions and introducing advanced technologies from the eastern region. The eastern region is economically developed and the cradle of advanced technologies, with the geographical advantage of introducing internationally advanced technologies and high-level human investment to improve the national technology to reduce carbon emissions. The government should also set up special funds, improve the green inclusive financial system, and build a green insurance system to reduce CEI.

(2)

RDI can reduce CEI through three channels, industrial upgrading, technological innovation and human capital. First, create a better business environment and improve regional infrastructure. We will set up special funds and tax incentives, introduce domestic and foreign advanced enterprises and scientific research institutions, and increase the proportion of low-carbon emission-intensive enterprises. Promote the technical knowledge overflow of advanced enterprises and scientific research institutions, promote the process of local industrial upgrading, and then realize the reduction of CEI. Second, the government sets up tax incentives to guide enterprises to strengthen their RDI. At the same time, with the government as an intermediary, form an R&D platform of enterprises-government-universities as a system to jointly promote local technological innovation and reduce CEI. Third, the government and enterprises should establish a long-term RDI mechanism. Mainly for training and introducing high-level talents, upgrading the technical level of ordinary employees, and increasing the related training expenses for skills and technicians at all levels. Establishing a salary mechanism, improving the promotion system, and creating an excellent competitive environment, thus enhancing local human capital and reducing CEI.

There are still some deficiencies in this paper: (1) Regarding R&D concern, this paper only measures the government’s R&D input intensity. Future research can explore the impact of R&D input intensity on carbon emission intensity from multi-dimensional perspectives such as government and business. (2) In the process of China’s realization of the “dual carbon” goal, it has gradually changed from reducing carbon emissions to “dual carbon control”. In the future, the impact mechanism and path selection of R&D concern on “carbon dual control” can be further studied.