Implementation Pathways for Carbon Emission Reduction Through Environmental Regulations: Synergistic Mechanisms of Industrial Intelligence and Green Technological Innovation

Abstract

In the context of the “dual-carbon” goal to promote the green and low-carbon transformation of the economy, the mechanism of environmental regulation as a core policy tool for carbon emission reduction remains theoretically controversial. Based on this, this paper uses panel data from 30 provinces in China from 2015 to 2022 and adopts a two-way fixed-effects analysis method to examine the direction and intensity of the impact of environmental regulations on carbon emissions, introducing industrial intelligence and green technological innovation as mediating variables. Research indicates that (1) for every 1-unit increase in the intensity of environmental regulation, carbon emissions are reduced by about 0.9866 units on average, and its carbon emission reduction effect is more significant in the eastern region, where the proportion of secondary industry is medium and high, as well as in non-technology-intensive regions. (2) Industrial intelligence and green technological innovation play a partial mediating role between environmental regulations and carbon reduction. (3) After categorizing green technology innovations, it is found that environmental regulations do not significantly incentivize substantive green technology innovations, but they can contribute to carbon emission reduction by promoting the development of strategic green technology innovations. (4) The analysis of spatial effects shows that carbon emissions in China’s provinces are characterized by significant spatial agglomeration. Enforcement of environmental regulations also exerts a suppressive effect on carbon emissions in adjacent provinces, and its carbon emission reduction effect is characterized by “total effect > indirect effect > direct effect”. Compared with existing studies, this paper elucidates the transmission mechanism whereby environmental regulation achieves carbon emission reductions through industrial intelligence and green technological innovation, thereby contributing a novel analytical framework for examining regulatory impacts on carbon emissions while furnishing actionable policy implications for facilitating socioeconomic greening and low-carbon transitions.

1. Introduction

Climate change ranks among the most pressing challenges confronting sustainable development across nations worldwide, with the continuous rise in global carbon emissions exerting severe adverse impacts on the climate system. The Intergovernmental Panel on Climate Change (IPCC) emphasizes that anthropogenic carbon dioxide emissions are the primary driver of global warming. Therefore, achieving carbon neutrality, reducing greenhouse gas emissions, and increasing carbon sinks have become important strategies for the global response to climate change. In response to this challenge, countries around the world have successively introduced carbon neutrality policies to accelerate the transition to low-carbon economies. As the most populous nation across the globe, China serves as both a key contributor to global carbon emission reduction efforts and one of the world’s largest carbon-emitting countries [1]. Its carbon neutrality process has both global significance and unique challenges. China formally articulated its “dual carbon” goals in 2020, stemming from intrinsic imperatives for sustainable development and global responsibilities in advancing a shared future for mankind.

China is presently undergoing a phase characterized by rapid progress in new industrialization, digital transformation, and modernization of agricultural production. The foundation for achieving a comprehensive green transition is relatively weak. China’s resource characteristics as a “coal-rich” country have determined its “coal-based” energy consumption pattern [2], which poses a huge challenge to attaining the “dual carbon” goals. Based on this, China is adopting increasingly stringent environmental regulations to foster coordinated development across economic, social, and environmental domains [3]. Environmental regulation, as a key policy instrument for enhancing environmental quality and preventing and curbing environmental pollution, constitutes an integrated fusion of government-initiated policy measures and the constraining impact of social entities and the general public on the conduct of pollution emitters [4]. The deep integration of current environmental regulations with carbon emission reduction targets fundamentally stems from China’s top-level institutional arrangements under its “dual carbon” strategy. Following the proposal of the “dual carbon” targets in 2020, policy documents such as the “Action Plan for Carbon Peaking Before 2030” and the “Administrative Measures for Carbon Emission Trading” have explicitly elevated carbon emission reductions from a derivative objective within traditional environmental protection to the core guiding principle of the environmental governance system. Taking the national carbon emissions trading market as an example, China launched this market in 2021, incorporating key emitters in the power generation sector. Through a regulatory model where “the government sets emission quotas and enterprises adjust through market-based transactions,” high-energy-consuming enterprises are compelled to reduce emissions. In 2025, it became the world’s largest carbon market in terms of greenhouse gas coverage. This practice directly demonstrates the institutional integration of environmental regulations and carbon emission reduction targets. It should be noted that the carbon reduction effects of environmental regulations do not exist in isolation; their actual effectiveness is highly dependent on the synergistic support of industrial system transformation and technological innovation. In recent years, industrial intelligence, typified by industrial robots, has experienced explosive growth in China. As an emerging technology that integrates industrial intelligence with industrial production activities, industrial intelligence not only helps industrial enterprises complete the intelligent transformation and upgrading of their production chains [5] but also generates significant energy-saving and technological advancement effects [6]. Concurrently, green technological innovation serves as the pivotal driver of green and low-carbon economic advancement, fostering a dynamic equilibrium between economic expansion and environmental governance [7]. Nevertheless, existing research has some significant theoretical limitations. Firstly, theoretical frameworks remain fragmented, as existing literature on the interactions among environmental regulation, industrial intelligence, and green technological innovation largely stays confined to pairwise correlation analyses, without yet systematically exploring the mediating transmission mechanisms of industrial intelligence and green technological innovation in the carbon emission reduction effect of environmental regulation. Secondly, there is disagreement within the academic community regarding the effectiveness of environmental regulation policies. Some scholars have raised the “green paradox” phenomenon to question their effectiveness [8], while others believe that, as natural resources continue to be depleted, pollution emissions will automatically tend to decrease, arguing that environmental regulation is redundant [9]. Against the backdrop of this intertwining of theoretical controversy and reality, investigating how environmental regulations influence carbon reductions and analyzing their synergistic relationship with industrial intelligence and green technology innovation holds significant academic merit by addressing research gaps, while also offering practical insights for enhancing environmental governance policies.

2. Literature Review

As a policy tool for preventing and controlling environmental pollution and improving environmental quality, environmental regulation involves multiple aspects such as society, economy, and ethics [4], and is one of the research focuses within the field of climate economics. Most literature classifies environmental regulations when conducting related research. From the viewpoint of the main body of implementation, environmental regulation comprises two categories: formal regulation enforced by state actors and informal regulation driven by public or social groups [10,11,12]. The former is primarily composed of command-and-control and market-incentive environmental regulation, while the latter is mainly characterized by voluntary environmental regulation [13]. Depending on their scope of application, environmental regulations can be divided into three types: exporting country regulations, importing country regulations, and multilateral environmental regulations [14]. Different types of environmental regulation are subject to different measurement approaches. Formal environmental regulation is usually measured using performance-based indicators [15] and cost-based indicators [16,17], while informal environmental regulation is usually measured using composite indicator measures. In summary, current literature on environmental regulation is mostly based on its implementation in the main body or the scope of application of the classification, and the lack of environmental regulation based on a holistic perspective on the unified measurement of environmental regulation.

Furthermore, research on environmental regulation predominantly examines its impact on new productive forces [18], industrial structure upgrading [19], green innovation [20], and related areas, while there is still no consensus on its effectiveness in reducing carbon emissions. The Porter Hypothesis, first articulated in 1991, holds that appropriately designed environmental regulations may act as a catalyst for corporate technological innovation. Such regulatory measures enhance production efficiency and output capacity, while simultaneously generating innovation offsets that compensate for compliance expenditures [21], and achieve the “double dividend” of improved environmental quality and increased economic benefits [22,23]. Furthermore, some scholars have conducted empirical tests and concluded that enforcing environmental regulations can achieve carbon emission reductions through the optimization of energy consumption structures and technological innovation [24,25], and that their inhibitory effects increase with improvements in technological innovation and economic development [26]. However, the “green paradox”, first put forward by Sinn [27], points out that environmental regulation may result in a rebound in carbon emissions due to the strategic behavior of market players, which exacerbates global warming. When using consumption-based carbon emissions as an indicator of environmental degradation, most developed economies do not achieve real emissions reductions, which is due to the fact that developed economies pay taxes on carbon emissions while relying on low-income economies to satisfy their own demand for carbon-intensive products, which leads to global carbon emissions reductions and the process of sustainable development being impeded [28]. Concurrently, extant literature identifies a statistically significant non-linear correlation between environmental regulatory stringency and carbon emission intensity, characterized by an inverted U-curve pattern. In the early stages, environmental regulations were relatively weak, and companies seeking short-term profits may have prioritized expanding production over reducing emissions, leading to an increase in carbon emissions as the economy grew. In later stages, environmental regulations achieved a sustained decline in emissions through green finance, technological innovation, and industrial restructuring [29,30], and the impact of environmental regulations on carbon emissions showed significant regional heterogeneity [31,32]. In summary, existing literature on whether environmental regulations can reduce carbon emissions varies widely. Against this backdrop, in-depth studies on how environmental regulations affect carbon emissions offer theoretical references and practical implications for grasping the policy merits of environmental regulations and sustainable environmental policies.

Furthermore, although industrial intelligence and green technological innovation have been regarded as the key transmission carriers of the impact of environmental regulations on carbon emissions, there is still controversy regarding their specific efficacy in the transmission process. On one hand, environmental regulations can drive the upgrading of industrial intelligence through technological incentives and cost-driven measures, and with the deepening of capital investment, their promoting effect shows a marginal increasing characteristic [33]. Industrial intelligence can further curtail carbon emissions by advancing industrial structure upgrading and enhancing energy use efficiency [34,35,36], with its carbon reduction effect exhibiting a notable spatial spillover effect. However, the large-scale application of intelligent equipment may intensify the energy consumption intensity of coal-dependent countries [37], and in scenarios where environmental regulation is insufficient, the industrial intelligent transformation of the economy could potentially trigger a rise in carbon emissions levels. This means the actual efficacy of environmental regulations in achieving carbon reduction through driving industrial intelligence still needs to be verified. On the other hand, environmental regulations can be effective in promoting a green economy and encouraging green technological innovation [38]. Encompassing energy conservation, emission reduction technologies, CO₂ capture systems, and allied solutions, green technological innovation constitutes the pivotal driver for regional ecological restoration [39], which can effectively reduce carbon emissions [40,41]. Nevertheless, the stimulatory influence of environmental regulations on green technological innovation may involve temporal lags and uncertainties. Concurrently, such innovation could elevate carbon outputs through rebound effects [42], resulting in the volatility of the carbon reduction effect of environmental regulations.

The key contributions of this study are as follows: First, the frequency of terms related to “environmental protection” in local government work reports serves as a proxy variable for environmental regulation. Existing research on the measurement of environmental regulations is scattered and lacks a unified quantitative framework based on a holistic perspective, which may be one of the reasons for the divergent conclusions in studies examining the interplay between environmental regulations and carbon emissions. Compared with traditional indicators such as performance-based and cost-based metrics, the frequent use of environmental protection terminology in policy documents better reveals the determination of local governments to govern and provides a more comprehensive reflection of the intensity and full scope of China’s various forms of environmental regulation. In addition, the work report is published at the beginning of each year, which is not affected by the economic behaviors occurring in the current year, and the environmental pollution is usually published at the end of the year or at the beginning of the next year, which can better mitigate the endogenous problems [43]. Second, in light of the existing discrepancies in research conclusions regarding how environmental regulation affects carbon emissions, this paper conducts a systematic analysis of its effects on carbon emissions dynamics at the macroeconomic level. It examines the heterogeneity of the relationship between environmental regulation and carbon emissions across different regional characteristics and incorporates spatial effect analysis to explore whether the carbon reduction effects of environmental regulation exhibit spatial spillover. Third, leveraging the synergistic interactions among environmental regulation, industrial intelligence, and green technology innovation in governing carbon emissions, this study delineates their internal transmission pathways. This approach empirically tests the existence of the “environmental regulation–industrial intelligence–carbon emissions” and “environmental regulation–green technology innovation–carbon emissions” causal mechanisms. In view of this, this study measures environmental regulation from a macro perspective using panel data from 30 Chinese provinces spanning 2015 to 2022. It uses a two-way fixed-effects analysis to examine the relationship between environmental regulation and carbon emissions, and introduces industrial intelligence and green technological innovation as mediating variables. This paper explores the impact of environmental regulation on carbon emissions and its mechanism of action based on the Chinese context, aiming to provide new insights for evaluating the effectiveness of environmental regulation in carbon emission reduction, and provide theoretical references for the optimization of environmental policy instruments under the dual constraints of economic growth and climate responsibility in emerging economies.

3. Theoretical Analysis and Research Hypotheses

3.1. Environmental Regulation and Carbon Emissions

Environmental regulation has traditionally centered on regional pollution prevention and control, including emissions permits and environmental taxes, while carbon reduction policies have been specifically designed to address greenhouse gases. Driven by China’s “dual-carbon” strategy, carbon reduction targets have been deeply internalized as a core component of environmental regulation, forming a synergistic governance framework. Environmental regulation constitutes a critical component of governmental social regulation. The government realizes green and low-carbon development by virtue of multiple measures, including pollutant discharge licenses, administrative sanctions, and enterprise-targeted environmental taxes [40]. It builds a systematic carbon emission reduction governance system based on the distinctive features of policy tool diversity, implementation entity diversity, and the broad scope of application. Instruments encompass three categories: command–control, market-based incentives, and public participation mechanisms. The state implements differentiated policy tools based on the heterogeneous characteristics of different cities, such as their economic development levels, industrial technology bases, and resource endowments, which can effectively improve the compatibility between environmental regulations and local emission reduction requirements. Secondly, the main bodies of environmental regulation include the government, market players, and social forces, among which the government has established carbon emission reduction as the core objective of environmental regulation and promoted its realization through the construction of the carbon emission trading market and other policy tools. Market entities reduce emission intensity through technological innovation and energy substitution. Social forces form a public consensus on carbon emission reduction through public opinion supervision and green consumption. The three parties promote the achievement of carbon emission reduction targets through a coordinated mechanism of “policy guidance, market response, and social supervision.” Third, environmental regulations mainly target high-energy-consuming industries such as manufacturing, construction, and energy supply, as well as large, key emitting enterprises, thereby achieving greenhouse gas emission control targets at the city level. Consequently, this paper postulates the following research hypothesis:

Hypothesis H1.

Environmental regulations contribute to reducing carbon emissions.

3.2. Environmental Regulation, Industrial Intelligence, and Carbon Emissions

Industrial intelligence refers to a production method integrating advanced manufacturing, information, and artificial intelligence technologies to achieve automation, intelligence, and efficiency in industrial design, production, management, and service processes [44]. It plays an important role as a link between environmental regulations and carbon emissions. First, technological innovation in enterprises is characterized by high costs, long-term nature, and uncertainty of returns, which leads to a lack of sufficient incentives for innovation among managers [45]. While enforcing environmental regulations, the government extends fiscal incentives to enterprises for stimulating technological innovation. As the fundamental driver of industrial intelligence, technological innovation utilizes pioneering technologies including artificial intelligence, big data, and the Internet of Things. This enables the transition of industrial production from established automation systems toward elevated intelligent operations.

Secondly, as a typical representative of general-purpose technology [46], industrial intelligence development facilitates carbon abatement via industrial structure upgrading and energy conservation [47]. On the one hand, China’s industrial development level has improved rapidly since the reform and opening up, but it has also suffered from the negative impacts of extensive development, especially environmental pollution caused by the heavy chemical industry [48]. Carbon emissions mainly come from industrial production activities [49]. Digitalization and intelligence are important means to transform traditional manufacturing industries and enable them to achieve low-carbon transformation [50]. As a transformative driver of emerging leading industries, industrial intelligence facilitates the transition of traditional sectors toward cleaner production, ultimately achieving substantial carbon emission reductions [51]. On the other hand, the coordination of smart technology with air conditioning, ventilation, and lighting systems can quickly sense changes in temperature, humidity, and lighting conditions, thereby rapidly automating the adjustment of smart device operating states, reducing energy consumption, and minimizing pollution emissions. Consequently, this paper postulates the following research hypothesis:

Hypothesis H2.

Industrial intelligence plays a mediating role between environmental regulation and carbon emission reduction, i.e., environmental regulation achieves a reduction in carbon emissions by increasing the level of industrial intelligence.

3.3. Environmental Regulation, Green Technology Innovation, and Carbon Emissions

Green technological innovation is an innovative activity that avoids or reduces environmental pollution and ecological damage while following ecological laws and conserving resources [52]. Compared with general technological innovation, it focuses more on the synergistic optimization of the efficient use of resources and ecological protection while improving technological efficiency, which is the key medium between environmental regulation and carbon emission. Firstly, environmental regulation plays a significant role in promoting green technological innovation. On the one hand, by releasing green development signals, environmental regulation guides the flow of research and development resources to green technologies and realizes innovation-driven green development [53]. On the other hand, escalating pollution control expenditures compel enterprises to redirect investments toward green technology R&D, thereby enhancing resource utilization efficiency while curtailing pollutant emissions. Concurrently, the government is guiding capital, talent, and other factors toward the clean technology sector through supporting measures such as R&D subsidies and green credit policies, creating economies of scale for innovative factors, and further promoting the development of green technology innovation.

Secondly, circular economy theory holds that green technological innovation can improve corporate production efficiency, reduce energy consumption, and curb carbon emissions through improvements in product design and the application of circular thinking [54]. From an industrial production perspective, green technological innovation can improve resource recycling rates through innovative process technologies and optimized production processes, and adopt advanced circular economy models to enable effective recycling and reuse of waste, thereby reducing carbon emissions intensity in industrial production processes. From an energy perspective, green technology innovation provides key technological support for the realization of carbon emission reduction targets by enhancing energy storage and conversion efficiency and promoting the application of energy-saving technologies. Consequently, this paper postulates the following research hypothesis:

Hypothesis H3.

Green technology innovation plays a mediating role between environmental regulation and carbon emission reduction, i.e., environmental regulation realizes the reduction in carbon emission by improving the level of green technology innovation.

Furthermore, green technological innovation manifests in two distinct paradigms: substantive green technological innovation and strategic technological green innovation. The former is an innovation that meets the requirements of creativity and practicality with the core objective of solving environmental problems, and is characterized by high R&D investment, long cycle time, and high risk. The latter focuses on the practical improvement of product structure or combination, which has the advantages of a short cycle, low threshold, and quick effect [55]. Under the pressure of environmental regulation, enterprises, driven by short-term cost minimization and the urgency of compliance, tend to give priority to strategic green technological innovations, and such innovations can achieve carbon emission reductions through technological adjustments such as equipment upgrading and industrial improvement, forming a transmission chain of “regulatory pressure-strategic green technological innovation-carbon emission reduction”. In contrast, because of its long R&D cycle and high technological uncertainty, it is difficult for substantive innovations to respond to regulatory pressures and form considerable emission reduction effects in a short period of time, and therefore, it is difficult to play an intermediary role in the transmission pathway between environmental regulation and carbon emission reductions. Consequently, this paper postulates the following research hypothesis:

Hypothesis H4.

Strategic technological green innovation has a mediating effect between environmental regulation and carbon emission reduction, while substantive green technological innovation does not mediate between environmental regulation and carbon emission reduction.

3.4. Heterogeneity Analysis of Carbon Emission Reduction by Environmental Regulations

The carbon emission reduction effect of environmental regulations is not uniform across different contexts. Its effectiveness depends on regional development characteristics, industrial structure attributes, and technological capability endowments, etc.

From the perspective of regional development, there are significant differences in law enforcement capabilities and technological innovation capabilities among the eastern, central, and western regions of China. The eastern regions typically have more complete and strict environmental monitoring, law enforcement capabilities, and regulatory systems, enabling the policy signals of environmental regulations to be more effectively transmitted to enterprises and generating actual emission reduction pressure. Conversely, central and western regions exhibit deficient investment in environmental governance resources, technological capacities, and enforcement rigor. On the other hand, the eastern regions have a large number of research and development institutions and talents, with strong technological innovation capabilities, enabling them to develop and apply energy-saving and emission-reduction technologies more quickly to cope with environmental regulations. Meanwhile, the central and western regions may be relatively lagging in technology acquisition, digestion, and re-innovation capabilities, which limits the emission reduction effect achieved through the technological upgrading path of environmental regulations. Consequently, this paper postulates the following research hypothesis:

Hypothesis H5.

The carbon abatement efficacy of environmental regulations demonstrates significant regional heterogeneity. The abatement efficacy in eastern regions markedly outpaces that observed in central and western counterparts.

From the perspective of industrial structure, there are significant differences in energy consumption intensity and carbon emissions among different industries. The secondary industry, as the dominant sector for energy consumption and carbon emissions, gathers high-energy-consuming industrial clusters such as electricity, heat, and fuel. These industries have high energy consumption intensity and a high carbon emission base. The carbon emission reduction effect produced by environmental regulations through forcing production technology upgrades and optimizing energy consumption structures is more direct. While regions with a lower proportion of the secondary industry are dominated by the service sector, with lower carbon emission intensity, the marginal carbon emission reduction effect of the regulation policy is difficult to manifest. Consequently, this paper postulates the following research hypothesis:

Hypothesis H6.

The carbon emission reduction effect of environmental regulations exhibits industrial structure heterogeneity. Environmental regulations can significantly reduce carbon emissions in regions with a higher proportion of the secondary industry, while statistically insignificant carbon mitigation outcomes occur in regions exhibiting a lower proportion of the secondary industry.

From the perspective of technological endowment, the policy implementation priority and enterprise behavior logic in regions with different technological intensities differ. Non-technologically intensive areas are mainly dominated by traditional manufacturing or resource-based industries, and environmental regulations can significantly reduce carbon emissions through systematically phasing out obsolete production capacities while catalyzing structural transformation. While technologically intensive areas may have problems of weak policy implementation, local governments, in order to achieve economic growth targets, relax regulatory standards, thereby weakening the actual effectiveness of environmental regulations on carbon emission reduction. Consequently, this paper postulates the following research hypothesis:

Hypothesis H7.

The carbon emission reduction effect of environmental regulations exhibits technological intensity heterogeneity. Compared to technologically intensive areas, the carbon emission reduction effect of environmental regulations is more significant in non-technologically intensive areas.

The research flow chart is shown in Figure 1.

4. Study Design

4.1. Model Settings

To examine the mechanistic pathways through which environmental regulations influence carbon emissions in China, this paper establishes the following two-way fixed-effect model.

First, given the possible direct effects of environmental regulations on carbon emissions, this paper constructs Equation (1):

$$C{E}_{it}={\alpha }_0+{\alpha }_{1}E{R}_{it}+{\displaystyle \sum {\alpha }_m{z}_m}+{\mu }_i+\lambda i{t}_t+{\epsilon }_{it}$$

(1)

Among these, the dependent variable $C{E}_{it}$ refers to the carbon emissions of province $i$ in year $t$, the core independent variable $E{R}_{it}$ refers to the environmental regulations of province $i$ in year $t$, $z_m$ refers to a series of control variables, specifically including industrial structure, economic scale, government expenditure, energy consumption, and financial development, ${\mu }_i$ represents individual-fixed effects, ${\lambda }_t$ represents time-fixed effects, and ${\epsilon }_{it}$ represents random disturbance terms.

Second, this paper introduces industrial intelligence and green technological innovation as mediating variables to explore whether the transmission paths of “environmental regulation–industrial intelligence–carbon emissions” and “environmental regulation–green technological innovation–carbon emissions” exist. Referring to the mediation effect testing process proposed by Wen Zhonglin and Ye Baojuan [56], this paper establishes the following model based on Equation (1):

Industrial intelligence and green technology innovation were introduced as explanatory variables to establish Equation (2) to investigate the impact of environmental regulations on industrial intelligence and green technology innovation:

$$Me{d}_{it}={\beta }_0+{\beta }_{1}E{R}_{it}+{\displaystyle \sum {\beta }_m{z}_m}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(2)

Among them, $Me{d}_{it}$ is the mediating variable selected in this paper, namely industrial intelligence and green technological innovation.

Finally, environmental regulations and mediating variables are simultaneously incorporated into Equation (3) to investigate whether industrial intelligence and green technological innovation mediate the relationship between environmental regulations and carbon emissions:

$$C{E}_{it}={\gamma }_0+{\gamma }_{1}E{R}_{it}+{\gamma }_{2}Me{d}_{it}+{\displaystyle \sum \gamma z_m}+{\mu }_i+{\lambda }_t+{\epsilon }_{it}$$

(3)

The approach for testing the mediating effect is as follows: first, based on Equation (1), test whether environmental regulations have an impact on carbon emissions. If the coefficient ${\alpha }_1$ is significant, proceed to the second step; if it is not significant, stop the mediating effect test. Second, based on Equation (2), examine the role of environmental regulations in industrial intelligence and green technological innovation, obtaining the coefficient ${\beta }_1$. Third, this paper conducts a regression analysis on Equation (3). If both ${\beta }_1$ and ${\gamma }_2$ are significant, it proves that the mediating effect of industrial intelligence and green technology innovation is established.

4.2. Variable Descriptions

4.2.1. Core Independent Variable

Environmental regulations (ERs) refer to environmental standards formulated and issued by the government. They serve as a supervisory mechanism for market entities, enabling enterprises within a region to effectively implement environmental policies and reduce violations of environmental regulations [57]. There is no consistent way to measure “environmental regulation” in the existing literature. This study draws on prior research [58,59] and uses the ratio of the word count of terms associated with “environmental protection” in local government work reports relative to the total word count of the reports as a proxy for environmental regulation. The specific steps for the construction of its indicators are as follows: first, the government work reports of 30 provinces except Tibet from 2015–2022 were collected manually from the website of the Central People’s Government of the People’s Republic of China, and the government work reports of each region were processed using Python 3.8 version for the binomialization of the government work reports. Second, the frequency of words related to environmental regulation was extracted from the government work reports of each region using the crawler method of Python software. Finally, the count of environment-linked terms in government work reports was computed as a ratio to the total word count of the full text of these reports to measure environmental regulation. In addition, specific terms related to “environmental protection” are listed in detail in

Table A1. Description of terms related to “environmental protection”.

Classification	Specific Vocabulary
Basic environmental protection terminology	Environmental Protection, Environmental Protection, Pollution, Energy Consumption, Emission, Green
“Carbon” related terminology	Low Carbon, Carbon Dioxide, Emission Reduction, Air, Ecology
Pollutant and indicator terminology	Chemical Oxygen Demand, Sulfur Dioxide, PM10, PM2.5

of Appendix A for readers’ reference.

4.2.2. Dependent Variable

Carbon emissions (CEs). In this paper, carbon emissions are measured by logarithmic treatment of total CO₂ emissions.

4.2.3. Mediating Variable

(1) Industrial Intelligence (II). This paper draws on the experience and practices developed by Sun Zao et al. [60], calculating the weighting of industrial robot installations in each province based on the share of manufacturing (manufacturing sales output as a percentage of the national total). This weighting is then multiplied by the robot installation data for China’s manufacturing industry from the International Federation of Robotics database to obtain the industrial robot investment data for each province’s manufacturing sector. In addition,

Table A2. Description of data related to industrial intelligence.

Indicator Categories	Specific Indicators	Data Sources	Coverage	Explanation of Limitations
Digital technology applications	Frequency count of keywords related to “digital technology”	Annual reports of listed companies	Listed Enterprises 2001–2023	Frequency of terms related to “digital technology” is inflated
Industrial Internet Platform	Number of Industrial Internet Enterprises	Social Science Big Data Platform	Prefecture-level cities 2010–2020	Serious data missing and no provincial data
5G technology applications	Number of 5G base stations	OpenCelliD database	Prefecture-level cities 2023–2024	Significantly missing data and no provincial data
	5G Mobile Subscribers	Ministry of Industry and Information Technology	Provinces 2012–2022	Serious Data Missing
Intelligent Talent Input	Number of R&D Personnel	China Statistical Yearbook	Provinces 2011–2022	Distorted data on investment in smart talent
Smart Manufacturing Output	Number of Smart Patents	CNRDS Database	Listed Enterprises 2007–2023	Missing Patent Data for SMEs

provides other relevant data on industrial intelligence for readers’ reference.

(2) Green Technology Innovation (GTI). A patent is a core indicator that can effectively measure technological innovation [61], and the authorization process is subject to a strict novelty, inventiveness, and utility review, which can truly reflect the actual output quality and market value of technological innovation. Based on this, this paper refers to the research of scholars such as Lin Chunyan [62] and Yu Zhihan [63] and employs the number of green patents granted in each province plus 1, then applies the natural logarithm to measure green technological innovation. In addition, to capture the variations in green technological innovation, this study categorizes it: substantive green technological innovation (GTI1) is measured by taking the natural logarithm of the number of granted green invention patents plus one, and strategic green technological innovation (GTI2) is measured by taking the natural logarithm of the number of granted green utility model patents plus one [64].

4.2.4. Control Variables

This study follows the approaches of scholars including Guo Feng et al. [65], Zhang Zhengyan et al. [66] and Miao Lujun et al. [67] to account for other factors influencing carbon emissions, specifically including (1) industrial structure (IS), represented by the added value of the secondary industry; industrial upgrading can encourage enterprises to adopt more advanced production technologies and management methods, improve energy utilization efficiency, and thereby achieve the goal of reducing energy consumption and carbon emissions; (2) economic scale (ES), measured by per capita gross domestic product (GDP); as per capita GDP increases, people’s consumption demands and production activities will also increase accordingly, which results in higher energy consumption and thereby affects carbon emissions; (3) government spending (GS) is measured by general public budget expenditure; expenditure in the general public budget for energy conservation and environmental protection, such as pollution control, ecological protection, and energy conservation and emission reduction projects, can effectively reduce carbon emissions; (4) energy consumption (EC), expressed in terms of electricity consumption; currently, the majority of the world’s electricity is still generated by burning fossil fuels (such as coal, oil, and natural gas), and during the process of burning these fossil fuels, a large amount of greenhouse gases such as carbon dioxide are released; and (5) financial development (FD), gauged by the loan balance of financial institutions. It exerts a facilitative effect on advancing low-carbon economic growth and curbing carbon emissions through green financial policies, capital markets, and backing for technological innovation.

4.3. Source and Explanation of Variables

In late 2014, the National Development and Reform Commission issued the “Interim Measures for the Management of Carbon Emission Rights Trading”, marking the first step in the construction of China’s national carbon market. Therefore, this study commences its temporal coverage in 2015. In addition, due to the excessive data loss after 2022, the availability of the data became very low. Therefore, in this paper, the year 2022 is taken as the final year. Building upon this foundation, this paper conducts empirical tests using panel data from 30 provinces in China spanning 2015–2022. Furthermore, due to the extremely backward development level of the Tibet region, in order to avoid extreme data causing excessive errors in the test results, this paper specifically excludes the data of the Tibet region, so as to more thoroughly investigate how environmental regulations influence carbon emissions and the transmission paths of “environmental regulations–industrial intelligence–carbon emissions” and “environmental regulations–green technological innovation–carbon emissions”. Among them, the data utilized in this study are primarily derived from sources such as the China Statistical Yearbook, the China Energy Statistical Yearbook, and government work reports. A breakdown of the specific variables chosen, along with their corresponding data sources, is presented in

Table 1. Selection and explanation of variables.

	Variable Name	Variable Abbreviation	Units	Measurement Method	Data Source
Core Independent Variable	Environmental regulations	ER	/	Environmental regulation word frequency and/text total length	Government Work Report
Dependent Variable	Carbon emissions	CE	million tons	ln (Total carbon dioxide emissions)	China Energy Statistics Yearbook
Mediating Variable	Industrial Intelligence	II	units	ln (Number of industrial robots installed)	China Statistics Yearbook International Federation of Robotics (IFR)
	Green Technology Innovation	GTI	pieces	ln (Number of green patent licenses + 1)	CNRDS Database
Control variables	Industrial structure	IS	trillion dollars	Secondary industry added value	China Statistics Yearbook
	Economic scale	ES	trillion dollars	Per capital gross domestic product	China Statistics Yearbook
	Government spending	GS	trillion dollars	General public budget expenditure	China Statistics Yearbook
	Energy consumption	EC	trillion kWh	Electricity consumption	China Statistics Yearbook
	Financial development	FD	trillion dollars	Financial institution loan balance	China Statistics Yearbook

.

5. Empirical Analysis

5.1. Descriptive Statistics of Variables

Descriptive statistics for the variables included in this study are summarized in

Table 2. Descriptive statistics of variables.

Variable	Sample Size	Mean	Standard Deviation	Min	Max
CE	240	0.1010	0.0085	0.0796	0.1193
ER	240	0.0034	0.0009	0.0016	0.0064
II	240	8.0598	1.0948	4.8616	10.4737
GTI	240	8.3841	1.1999	5.2781	10.9333
ES	240	6.8231	3.1849	2.6165	19.0313
GS	240	0.6268	0.3266	0.1138	1.8510
IS	240	1.2499	1.1022	0.0761	5.5889
EC	240	0.2355	0.1685	0.0272	0.7870
FD	240	0.0005	0.0004	0.0000	0.0022

Source: Author’s analysis based on data collected using Stata 17.0.

. Among them, the standard deviations for II and GTI in China’s provinces are 1.0948 and 1.1999, respectively, indicating significant differences in the levels of II and GTI across regions. In addition, the standard deviation of ER is 0.0009, with a maximum value of 0.0064 and a minimum value of 0.0016. The standard deviation of CE is 0.0085, with a maximum value of 0.1193 and a minimum value of 0.0796. These figures suggest that ER and CE levels in different provinces are relatively concentrated, but there are still differences.

5.2. Unit Root Test and Multicollinearity Analysis

In order to avoid pseudo-regression, this paper conducts a unit root test for the smoothness of variables based on the LLC model to eliminate heteroskedasticity, and the results of the test are shown in

Table 3. Unit root test and multicollinearity test results for variables.

	(1)	(2)	(3)	(4)	(5)
Variables	Statistics	p-Value	Presence of Unit Root	VIF	1/VIF
CE	−7.8897	0.0000	No	/	/
ER	−13.4736	0.0000	No	/	/
II	−32.3274	0.0000	No	/	/
GTI	−16.0184	0.0000	No	/	/
ES	−37.2470	0.0000	No	1.6500	0.6060
GS	−5.9808	0.0000	No	7.2600	0.1377
IS	−11.9439	0.0000	No	8.5700	0.1167
EC	−18.2069	0.0000	No	5.7500	0.1738
FD	−6.5030	0.0000	No	5.2900	0.1891
Mean VIF	/	/	/	5.7000	/

, columns (1) to (3). The results show that the smoothness test of each variable rejects the original hypothesis of the existence of a unit root in panel data, indicating that the variables are smooth. Additionally, multicollinearity diagnostics were undertaken for all control variables, with outcomes detailed in Table 3, columns (4) to (5), which indicate that the value of variance inflation factor between each control variable is less than the critical value of 10 and the mean value of the variance inflation factor is only 5.7, indicating an absence of significant multicollinearity among control variables, thereby validating the suitability of regression analysis.

5.3. Benchmark Regression Results

This study employs a two-way fixed-effects model to empirically examine the effect of ER on CE from a full-sample perspective. The test findings are shown in

Table 4. Results of benchmark regression tests.

Variables	(1)	(2)
	CE	CE
ER	−1.0122 ***	−0.9866 ***
	(0.2654)	(0.2841)
ES		0.0001
		(0.0002)
GS		−0.0025
		(0.0045)
IS		−0.0001
		(0.0017)
EC		−0.0006
		(0.0124)
FD		2.0477
		(2.1226)
Constant	0.0988 ***	0.0975 ***
	(0.0015)	(0.0037)
Province FE	Yes	Yes
Year FE	Yes	Yes
Observations	240	240
R-squared	0.9361	0.9372
F-statistic	378.8500	294.7800

Note: *** denote significance at the 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Column (1) displays results excluding control variables, whereas column (2) incorporates ES, GS, IS, EC, and FD as control variables. As can be seen from Table 4, the coefficient of the effect of ER on the inverse of CE is significantly negative at the 1% level, indicating that ER significantly suppresses China’s total CE. Specifically, for every 1-unit increase in the intensity of ER, CE is reduced by about 0.9866 units on average, and hypothesis H1 is valid.

5.4. Robustness Test and Endogenous Processing

5.4.1. Robustness Test

(1) Replacement of Independent Variables. This study systematically analyzes subnational government work reports to delineate environmental regulation policy intent. However, the validity of the proxy variable for ER constructed by word frequency is debatable because it may be affected by the length of the text and the writing style, so this paper utilizes the logarithmic value of the investment in industrial pollution control as a proxy variable for ER to conduct a robustness test, and the results are shown in

Table 5. Robustness test results.

Variables	Replacement of Independent Variables		Exclusion of Samples from Municipalities		Exclusion of High Energy Consumption Samples		Shortening the Time Window		Tail Trimming
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)
	CE	CE	CE	CE	CE	CE	CE	CE	CE	CE
ER	−0.0270 ***	−0.0276 ***	−0.9125 ***	−0.8919 ***	−0.9344 ***	−0.8140 ***	−1.0201 **	−1.0609 **	−1.0245 ***	−1.0279 ***
	(0.0059)	(0.0058)	(0.2670)	(0.2980)	(0.2904)	(0.2974)	(0.3997)	(0.4137)	(0.2680)	(0.2870)
ES		0.0004 **		−0.0009 **		0.0007 ***		−0.0002		0.0000
		(0.0002)		(0.0004)		(0.0002)		(0.0006)		(0.0002)
GS		0.0020		0.0041		−0.0203 ***		−0.0049		−0.0050
		(0.0043)		(0.0055)		(0.0045)		(0.0099)		(0.0044)
IS		−0.0008		0.0011		−0.0014		−0.0019		0.0009
		(0.0015)		(0.0018)		(0.0025)		(0.0030)		(0.0018)
EC		0.0034		−0.0009		0.0089		0.0229		−0.0042
		(0.0129)		(0.0115)		(0.0214)		(0.0211)		(0.0128)
FD		1.0831		1.2435		3.2974 *		0.0002		2.5169
		(1.0947)		(1.2777)		(1.9285)		(0.0002)		(2.1626)
Constant	0.1817 ***	0.1765 ***	0.1178 ***	0.1169 ***	0.0983 ***	0.0978 ***	0.1010 ***	0.1052 ***	0.0989 ***	0.0997 ***
	(0.0186)	(0.0188)	(0.0011)	(0.0035)	(0.0015)	(0.0032)	(0.0021)	(0.0111)	(0.0015)	(0.0038)
Province FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	240	240	208	208	144	144	150	150	240	240
Number of id	0.9437	0.9455	0.9404	0.9435	0.9205	0.9341	0.9283	0.9300	0.9358	0.9371
F-statistic	231.1600	192.9800	184.4200	161.8000	138.8900	122.8000	243.3000	194.8100	366.2200	264.3400

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

, columns (1) to (2). After replacing the independent variables, the coefficient of the effect of ER on CE is significantly negative at the 1% level, which is consistent with the results in Table 4.

(2) Exclusion of Samples from Municipalities. Municipalities have a higher administrative level, which often gives them special advantages in terms of resource allocation and policy support, and may be more beneficial to the implementation of CE reduction. Based on this, drawing on the approach adopted by Li Ronghua et al. [68], this paper excludes the samples from municipalities directly under the central government and conducts the test, with the results shown in columns (3) to (4) of Table 5. After excluding the samples from municipalities directly under the central government, the coefficient of the effect of ER on CE is significantly positive at the 1% level, which is consistent with the findings in Table 4.

(3) Exclusion of High Energy Consumption Samples. High energy-consuming provinces have energy-intensive industries as their core pillars, with large CE bases and rigid industrial structures, and their response mechanisms to ER are significantly different from those of other provinces. Based on this, this paper calculates the total energy consumption of each province from 2012 to 2022 and selects the ten provinces with the highest total energy consumption as high energy-consuming samples to be excluded and re-examined empirically, and the results are shown in Table 5, columns (5) to (6). It can be seen from Table 5 that the inhibitory effect of ER on CE still exists significantly after the high-energy consumption sample is excluded.

(4) Shortening the Time Window. After China listed the environmental protection industry as the top of its seven strategic emerging industries, 2017 was a year of explosive growth in environmental protection policies, with various environmental protection policies being introduced and continuously improved. The period from 2015 to 2017 was a critical time for the policy to move from initial formulation to gradual refinement. The implementation effects and impacts of the policy had not yet stabilized, and the data may have been subject to significant fluctuations, which could have interfered with the empirical results. Based on this, the paper re-runs the regression test after excluding 2015–2017, and the test results are presented in Table 5, columns (7) to (8). In Table 5, the regression coefficient of ER on CE remains significantly negative after shortening the time window.

(5) Tail Trimming. To prevent extreme values of each variable from affecting the empirical results, this paper re-runs the regression after shrinking the tail of each variable by 1%, and the results are shown in columns (9) to (10) of Table 5. In Table 5, the coefficient for ER’s effect on CE remains significantly negative at the 1% significance level following variable trimming, confirming the robustness of this study’s conclusions.

5.4.2. Endogenous Processing

The preceding discussion confirms that ER can significantly promote CE reductions, but the above studies ignore the possibility of endogeneity in the model, thereby reducing the reliability of the conclusions. To guarantee the scientific validity and rationality of the research findings, this paper tackles endogeneity from the following two perspectives:

(1) Two-stage least-squares (2SLS) method. In order to reflect the heterogeneity of provincial governments’ willingness and intensity of environmental governance, this paper refers to the practice of Xu Long and other scholars [69], multiplying the total water resources of each province by the percentage of “environmental protection”-related words in the government’s work report as an instrumental variable (IV) for environmental regulation. Provinces with abundant water resources tend to pay more attention to water pollution control and adopt more stringent environmental protection measures. The model is further tested for endogeneity using the two-stage least-squares method, and the results are shown in columns (1) to (2) of

Table 6. Results of endogeneity tests.

Variables	Two-Stage Least Squares Method		GMM Dynamic Panel Estimation
	(1)	(2)	(3)
	CE	CE	CE
ER		−1.1786 ***	−0.5841 **
		(−3.3026)	(0.268)
IV	0.0003 ***
	(9.5248)
L.CE			0.8866 ***
			(0.074)
ES	−0.0001 **	0.0001	0.0001
	(−1.9966)	(0.3266)	(0.000)
GS	−0.0023 **	−0.0030	0.0005
	(−2.2536)	(−0.6964)	(0.001)
IS	0.0004	−0.0001	0.0002
	(1.2785)	(−0.0493)	(0.001)
EC	0.0018	−0.0005	−0.0000
	(0.8505)	(−0.0410)	(0.005)
FD	0.2656	2.0571	0.3633
	(1.2714)	(1.0142)	(0.383)
Province FE	Yes	Yes	Yes
Year FE	Yes	Yes	Yes
Kleibergen–Paap rk LM statistic	38.5000 ***
Kleibergen–Paap rk Wald F statistic	90.7230
AR(1)			0.006
AR(2)			0.429
Hansen			0.993
Observations	240	240
R-squared	0.3498	0.8117
F-statistic	61.3200

Note: **, and *** denote significance at the 5% and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. From the results obtained, the Kleibergen–Paap rk Wald F statistic is greater than the Stock–Yogo’s 10% level critical value and passes the weak instrumental variable test, and the Kleibergen–Paap rk LM statistic of the non-identifiability test rejects the original hypothesis at the 1% level and satisfies instrumental variable identifiability. The first-stage regression results in Column (1) show that the coefficients of IV are significantly positive at 1% level; the second-stage regression results in Column (2) show that the coefficients of ER are significantly negative at 1% level, which indicates that after considering the potential endogeneity problem, ER still significantly reduce CE, which is the same as the results compared with Table 4, and further confirms that the results of this paper are robust.

(2) GMM dynamic panel estimation. CEs tend to have a certain degree of persistence and serial correlation, and this paper is based on the SYS-GMM model to test the robustness of the previous findings. The test statistics shown in column (3) of Table 6 indicate that the instrumental variables are all valid and satisfy the conditions for the use of SYS-GMM. The regression results show that the regression coefficient of ER is significantly negative at the 5% level, indicating that after taking into account the property of serial correlation of CE (controlling for L.CE and the endogeneity it leads to), ER still contributes to the reduction in CE, which proves that the conclusions of the previous paper are robust.

5.5. Heterogeneity Test

5.5.1. Regional Heterogeneity

Different regions in China vary significantly in terms of geographical environment, policy planning, and development goals. The differences in the regions may also have an impact on the CE reduction effect of ER. Based on this, this paper divides the 30 provinces in China into three major regions, namely, East, Central, and West, according to the classification standard of China Statistical Yearbook, and analyzes them empirically. The test findings are presented in

Table 7. Regional heterogeneity regression results.

Variables	Eastern Region		Central Region		Western Region
	(1)	(2)	(3)	(4)	(5)	(6)
	CE	CE	CE	CE	CE	CE
ER	−0.4086 *	−0.5244 **	−0.7222	−0.2089	−1.3209 ***	−0.7003
	(0.2168)	(0.2426)	(0.4806)	(0.3898)	(0.4614)	(0.4796)
ES		0.0002 *		0.0008 *		−0.0000
		(0.0001)		(0.0005)		(0.0009)
GS		−0.0064 **		−0.0195 ***		0.0127
		(0.0028)		(0.0070)		(0.0099)
II		−0.0009		−0.0080 *		−0.0175 **
		(0.0007)		(0.0046)		(0.0078)
ES		0.0068		−0.0702		−0.0045
		(0.0057)		(0.0452)		(0.0206)
FD		−0.2752		42.3735 ***		26.2074 ***
		(0.5045)		(7.7011)		(9.4772)
Constant	0.0959 ***	0.0971 ***	0.1136 ***	0.1268 ***	0.1149 ***	0.1156 ***
	(0.0010)	(0.0017)	(0.0021)	(0.0059)	(0.0020)	(0.0090)
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	88	88	64	64	88	88
R-squared	0.9885	0.9915	0.9393	0.9714	0.9405	0.9555

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Table 7 reveals that ER can markedly lower CE in the eastern region, whereas in the central and western regions, ER’s impact on CE reduction is not significant, and hypothesis H5 is established.

5.5.2. Industrial Structure Heterogeneity

There are significant differences in the sensitivity of various industries to ER. Moreover, as the main source of energy consumption and CE, the proportion of the secondary industry can precisely reflect the “carbon dependence” level of the regional economy. Based on this, this paper calculates the average proportion of the secondary industry in each province from 2012 to 2022. Based on the quintile method, each province is divided into three groups: those with a relatively high proportion of the secondary industry, those in the middle, and those with a relatively low proportion. Separate empirical tests are conducted for each group. The findings are presented in

Table 8. Industrial structure heterogeneity regression results.

Variables	High Share of Secondary Sector		Medium Share of Secondary Sector		Low Share of Secondary Sector
	(1)	(2)	(3)	(4)	(5)	(6)
	CE	CE	CE	CE	CE	CE
ER	−1.4513 **	−1.2786 *	−0.9373 ***	−0.7459 **	−1.1924 ***	−0.2436
	(0.6489)	(0.7502)	(0.3572)	(0.3370)	(0.3967)	(0.2952)
ES		0.0016		−0.0012 **		−0.0000
		(0.0021)		(0.0005)		(0.0004)
GS		−0.0058		0.0099 *		−0.0439 ***
		(0.0155)		(0.0057)		(0.0082)
II		−0.0004		0.0037		−0.0011
		(0.0084)		(0.0024)		(0.0040)
ES		−0.0218		−0.0291 **		0.0325
		(0.0695)		(0.0141)		(0.0245)
FD		7.2531		1.5043		30.1630 ***
		(4.4347)		(0.9673)		(8.4725)
Constant	0.1156 ***	0.1148 ***	0.1027 ***	0.1116 ***	0.0989 ***	0.1000 ***
	(0.0025)	(0.0143)	(0.0014)	(0.0052)	(0.0016)	(0.0059)
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	48	48	144	144	48	48
R-squared	0.9151	0.9297	0.9506	0.9564	0.9019	0.9685

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. The results indicate that ER can significantly reduce CE in regions with a relatively high proportion of the secondary industry and those in the middle of the proportion. However, their CE reduction effect is not significant in regions with a relatively low proportion of the secondary industry. Hypothesis H6 is thus confirmed.

5.5.3. Technology-Intensive Heterogeneity

Furthermore, this paper refers to the approach taken by Chao Xiaojing and Zhou Wenjing [70] and divides the 30 provinces of China into technology-intensive and non-technology-intensive regions. The specific calculation formula for this division is as follows:

$$L={\displaystyle \sum _{i=1}^3{y}_i}\times i(1\le L\le 3)$$

(4)

Among them, $y_i$ represents the proportion of the total output value of the $i$ industry, and $L$ represents the degree of industrial structure upgrading used to determine whether an area is technology-intensive. If the average $L$ of a province during the sample period is greater than the average $L$ of all provinces during the sample period, it is defined as a technology-intensive area. Based on the calculation results, 10 regions, including Beijing, Tianjin, Shanxi, Liaoning, Shanghai, Jiangsu, Zhejiang, Shandong, Guangdong, and Chongqing, were classified as technology-intensive regions, while other regions were classified as non-technology-intensive regions. This paper conducts separate empirical tests on technology-intensive and non-technology-intensive areas, and

Table 9. Technology intensity heterogeneity regression results.

Variables	Non-Technology-Intensive Areas		Technology-Intensive Areas
	(1)	(2)	(3)	(4)
	CE	CE	CE	CE
ER	−0.9672 ***	−0.9312 ***	−1.4680 *	−1.4719
	(0.2674)	(0.2835)	(0.8361)	(1.0565)
ES		0.0002		0.0001
		(0.0005)		(0.0003)
GS		−0.0009		−0.0057
		(0.0051)		(0.0056)
IS		−0.0011		−0.0052
		(0.0032)		(0.0033)
EC		−0.0008		0.0388
		(0.0130)		(0.0291)
FD		3.8546		1.1070
		(2.8630)		(1.6390)
Constant	0.1179 ***	0.1176 ***	0.1001 ***	0.1005 ***
	(0.0011)	(0.0038)	(0.0031)	(0.0072)
Province FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	168	168	72	72
R-squared	0.9558	0.9565	0.9028	0.9087

Note: *, and *** denote significance at the 10% and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

shows the corresponding results. As shown in Table 9, ER significantly contributes to the reduction in CE in non-technology-intensive areas, while in technology-intensive areas, ER does not play a significant role in the reduction in CE. Hypothesis H7 holds.

5.6. Mechanism Verification

5.6.1. Industrial Intelligence

This paper adopts a two-way fixed model and introduces II as an intermediary variable based on Table 4 to explore whether the mechanism of “ER-II-CE” exists. The findings of the test are presented in

Table 10. The test results of the mediating effect of industrial intelligence.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	CE	II	CE	CE	II	CE
ER	−1.0122 ***	16.9821 *	−0.8470 ***	−0.9866 ***	19.4690 **	−0.7652 ***
	(0.2654)	(9.1762)	(0.2501)	(0.2841)	(8.6190)	(0.2596)
II			−0.0097 ***			−0.0114 ***
			(0.0028)			(0.0028)
ES				0.0001	0.0260 ***	0.0004 *
				(0.0002)	(0.0086)	(0.0002)
GS				−0.0025	−0.5810 ***	−0.0091 **
				(0.0045)	(0.1422)	(0.0041)
IS				−0.0001	0.0993	0.0010
				(0.0017)	(0.0607)	(0.0013)
EC				−0.0006	0.7729 *	0.0082
				(0.0124)	(0.4097)	(0.0113)
FD				2.0477	−56.2719	1.4076
				(2.1226)	(73.8258)	(1.3952)
Constant	0.0988 ***	6.7589 ***	0.1645 ***	0.0975 ***	6.6771 ***	0.1734 ***
	(0.0015)	(0.0440)	(0.0187)	(0.0037)	(0.1289)	(0.0196)
Bootstrap test for confidence intervals				[−0.5860, −0.0364]
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	240	240	240	240	240	240
R-squared	0.9361	0.9956	0.9429	0.9372	0.9963	0.9451

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Columns (1) to (3) in Table 10 present the test results without control variables, while columns (4) to (6) introduce ES, GS, IS, EC, and FD as control variables on this basis. Columns (2) and (5) indicate that the coefficients of ER on II are all positive, indicating that ERs have promoted the development of II. This may be due to ER policies that encourage companies to actively seek more efficient and environmentally friendly production methods. The government’s environmental requirements for companies are constantly increasing, driving enterprises to raise their investment in technological innovation. II, as an innovative means of improving production efficiency, reducing resource consumption, and reducing environmental pollution, has become an important way for companies to meet ER requirements. Furthermore, columns (3) and (6) present the joint significance test results of ER and II on carbon emissions. The regression coefficients for both are significantly negative. At the same time, from the results of columns (4) to (6) as a whole, it can be known that II is an intermediary factor between ER and CE. Furthermore, this paper uses Bootstrap to validate the results of the mechanism test by repeatedly sampling 1000 times based on fixed individuals and years. It can be seen that the 95% confidence interval is [−0.5860, −0.0364], which does not include 0, indicating that the mediating effect of II exists, i.e., the transmission path of “ER-II-CE” is valid, and Hypothesis H2 is valid.

5.6.2. Green Technology Innovation

Similarly, this study employs GTI as a mediating variable to empirically examine the transmission mechanism of “ER–GTI–CE.” The findings are presented in

Table 11. The test results of the mediating effect of green technological innovation.

Variables	(1)	(2)	(3)	(4)	(5)	(6)
	CE	GTI	CE	CE	GTI	CE
ER	−1.0122 ***	40.3362 **	−0.898 8 ***	−0.9866 ***	34.8522 *	−0.8974 ***
	(0.2654)	(19.7817)	(0.2708)	(0.2841)	(19.6879)	(0.2934)
GTI			−0.0028 **			−0.0026 **
			(0.0012)			(0.0013)
ES				0.0001	−0.0191	0.0001
				(0.0002)	(0.0171)	(0.0002)
GS				−0.0025	0.3197	−0.0017
				(0.0045)	(0.3193)	(0.0043)
IS				−0.0001	−0.0980	−0.0004
				(0.0017)	(0.1071)	(0.0016)
EC				−0.0006	0.4944	0.0007
				(0.0124)	(0.6204)	(0.0120)
FD				0.0002	−0.0153	0.0002
				(0.0002)	(0.0098)	(0.0002)
Constant	0.0988 ***	8.6846 ***	0.1232 ***	0.0975 ***	8.8752 ***	0.1202 ***
	(0.0015)	(0.1118)	(0.0110)	(0.0037)	(0.3152)	(0.0118)
Bootstrap test for confidence intervals				[−0.4568, −0.0037]
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	240	240	240	240	240	240
R-squared	0.9361	0.9871	0.9382	0.9372	0.9877	0.9389

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Columns (2) and (5) show that ER has significantly promoted the development of GTI. Columns (3) and (6) present the joint significance test results of ER and GTI on carbon emission reduction. The coefficients for both exhibit significant negativity. From the results of columns (4) to (6) as a whole, it can be concluded that GTI exerts a mediating effect on the link between ER and CE. In addition, based on a two-way fixed-effect model, Bootstrap conducted 1000 repeated samples to test this result. The results showed that the 95% confidence interval was [−0.4568, −0.0037], which did not include 0, proving that the mediating effect of GTI existed, i.e., the transmission path of “ER–GTI–CE” existed, and hypothesis H3 is valid.

Furthermore, this paper divides GTI into substantive green technological innovation (GTI1) and strategic green technological innovation (GTI2) and conducts empirical tests on each of them separately. The test findings are presented in

Table 12. Green technology innovation sub-item inspection results.

	(1)	(2)	(3)	(4)	(5)	(6)
	CE	GTI1	CE	CE	GTI2	CE
ER	−0.9866 ***	−22.8983	−1.0038 ***	−0.9866 ***	42.9392 *	−0.8978 ***
	(0.2841)	(18.2393)	(0.2863)	(0.2841)	(22.2093)	(0.2940)
GTI1			−0.0007
			(0.0011)
GTI2						−0.0021 *
						(0.0011)
ES	0.0001	−0.0064	0.0001	0.0001	−0.0206	0.0001
	(0.0002)	(0.0152)	(0.0002)	(0.0002)	(0.0199)	(0.0002)
GS	−0.0025	0.2232	−0.0024	−0.0025	0.4049	−0.0017
	(0.0045)	(0.2533)	(0.0045)	(0.0045)	(0.3746)	(0.0044)
IS	−0.0001	0.1826 *	0.0000	−0.0001	−0.1525	−0.0004
	(0.0017)	(0.0944)	(0.0017)	(0.0017)	(0.1318)	(0.0016)
EC	−0.0006	−1.1226 *	−0.0014	−0.0006	0.6186	0.0007
	(0.0124)	(0.6728)	(0.0127)	(0.0124)	(0.7149)	(0.0121)
FD	0.0002	−0.0049	0.0002	0.0002	−0.0172	0.0002
	(0.0002)	(0.0041)	(0.0002)	(0.0002)	(0.0119)	(0.0002)
Constant	0.0975 ***	8.1531 ***	0.1036 ***	0.0975 ***	8.2663 ***	0.1146 ***
	(0.0037)	(0.2695)	(0.0100)	(0.0037)	(0.3699)	(0.0098)
Bootstrap test for confidence intervals	[−0.0238, 0.1568]			[−0.4455, −0.0019]
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	240	240	240	240	240	240
R-squared	0.9372	0.9891	0.9373	0.9372	0.9839	0.9386

Note: * and *** denote significance at the 10% and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Column (2) of Table 12 reveals that the coefficient of ER’s impact on GTI1 is negative and fails to pass the 10% significance test. Therefore, the test on the “ER–GTI1–CE” transmission path is stopped. In addition, from column (5) to column (6), it can be concluded that ER can facilitate the enhancement of the level of GTI2, and the combined coefficient of the effect of ER and GTI2 on CE is significantly negative, indicating that GTI2 plays an intermediary role between ER and CE. In addition, this paper verifies the results based on Bootstrap through repeated sampling 1000 times on the basis of two-way fixation, which shows that 95% of the confidence intervals for GTI1 contain 0, while 95% of the confidence intervals for GTI2 do not contain 0. This proves that the previous conclusions are robust, and hypothesis H4 is valid.

5.7. Spatial Effects of Environmental Regulation on Carbon Emissions

5.7.1. Spatial Correlation Test

To comprehensively examine the spatial autocorrelation of China’s CE, this study computed the global Moran’s I index for China’s CE from 2015 to 2022 via the spatial adjacency matrix using Stata 17, as presented in

Table 13. The result of the spatial correlation test.

Year	Moran’s I	E(I)	SD(I)	Z-Value	p-Value
2015	0.300 ***	−0.034	0.107	3.118	0.001
2016	0.283 ***	−0.034	0.107	2.980	0.001
2017	0.269 ***	−0.034	0.107	2.852	0.002
2018	0.277 ***	−0.034	0.108	2.894	0.002
2019	0.261 ***	−0.034	0.108	2.732	0.003
2020	0.152 **	−0.034	0.101	1.846	0.032
2021	0.130 *	−0.034	0.102	1.611	0.054
2022	0.104 *	−0.034	0.103	1.344	0.089

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

. The findings indicate that the global Moran’s I indices for China’s provincial CE all exhibit significant positivity, indicating that China’s provincial CE exhibit obvious spatial clustering characteristics, i.e., high-carbon-emitting provinces tend to be adjacent to high-emission provinces, while low-carbon-emitting provinces tend to be adjacent to low-emission provinces. In addition, although the positive spatial correlation persists, its strength shows a decreasing trend year by year. Especially after 2020, the spatial correlation shows a significant weakening trend, which may be due to the differentiation characteristics of regional low-carbon transition triggered by the implementation of the “dual-carbon” policy, as well as the weakening effect of the diffusion of new energy technologies and the construction of a carbon market mechanism on the influence of traditional geographic proximity.

5.7.2. Empirical Analysis of Spatial Effects

Further, this study employs the two-way fixed effect spatial Durbin model to empirically test the link between ER and CE, with the findings presented in

Table 14. Spatial Durbin model regression results.

	(1)	(2)	(3)	(4)	(5)
Variables	Carbon Emissions	Spatial Lag Term	Direct Effect	Indirect Effect	Total Effect
ER	−0.5896 ***	−1.1176 **	−0.7947 ***	−2.5703 ***	−3.3650 ***
	(0.2221)	(0.4438)	(−3.1724)	(−2.9372)	(−3.3058)
ES	−0.0002	−0.0005	−0.0003	−0.0011	−0.0014
	(0.0002)	(0.0005)	(−1.1963)	(−1.1652)	(−1.2295)
GS	−0.0008	−0.0147 *	−0.0024	−0.0252	−0.0277
	(0.0041)	(0.0089)	(−0.5075)	(−1.3991)	(−1.2726)
II	−0.0000	0.0067 *	0.0009	0.0120 *	0.0129
	(0.0016)	(0.0034)	(0.4758)	(1.6980)	(1.4972)
ES	−0.0044	−0.0171	−0.0080	−0.0383	−0.0463
	(0.0100)	(0.0230)	(−0.6254)	(−0.8043)	(−0.7942)
FD	1.5075	1.2866	1.8551 *	3.8734 *	5.7285 *
	(0.9518)	(1.0425)	(1.7939)	(1.6862)	(1.9178)
rho	0.4955 ***
	(0.0730)
sigma2_e	0.0000 ***
	(0.0000)
Observations	240	240	240	240	240
R-squared	0.0450	0.0450	0.401	0.401	0.401
Number of id	30	30	30	30	30

Note: *, **, and *** denote significance at the 10%, 5%, and 1% significance levels, respectively; heteroskedasticity robust standard errors are in parentheses.

. Column (1) shows that the coefficient of ER on CE exhibits a significantly negative value, consistent with previous empirical results, and column (2) indicates that China’s implementation of ER has the same inhibitory effect on CE of neighboring provinces. Further, columns (3) to (5) reflect the direct, indirect, and overall effects that ER exerts on CE, and the coefficients of ER all demonstrate significant negativity at the 1% significance level, and their CE reduction effects are characterized by “total effect > indirect effect > direct effect”. Upon analyzing the reasons, first, the local regulation faces short-term compliance cost pressure and innovation compensation effect time lag, and administrative boundary limitations make it difficult to fully cover the trans-regional pollution problem, so the direct effect is relatively weak; second, the ER in the neighboring regions through the constraints on the transfer of pollution, drive the technological spillover and strengthen the synergy of policies, the formation of trans-regional suppression of the local CE, and the breakthrough of a single regional regulation limitations to form a trans-regional The ER of neighboring regions can inhibit local CE across regions by restraining pollution transfer, driving technology spillover and strengthening policy synergy, thus breaking through the limitations of single-region regulation and forming cross-regional emission reduction synergy. From the above analysis, it is evident that the suppression of CE by ER in the spatial dimension is multi-dimensional and synergistic, which provides important empirical evidence and theoretical support for the development of a cross-regional environmental governance system and the optimization of environmental regulation policies.

6. Research Conclusions and Policy Recommendations

This paper relies on panel data spanning 2015–2022 from 30 Chinese provinces to empirically examine the relationship between ER and CE. Robustness tests, including the exclusion of municipal-level samples and the instrumental variables method, validate the validity of the findings of this study. The study further explores whether transmission pathways exist between “ER–II–CE” and “ER–GTI–CE”. The key conclusions derived from this study can be summarized as follows:

First, based on the full-sample perspective, ER contributes to the reduction of CE.

Second, from the heterogeneity perspective, the CE mitigation effect of ER is more pronounced in the eastern region, regions with a high share of secondary industry, regions with a medium share of secondary industry, and non-technology-intensive regions.

Third, II and GTI play a mediating role between ER and CE. In addition, ER does not significantly incentivize GTI1, but they can promote CE reduction by facilitating the development of GTI2.

Fourth, CEs in China’s provinces show obvious spatial clustering characteristics, and the implementation of ER also has an inhibitory effect on CE in neighboring provinces.

To sum up, this study proposes the following policy recommendations:

First, policymakers should strengthen the intensity of environmental regulations and implement differentiated innovation strategies. On the one hand, they should improve diversified regulatory mechanisms and constraint systems: utilize techniques including big data and the IoT to achieve precise and real-time monitoring of emissions, thereby compelling firms to pursue green innovation; refine environmental standards and penalty regulations; significantly increase the cost of violations (such as setting strict emission limits for high-polluting enterprises, imposing heavy fines, and suspending production for rectification); and encourage public lawsuits, forming a synergy between legal and social supervision. On the other hand, they should implement regional differentiation policies: in regions with weaker environmental regulations, prioritize subsidies from higher-level governments to improve real-time carbon emission monitoring systems for high-energy-consuming enterprises and strengthen penalties for enterprises that falsify data or fail to implement monitoring requirements. In regions with stronger environmental regulations, they can rely on mature monitoring and supervision systems to establish cross-regional environmental governance collaboration platforms and share monitoring technologies and emission reduction experiences with surrounding regions with weaker regulations.

Second, with industrial intelligence as the core, policymakers should promote the achievement of the dual carbon goals. For regions with relatively low levels of industrial intelligence, policy support should be strengthened and special support funds established, with a focus on supporting the innovative application of key technologies such as artificial intelligence, the Internet of Things, and big data in areas such as green manufacturing and energy optimization. In regions with relatively high levels of industrial intelligence, enterprises should be encouraged to adopt intelligent means to upgrade and transform existing production equipment, thereby reducing energy waste. Additionally, leveraging the established foundation of intelligent industries, cross-regional technology-sharing platforms can be established to provide standardized intelligent emissions reduction solutions and technical training to regions with lower levels of industrial intelligence. Strengthening industry–academia–research collaboration can further accelerate the commercialization of intelligent technology outcomes, cultivate green intelligent demonstration enterprises, and comprehensively enhance the green development level of the industrial sector, thereby injecting strong momentum into the achievement of the dual carbon goals.

Third, we recommend establishing differentiated incentive mechanisms and making categorized breakthroughs during patent prosecution for green utility models and inventions. In response to the differentiated impacts of environmental regulations on the two types of green technology patents, it is recommended that differentiated policies be implemented: for green utility model patents, incorporate them into local government environmental performance evaluation systems and expand the scope of technology promotion through policies such as R&D expense deductions and preferential interest rates for green loans. For green invention patents, we recommend leveraging environmental regulatory tools and carbon quota conversion incentives to synergistically drive innovation and convert the R&D costs of green invention patents into quantifiable carbon asset returns. Additionally, patent review standards should be improved to mandate that high-emission enterprises allocate a certain proportion of their R&D investments to original green technology research and development.

Fourth, a cross-provincial and regional coordinated emission reduction system ought to be established to unleash the spatial impact of carbon emission reduction. A centrally coordinated cross-regional environmental regulation coordination mechanism should be established to clarify the weight of each province’s responsibility for emission reductions and to set up a platform for green technology and information sharing. In addition, a system of ecological compensation and benefit balancing should be established, with central financial transfers to provinces that have made outstanding contributions to emission reduction, tax incentives for cross-regional pollution control projects, and the mechanism of “cost-sharing and benefit-sharing” to stimulate the power of coordinated emission reduction in all provinces, so as to form a nationwide spatial synergy for carbon emission reduction.