Threshold Effect of Environmental Decentralization on Environmental Regulation and Carbon Emissions

Abstract

Under the “dual carbon” targets, the influence of environmental regulation on carbon emissions is critical, and the moderating role of environmental decentralization should not be overlooked. Using provincial panel data from China, this study builds a Panel Smooth Transition Regression Model (PSTR) with environmental decentralization as the threshold variable to examine the nonlinear relationship between environmental regulation and carbon emissions. The study finds that when environmental decentralization is below the threshold, raising the intensity of environmental regulation leads to a significant reduction in carbon emissions; however, once decentralization surpasses the threshold, strengthened environmental regulation may result in a rise in carbon emissions. Three subcategories of decentralization exhibit similar threshold effects, but their direct emission reduction effects are heterogeneous. This research offers empirical evidence supporting the optimization of the distribution of environmental responsibilities across central and local governments, as well as for formulating regionally differentiated emission reduction policies.

1. Introduction

With global climate change intensifying [1], China, the world’s largest carbon emitter, clearly set the “carbon peak and carbon neutrality” (referred to as the “dual carbon” goals) target in 2020. The report from the 20th National Congress of the Communist Party further emphasizes “promoting green development and fostering harmony between humanity and nature”, and it explicitly calls for “improving the regulation of energy consumption and intensity, and transitioning toward a dual control framework for carbon emissions”. Achieving this target relies on a scientifically effective environmental governance system, and environmental regulation, as the core policy tool for government intervention in carbon emissions [2,3]. Since the 19th National Congress, China has continued to advance ecological civilization reform, clearly proposing to “Build a modern environmental governance framework” and the “14th Five-Year Plan” demands “optimization of the central–local environmental responsibility division” and the strengthening of local governments’ responsibilities.

Environmental regulation serves as the direct tool for environmental protection agencies to manage environmental governance and control carbon emissions. Environmental decentralization refers to the institutional arrangement for implementing these regulations. For local governments to effectively engage in environmental governance and reduce carbon emissions, it is essential to decentralize governance powers based on specific responsibilities, enabling them to effectively enforce and utilize environmental regulations. Therefore, the environmental decentralization system is a key institutional background for understanding the effectiveness of environmental regulation in China [4,5,6]. Environmental decentralization refers to the dynamic allocation of management responsibilities of the environment across different levels of government. It covers the decentralization of environmental administrative powers (such as policy formulation and project approval), monitoring rights (data collection and information disclosure), and supervisory rights (law enforcement supervision and penalties) [7]. While decentralization reform provides local governments with greater flexibility, it may also trigger a ‘race-to-the-bottom’ competition, or “symbolic governance” in the implementation of environmental regulations, driven by pressures for economic growth, performance assessments, and regional development disparities [8,9].

Existing studies have thoroughly investigated how environmental regulation affects carbon emissions, but there are several limitations. First, few studies integrate both environmental regulation and environmental decentralization into a single framework to investigate their combined impact on carbon emissions. Second, most literature analyzes the emission reduction effects of environmental regulation under linear assumptions, ignoring the dynamic nonlinear characteristics of policy effects in a decentralized system. Third, as an essential institutional arrangement in China’s environmental governance system, the multidimensional nature of environmental decentralization (such as administrative, monitoring, and supervision decentralization) is not fully explored in its differentiated impact on carbon emissions. To address these issues, this paper focuses on environmental decentralization and develops a Panel Smooth Transition Model (PSTR) to explore the nonlinear effects and regional variations in how environmental regulation influences carbon emissions. The aim is to offer theoretical insights and policy recommendations for improving the environmental governance system in line with the “dual carbon” goals.

The marginal contributions of this paper are manifested in three aspects: Firstly, it incorporates both environmental decentralization and environmental regulation into the research framework. And it examines their impacts on carbon emissions, pioneering the investigation of nonlinear effects of environmental regulation on carbon emissions while uncovering the threshold effect of environmental decentralization within this mechanism. Concurrently, three subdivided decentralizations are introduced, enabling a more granular analysis of threshold effects across distinct administrative responsibilities. Methodologically, the explanatory power of nonlinear modeling is strengthened through the application of the PSTR. Consequently, this provides empirical foundations for optimizing the allocation of environmental governance authority between central and local governments and formulating regionally differentiated emission reduction policies.

The structure of the paper is organized as follows: Section 2 provides a review of studies on the link between environmental regulation, decentralization systems, and carbon emissions; Section 3 constructs a theoretical framework and comes up with research hypotheses; Section 4 outlines the model setting and data sources; Section 5 presents empirical findings and analyzes nonlinear effects; Section 6 verifies the reliability of conclusions through robustness tests and heterogeneity analysis; Section 7 offers concluding insights and policy suggestions.

2. Literature Review

Environmental regulation, as the core policy tool for government control over carbon emissions, has attracted widespread attention regarding its mechanisms and effects. China’s unique decentralization system, which includes both environmental and fiscal decentralization, serves as a key institutional framework for implementing environmental governance in the country. These systems form the basic framework for China’s carbon emission governance. This paper summarizes existing research in three main areas: the direct impacts of carbon emissions from environmental regulations, the impacts of carbon emissions from environmental decentralization, and fiscal decentralization’s effect as a key complement to decentralization.

2.1. The Impact of Environmental Regulation on Carbon Emissions

The direct influence of environmental regulation on carbon emissions does not follow a simple linear pattern; instead, it exhibits significant threshold effects and heterogeneity. Yin et al. [10] observed that the relationship between environmental regulation intensity and carbon emissions follows an inverted U-shape [11]. They suggest that weaker environmental regulations might actually lead to higher carbon emissions, a phenomenon explained by the “green paradox” effect, while high-intensity regulation significantly reduces carbon emissions. Zhang et al. [12] confirmed this nonlinearity using a threshold model, highlighting that the impact of environmental regulation on emission reduction is stronger in regions with high foreign investment. However, Radulescu et al. [13] pointed out that the effectiveness of environmental policies is limited by institutional design, with a single policy tool being insufficient to achieve emission reduction goals.

In terms of the channels of effect, the carbon reduction impact of environmental regulation is mainly achieved by driving technological innovation and upgrading the industrial structure [14]. Pei et al. [15] found that technological efficiency partially mediates the connection between environmental regulation and carbon emissions, especially in high-energy-consuming sectors like petroleum refining. Chen et al. [16] further proposed that the degree of industrial structural optimization is a crucial factor for the success of environmental regulation—when the industrial structure is better optimized, the ability of environmental regulations to reduce emissions is significantly strengthened. Furthermore, Guo and Wang [17] emphasized that, in the long run, environmental regulation could facilitate carbon reduction by encouraging green technological innovation (such as clean energy technologies), although it may hinder technological investment in the short term due to increased costs for enterprises.

The effectiveness of environmental regulation varies significantly depending on regional development levels and industry characteristics. Lu, Wu, Yang, and Tu [3] observed that environmental regulation leads to greater emission reductions in the eastern regions compared to the central and western areas, which can be attributed to the higher capacity for technological adoption and more advanced market conditions in the east. Zhao et al. [18] demonstrated that market-based environmental regulations (such as emissions trading) lead to more significant emission reductions than command-and-control policies in the electricity industry. In the agricultural sector, Fan and Li [19] pointed out that informal environmental regulations (e.g., public environmental awareness) are more effective at reducing carbon emissions from food production than formal policy tools, highlighting the importance of social participation.

2.2. The Impact of Environmental Decentralization on Carbon Emissions

Environmental decentralization affects carbon emissions by redistributing environmental responsibilities across central and local governments, but its effectiveness largely depends on how decentralization is structured and the supporting institutional frameworks. Environmental decentralization encompasses administrative, monitoring, and supervisory dimensions, with varying emission reduction effects across these dimensions. Ran et al. [20] found that environmental administrative decentralization (local environmental department autonomy) can enhance emission reduction efficiency due to information advantages, whereas environmental supervision decentralization (delegating law enforcement authority) may weaken regulatory efforts due to local protectionism. Lin and Xu [21] used micro-enterprise data to show that environmental monitoring decentralization increases the risk of data falsification by enterprises, which is counterproductive to emission reduction. Case studies by Che et al. [22] demonstrated that the New Energy Demonstration Cities Program (NEDC), through the combination of “environmental decentralization + technology empowerment” policies, achieved significant collaborative emission reductions in western cities, validating the importance of matching decentralization with capacity building.

The impact of environmental decentralization is constrained through the strength of regional institutions. Hao et al. [23] noted that the suppression of environmental emergencies by environmental decentralization is significant in low-corruption regions but is entirely offset in high-corruption areas. Feng et al. [24] emphasized that the development of digital finance can alleviate the financing constraints under environmental decentralization, thereby enhancing the contribution of green technological innovation to emission reductions. Furthermore, environmental decentralization strengthens local governments’ environmental governance responsibilities, leading to a “Porter effect”, but this requires supplementary incentives, such as ecological performance assessments according to Li, Guo and Di [9].

Existing research on decentralization reform is divided. Xu et al. [25] advocate for moderately centralized environmental regulatory powers to reduce local interference, while Feng et al. [26] argue that governance-type decentralization (e.g., public participation in environmental decision-making) is more likely to stimulate innovation. Jiang et al. [27] propose a differentiated decentralization strategy: strengthening central vertical management in ecologically fragile regions while granting more environmental responsibilities to economically developed regions to leverage their governance capabilities.

2.3. Fiscal Decentralization as a Key Complement to Decentralization

In the study of the impact of decentralization on carbon emissions, fiscal decentralization is also an important research branch. Fiscal decentralization reflects the financial arrangements between the central and local governments and can, to some extent, serve as a complement to environmental decentralization. Therefore, this paper also reviews the relevant literature on the impact of fiscal decentralization on carbon emissions.

Currently, there is little consensus on the impact of fiscal decentralization on carbon emissions. On one hand, enhanced fiscal autonomy can motivate local governments to tailor environmental policies to local conditions [28,29]. On the other hand, local governments might ease environmental controls in their pursuit of GDP growth, resulting in a “race-to-the-bottom” effect [30].

Yang et al. [31] empirically showed that fiscal decentralization generally exacerbates carbon emissions, but its impact is weakened in the eastern regions due to industrial upgrading and amplified in the western regions due to increased energy dependence. Xu and Li [32] further indicated that China’s form of decentralization weakens the effectiveness of emission reduction policies by distorting factor markets (e.g., reducing the price of environmental factors), which leads to a “policy implementation gap”. Xia et al. [33] found that GDP-driven promotion incentives cause local governments to prioritize high-tax but high-pollution industries, with environmental regulation becoming a “symbolic policy”. Sun et al. [34] showed that, under fiscal decentralization, the vertical decentralization of powers (central–local government responsibility division) and horizontal competition (GDP competition between regions) jointly shape the trajectory of carbon intensity changes, with dynamic threshold effects. To address this dilemma, Dong et al. [35] proposed embedding green taxes into the decentralization system through fiscal–taxation linkage mechanisms to correct local government incentives.

Additionally, some literature has found that fiscal decentralization has a spatial spillover effect on carbon emissions. Liu and Yang [36] used a spatial Durbin model to find that increased fiscal decentralization in one region exacerbates carbon emissions in neighboring areas through industrial transfer and pollution leakage, forming a “free-rider” effect. Wang and Yu [37] further pointed out that environmental decentralization amplifies this spatial spillover, particularly when environmental regulatory powers are decentralized but regional joint prevention and control mechanisms are absent. In response, Luo et al. [38] suggested building a “cooperative environmental federalism” model to strengthen cross-regional collaborative governance through central environmental supervision and ecological GDP assessments.

Existing literature systematically reveals the complex mechanisms through which environmental regulation, fiscal decentralization, and environmental decentralization affect carbon emissions, but there is still room for further research. First, most existing studies examine either environmental decentralization or environmental regulation in isolation, neglecting the potential synergistic relationship between the two. Second, when incorporating decentralization into the environmental regulation framework, most studies focus on fiscal decentralization, which primarily concerns financial powers, while the allocation of environmental governance responsibilities reflects personnel arrangements and cannot fully capture the nuances of central–local governance. Additionally, different environmental functions and responsibilities may have heterogeneous effects on carbon emissions depending on their level of decentralization. This paper addresses these gaps by examining the combined effects of environmental regulation and environmental decentralization on carbon emissions and exploring the heterogeneous effects of three specific types of decentralization (

Table 1. Literature review.

		Authors	Time	Conclusion	References
Environmental Regulation and Carbon Emissions	Direct Influence	Yin et al.	2022	The relationship between environmental regulation intensity and carbon emissions follows an inverted U-shape.	[10]
		Zhang et al.	2020	The nonlinearity is confirmed by using a threshold model.	[12]
		Radulescu	2024	A single policy tool is insufficient to achieve emission reduction goals.	[13]
	Channels of Effect	Pei et al.	2019	Technological efficiency mediates the connection between environmental regulation and carbon emissions.	[15]
		Chen et al.	2019	Industrial structural optimization is a crucial factor for the success of environmental regulation.	[16]
		Guo and Wang	2018	Environmental regulation could facilitate carbon reduction by encouraging green technological innovation.	[17]
	Heterogeneity	Lu et. al	2022	Environmental regulation leads to greater emission reductions in the eastern regions.	[3]
		Zhao, et al.	2015	Market-based environmental regulations lead to more significant emission reductions than command-and-control policies.	[18]
Environmental Decentralization and Carbon Emissions	Direct Effectiveness	Fan and Li	2022	Informal environmental regulations are more effective at reducing carbon emissions.	[19]
		Ran et al.	2020	Environmental administrative decentralization can enhance emission reduction efficiency.	[20]
		Lin and Xu	2022	Environmental monitoring decentralization increases the risk of data falsification by enterprises.	[21]
		Che et al.	2023	NEDC policies achieve significant collaborative emission reductions in western cities.	[22]
	Influence Factor	Hao et al.	2022	The suppression of environmental emergencies by environmental decentralization is significant in low-corruption regions.	[23]
		Feng et al.	2022	The development of digital finance alleviates the financing constraints under environmental decentralization.	[24]
		Li et al.	2021	Environmental decentralization strengthens local governments’ environmental governance responsibilities, leading to a “Porter effect”.	[9]
	Decentralization Reform	Xu et al.	2023	They advocate for moderately centralized environmental regulatory powers to reduce local interference.	[25]
		Feng et al.	2020	Governance-type decentralization is more likely to stimulate innovation.	[26]
		Jiang et al.	2023	They propose a differentiated decentralization strategy.	[27]
Fiscal Decentralization as a Complement	Impact of Fiscal Decentralization	Yang et al.	2021	Fiscal decentralization generally exacerbates carbon emissions.	[31]
		Xu and Li	2023	China’s form of decentralization weakens the effectiveness of emission reduction policies by distorting factor markets.	[32]
		Xia et al.	2022	GDP-driven promotion incentives cause local governments to prioritize high-tax but high-pollution industries.	[33]
		Sun et al.	2024	The vertical decentralization of powers and horizontal competition jointly shape the trajectory of carbon intensity changes.	[34]
		Dong et al.	2023	They propose embedding green taxes into the decentralization system.	[35]
	Spatial Spillover Effect	Liu and Yang	2022	Increased fiscal decentralization exacerbates carbon emissions in neighboring areas.	[36]
		Wang and Yu	2023	Environmental decentralization amplifies spatial spillover.	[37]
		Luo et al.	2019	They suggest building a “cooperative environmental federalism” model to strengthen cross-regional collaborative governance.	[38]

).

3. Theoretical Analysis and Hypotheses

3.1. The Threshold Effect of Environmental Decentralization

Environmental decentralization is the core institutional arrangement of China’s environmental governance system. It profoundly impacts the policy execution logic of local governments by adjusting how environmental responsibilities are distributed across central and local authorities [39].

When the level of environmental decentralization falls below a certain threshold, the division of environmental governance responsibilities between the central and local governments tends to favor a more centralized system, with stricter control by the central government. Local environmental protection agencies often lack the necessary human, financial, and decision-making powers, which limits their autonomy. In this context, decentralizing governance powers allows for an expansion of the scale of local environmental protection departments [40].

This expansion of autonomy enables local authorities to leverage their information advantages, making environmental regulations more effective. Consequently, through promoting technological upgrades in enterprises and strictly enforcing environmental regulations, local governments can achieve carbon reduction goals [16].

When environmental decentralization surpasses the threshold, local governments gain greater autonomy, but conflicts between central and local government goals intensify. Local officials’ promotions are highly dependent on GDP growth rates. Deeper decentralization encourages resource allocation to high-revenue, high-carbon industries, leading to a “growth-first” strategy [4]. Local governments may lower environmental standards to attract investment, and even tacitly approve unauthorized projects, thereby exacerbating carbon emissions [9]. Local environmental departments, dependent on local governments for funding and personnel, may be influenced by local businesses, leading to weak enforcement, and undermining regulatory effectiveness [21]. These mechanisms together lead to “symbolic enforcement” of environmental regulations and may even trigger the “green paradox”—increased regulation may inadvertently stimulate businesses to offset costs through short-term capacity expansion, ultimately increasing carbon emissions [10] (Figure 1).

3.2. Heterogeneous Mechanisms of Subdivided Decentralization

The three types of subdivided environmental decentralization may produce heterogeneous moderating effects in the impact of environmental regulation on carbon emissions, which is likely closely related to the responsibilities of each type of decentralization.

Environmental administrative decentralization includes the affiliation of environmental protection agencies, personnel appointments, and fiscal relationships. They are primarily managed by local governments, with dual leadership from higher-level environmental protection departments [20]. Environmental administrative affairs encompass the formulation of local environmental regulations and protection plans, as well as responsibilities for personnel appointments and institutional management within environmental departments.

This dual focus reflects the local government’s prioritization of environmental governance.

Environmental supervision decentralization includes specific tasks such as pollution control and law enforcement penalties. Its functional scope is broad, covering the entire process of pre-emptive, in-process, and post-process supervision. It is considered a labor-intensive field [23].

Environmental monitoring decentralization forms the foundation for environmental management. Environmental monitoring stations are responsible for monitoring air quality, industrial pollution, and other aspects within a region. It is more directly related to public environmental services, with environmental monitoring data reflecting the quality of the regional ecological environment [41].

The three types of environmental decentralization may each generate a threshold effect in the impact of environmental regulation on carbon emissions, similar to the overall environmental decentralization threshold effect. However, due to their differing functional attributes, their direct impact on carbon emissions may vary.

A higher level of environmental administrative decentralization grants local governments more power to design and implement policies tailored to local conditions. This information advantage allows local governments to more efficiently implement emission reduction measures, improving the reduction efficiency.

A higher level of environmental monitoring decentralization allows local governments to better control the collection and reporting of environmental data. This enables local environmental protection departments to adopt more precise regulatory measures, optimizing emission reduction strategies through timely and accurate data feedback, leading to direct carbon emission reduction effects [22].

Environmental supervision decentralization, on the other hand, is labor-intensive. Although it involves many complex tasks, it is more difficult to reflect the policy orientation of local departments. Therefore, environmental supervision decentralization may have a limited direct impact on regional carbon emissions.

3.3. Research Hypotheses

Drawing from the theoretical mechanisms discussed, the subsequent hypotheses are proposed:

Hypothesis 1 (Overall Threshold Effect): 

Environmental decentralization exerts a nonlinear moderating influence on the emission reduction effectiveness of environmental regulation. If environmental decentralization is below a certain threshold, increasing regulatory intensity leads to a significant reduction in carbon emissions. However, once decentralization exceeds this threshold, the impact of regulatory measures on carbon emissions shifts from negative to positive.

Hypothesis 2 

(Heterogeneity of Subdivided Decentralization):

H2a: 

Environmental administrative decentralization (EDA), environmental supervision decentralization (EDS), and environmental monitoring decentralization (EDM) generate a significant threshold effect in the impact of environmental regulation on carbon emissions as the overall decentralization does.

H2b: 

The direct effects of the three types of decentralization are heterogeneous. Environmental administrative decentralization and environmental monitoring decentralization will have a significant inhibitory effect on carbon emissions, while the direct effect of environmental supervision decentralization is not significant.

4. Model Setting and Variable Sources

4.1. Model Setting

The Panel Smooth Transition Regression (PSTR) model is a tool for handling nonlinear and heterogeneous data and is widely used to capture nonlinear relationships between variables. Unlike traditional linear regression models, the PSTR allows the relationship between the explanatory variables and the dependent variable to smoothly change as certain variables (i.e., smooth transition variables) vary. In examining how environmental regulation influences carbon emissions, the PSTR is ideal for exploring whether its impact varies nonlinearly across different levels of environmental decentralization [42].

The PSTR model is a panel data regression model used to capture nonlinear relationships between dependent and independent variables. By introducing a smooth transition function, the model allows the connection between the dependent and independent variables to change gradually based on certain latent variables. It is suitable for cases where this relationship is nonlinear, with the nonlinear form described by a smooth function.

Specifically, the PSTR model can be expressed as follows:

$$y_{it}={\beta }_0+{\beta }_1{x}_{it}·G\left(z_{it};\gamma ,c\right)+{\epsilon }_{it},$$

(1)

In this model, $y_{it}$ represents the dependent variable in the panel data, and $x_{it}$ represents the independent variables. $G\left(z_{it};\gamma ,c\right)$ is the smooth transition function, where $z_{it}$ is the smooth transition variable (i.e., the threshold variable). $\gamma$ is the shape parameter of the smooth transition function, controlling the transition speed, while c represents the smooth transition threshold, marking the critical point at which the transition occurs, i.e., the threshold value.

The PSTR model specified in this chapter can be expressed as follows:

$${{CO}_2}_{it}={\beta }_0+{\beta }_1{ER}_{it}+{\beta }_2{ED}_{it}+{\beta }_3{ER}_{it}G\left(\gamma ,\delta ,{ED}_{it}\right)+\beta X_{it}+{\epsilon }_{it}$$

(2)

${{CO}_2}_{it}$ represents the dependent variable, which refers to the carbon dioxide emission level at the provincial level. The logarithm of this variable is taken for processing in this chapter. ${ER}_{it}$ and ${ED}_{it}$ are the core explanatory variables in this study, representing environmental regulation intensity and the degree of environmental decentralization, respectively. $G\left(\gamma ,\delta ,{ED}_{it}\right)$ is the smooth transition function, which reflects how environmental decentralization at different stages influences carbon emissions. $X_{it}$ denotes a vector containing control variables, including various elements that could significantly influence carbon emissions. ${\epsilon }_{it}$ is the error term, representing other influencing factors that are not explained by the model.

The smooth transition function captures the key characteristics of the nonlinear relationship between carbon emissions and environmental regulation. Its basic construction follows a logit function, and its specific form is as follows:

$$G\left(\gamma ,\delta ,{ED}_{it}\right)={\displaystyle \frac{1}{1+\exp (-\gamma ({ED}_{it}-\delta \left)\right)}}$$

(3)

In this model, $\gamma$ controls the rate of transition, indicating the sensitivity of environmental regulation to changes at different levels of environmental decentralization. $\delta$ represents the threshold value of environmental decentralization, which highlights the point at which the relationship between carbon emissions and environmental regulation undergoes a significant change when local governments’ implementation capacity or policy priorities reach a certain level. ${ED}_{it}$ is the threshold variable, representing the degree of environmental decentralization.

The core of the PSTR lies in capturing how environmental regulation influences carbon emissions at varying levels of decentralization through the smooth transition function $G\left(\gamma ,\delta ,{ED}_{it}\right)$. The design of the smooth transition function ensures that the effect of environmental decentralization on carbon emissions evolves gradually, rather than undergoing a sudden shift or following a linear pattern. This mechanism effectively reflects the varying economic and social challenges faced by local governments when implementing environmental regulations.

4.2. Variable Selection and Data Sources

Dependent Variable: Carbon dioxide emissions (${\mathrm{C}\mathrm{O}}_2$) are the dependent variable, primarily used to measure the carbon emissions level at the provincial level. This study follows the approach of Feng et al. [43] by utilizing provincial ${\mathrm{C}\mathrm{O}}_2$ emissions data derived from the China Energy Statistical Yearbook’s regional energy balance tables. These tables offer detailed information on energy consumption across provinces, including the use of coal, oil, and natural gas. By applying corresponding carbon emission factors, the consumption of various types of energy can be converted into ${\mathrm{C}\mathrm{O}}_2$ emissions. The data on carbon emissions is processed using a logarithmic transformation.

Core Explanatory Variables: This study adopts the methods used by Pei, Zhu, Liu, Wang, and Cao [15] and Yin, Liu, and Gu [10] to measure environmental regulation intensity, based on the share of pollution control investment and expenditure in regional GDP. Specifically, pollution control investment includes funds for environmental protection, investments in environmental facilities, and other relevant expenditures. These investments reflect local government spending on pollution control.

Threshold Variable: Environmental decentralization (ED) is the threshold variable. This paper follows the approach used in existing literature [23,30] by using the changes in personnel allocation between central and regional environmental protection departments as an indicator of environmental decentralization.

From the perspective of decentralization metrics, personnel allocation and fiscal decentralization constitute critical dimensions in measuring environmental decentralization. While substantial literature substitutes fiscal decentralization for environmental decentralization, these two concepts exhibit fundamental distinctions. Fiscal decentralization emphasizes the division of financial authority between central and local governments, particularly in tax revenue distribution, representing regional fiscal revenue-expenditure capacity. Although environmental protection is constrained by local fiscal authority, financial power does not equate to administrative authority. This relationship does not fully align with the administrative authority distribution principles under fiscal federalism.

Within environmental federalism frameworks, the administrative authority over environmental protection is more closely associated with institutional scale and workforce size. Utilizing personnel allocation characteristics within environmental protection agencies across government tiers to delineate China’s environmental decentralization practices demonstrates high operational feasibility and validity. More specifically, using personnel allocation as a measurement for environmental decentralization offers the following advantages:

(1)

Local environmental responsibilities encompass policy development, environmental oversight, pollution testing, infrastructure development, and environmental protection investments. Changes in the number of personnel reflect the government’s focus on different environmental affairs.

(2)

In China, the setup of institutions and personnel is the vehicle for government service provision. Given the relatively stable personnel numbers within the national environmental protection framework, the allocation of personnel between central and local departments reflects the government’s approach to dividing environmental regulatory responsibilities.

(3)

Environmental decentralization fundamentally represents the delegation of management authority. In contrast to fiscal spending, shifts in staffing composition more effectively capture the core of management decentralization.

Specifically, this paper uses the China Environmental Yearbook, which records the personnel arrangements of national and provincial environmental protection agencies. Before 2016, the yearbook recorded the overall staffing levels in both the national and provincial environmental protection systems. Additionally, it categorized personnel by specific functions, such as administrative departments, supervisory agencies, monitoring stations, research institutions, and public education institutions. This study selected personnel data from environmental protection agencies between 2005 and 2015, constructing an overall environmental decentralization index based on the proportion of staff within environmental protection agencies [4,20,36]. Furthermore, considering that environmental public affairs primarily focus on administrative management, environmental supervision, and environmental quality monitoring, three subdivided indicators of environmental decentralization were constructed. Environmental Administrative Decentralization (EDA) involves delegating administrative responsibilities to local authorities for policy implementation and enforcement. Environmental Supervision Decentralization (EDS) gives local authorities the responsibility to monitor and enforce environmental regulations. Environmental Monitoring Decentralization (EDM) allows local agencies to handle environmental data collection and analysis. The construction process is outlined as follows:

$${ED}_{it}=\left[{\displaystyle \frac{{LEPP}_{it}/{POP}_{it}}{{NEPP}_t/{POP}_t}}\right]\times [1-({\displaystyle \frac{{GDP}_{it}}{{GDP}_t}})]$$

(4)

$${EDA}_{it}=\left[{\displaystyle \frac{{LEAP}_{it}/{POP}_{it}}{{NEAP}_t/{POP}_t}}\right]\times [1-({\displaystyle \frac{{GDP}_{it}}{{GDP}_t}})]$$

(5)

$${EDS}_{it}=\left[{\displaystyle \frac{{LESP}_{it}/{POP}_{it}}{{NESP}_t/{POP}_t}}\right]\times [1-({\displaystyle \frac{{GDP}_{it}}{{GDP}_t}})]$$

(6)

$${EDM}_{it}=\left[{\displaystyle \frac{{LEMP}_{it}/{POP}_{it}}{{NEMP}_t/{POP}_t}}\right]\times [1-({\displaystyle \frac{{GDP}_{it}}{{GDP}_t}})]$$

(7)

In this model, ${LEPP}_{it}$, ${LEAP}_{it}$, ${LESP}_{it}$, and ${LEMP}_{it}$ denote the total number of staff in the environmental protection system, as well as the personnel assigned to environmental administration, supervision, and monitoring for province i in year t. ${NEPP}_t$, ${NEAP}_t$, ${NESP}_t$, and ${NEMP}_t$ denote the total personnel in the national environmental protection system, covering both central and local levels, along with the staff involved in environmental administration, supervision, and monitoring for year t. Meanwhile, ${POP}_{it}$ indicates the population size of province i in year t, and ${POP}_t$ represents the national population in the same year.

Additionally, considering the differences in economic development across provinces, the proportion of each province’s economic scale (${GDP}_{it}$) is included in the decentralization index calculation. When a province has a larger economic scale, its economic adjustment factor is smaller, thereby correcting for measurement errors caused by regional economic heterogeneity.

To ensure the robustness of the model and account for other potential factors affecting carbon emissions, several control variables are introduced based on existing literature [9,37,44,45]. Provincial foreign direct investment (FDI) is an important indicator of the local economic openness, and it shows important effect on carbon emission; infrastructure level (Infra) is measured by the per capita total infrastructure investment. This indicator reflects a region’s development status in transportation, energy, telecommunications, and related sectors. Well-developed infrastructure not only promotes economic growth but may also indirectly influence carbon emissions. Economic development level (GDP) is reflected by regional GDP size. Generally, there is a positive correlation between economic development and carbon emissions, as economic growth is often accompanied by increased energy consumption and rising carbon emissions. Research and development (R&D) investment is measured by the proportion of provincial R&D expenditure to GDP, reflecting regional investments in promoting green technologies, low-carbon industries, and innovation capacity. Industrial structure level (Str) is measured by the share of the secondary industry in GDP. In most cases, the secondary industry, particularly heavy industry and manufacturing, is typically associated with higher carbon emission levels. Given the availability of data, this study constructs a panel dataset for 30 provincial-level regions in mainland China (excluding Tibet) covering the period from 2005 to 2015. All data are sourced from official records, including the China Environmental Yearbook, China Environmental Statistical Yearbook, China Science and Technology Statistical Yearbook, Wind Energy Database, China Energy Statistical Yearbook, and the National Bureau of Statistics (

Table 2. Descriptive statistics of main variables used.

Variable	Obs	Mean	Std. Dev.	Min	Max
CO2	330	9.982139	0.7731715	7.397047	11.46513
ER	330	0.6887679	0.3764091	0.0010754	2.353331
ED	330	0.9752518	0.3573501	0.4736597	2.290762
FDI	330	0.0301037	0.0341161	0.0006826	0.1812011
Infra	330	0.8637694	0.5237504	0.0430569	2.52381
GDP	330	14344.13	13106.2	499.4	74732.4
RD	330	1.488879	1.074757	0.21	6.08
Str	330	0.4145277	0.0852752	0.2830286	0.7965269

).

5. Empirical Results and Analysis

5.1. Analysis of the Threshold Effect of Overall Environmental Decentralization

Before estimating the Panel Smooth Transition Regression (PSTR) model, it is necessary to estimate the parameters of the model to determine its specific form.

First, before constructing the PSTR, a nonlinear test is required to determine whether the model exhibits nonlinear effects, and to analyze whether there is a nonlinear regime in the model, that is, the parameter $\gamma$ in the smooth transition function $G\left(\gamma ,\delta ,{ED}_{it}\right)$. If $\gamma =0$, there is no nonlinear regime shift in the model, and a linear model should be used for estimation. If $\gamma \ge 1$, meaning the null hypothesis of the linear model is rejected, there is cross-sectional heterogeneity, indicating that the model has a nonlinear regime shift effect, and the PSTR can be applied for estimation.

To determine the presence of nonlinear regime shifts and the number of transitions in the model, this study uses the Wald test, Fisher test, and Likelihood Ratio Test (LRT).

The PSTR inherently accounts for individual heterogeneity in panel data while capturing nonlinear transition mechanisms. When nonlinearity is absent in regression specifications, the PSTR framework simplifies to a standard linear model. To statistically verify the presence of nonlinear effects, we conduct linearity testing through first-order Taylor expansion of the transition function $G\left(\gamma ,\delta ,{ED}_{it}\right)$, yielding an auxiliary regression equation that enables parameter estimation and hypothesis validation.

$${{CO}_2}_{it}={\beta }_0+{\beta }_1{ER}_{it}+{\beta }_2{ER}_{it}{ED}_{it}+{\beta }_3{ER}_{it}{ED}_{it}^2+\cdots +{\beta }_3{ER}_{it}{ED}_{it}^m+{\epsilon }_{it}$$

(8)

The linearity test specifically examines whether all interaction terms containing the threshold variable (ED) are statistically insignificant, i.e., ${\beta }_2={\beta }_3=\cdots ={\beta }_{\mathrm{m}}=0$. Three statistics are employed for this verification:

$$LM={\displaystyle \frac{TN\left({SSR}_0-{SSR}_1\right)}{{SSR}_O}}$$

(9)

$$LMF=({SSR}_0-{SSR}_1)/{\displaystyle \frac{mk}{{SSR}_1}}(TN-N-mk)$$

(10)

$$LRT=-2log{SSR}_1/{SSR}_0$$

(11)

The tests are conducted sequentially until the null hypothesis fails to be rejected in all cases.

Table 3. Nonlinearity test.

Hypothesis	Wald Tests (LM)	Fisher Tests (LMF)	LRT Tests (LRT)
$H_0:\gamma =0,H_1: \gamma =1$	W = 10.455	F = 9.783	LRT = 10.624
	p = 0.001	p = 0.000	p = 0.001
$H_0:\gamma =1,H_1:\gamma =2$	W = 29.682	F = 29.354	LRT = 31.103
	p = 0.000	p = 0.000	p = 0.000
$H_0:\gamma =2,H_1:\gamma =3$	W = 0.137	F = 0.061	LRT = 0.137
	p = 0.934	p = 0.941	p = 0.934

reports the three parameter tests for $\gamma$ equal to 1, 2, and 3, along with the corresponding statistics and p-values. In the parameter tests, the null hypotheses for $\gamma =0$ and $\gamma =1$ are rejected, but the null hypothesis for $\gamma =2$ cannot be rejected. This suggests that there is a nonlinear regime shift effect in the model, and the smooth transition function $G\left(\gamma ,\delta ,{ED}_{it}\right)$ contains two transition functions, with $\gamma =2$.

After determining that the number of transition functions, $\gamma$, is 2, the next step is to determine how many position parameters are needed, i.e., the threshold parameter $\delta$. This involves minimizing both the Bayesian Information Criterion (BIC) and the Akaike Information Criterion (AIC). Given that $\gamma =2$, the values of m are set to 1 and 2, and the corresponding BIC and AIC values of the current PSTR are calculated. The smaller the values of BIC and AIC, the more significant the model, and thus the more appropriate it is to use these parameters for PSTR estimation.

When $m=1$, AIC = −1.185 and BIC = −1.104. When $m=2$, AIC = −1.153, and BIC = −1.049. Regardless of whether AIC or BIC is considered, the values for $m=1$ are smaller than for $m=2$, indicating that the PSTR should have one threshold.

Based on $\gamma =2$ and $m=1$, the model was estimated. The results show that before the environmental decentralization level reaches the threshold, increasing the intensity of environmental regulation significantly reduces provincial carbon emissions. However, once environmental decentralization exceeds the threshold, the impact of environmental regulation on carbon emissions changes direction and becomes notably positive, with a net effect of 0.206 (i.e., −0.150 + 0.356). This indicates that after the environmental decentralization threshold is exceeded, strengthening environmental regulation no longer effectively reduces carbon emissions and may even significantly increase them.

However, when examined in isolation, environmental decentralization does not have a significant impact. Nevertheless, it plays a critical role in the overall impact of environmental regulation. This finding reveals the important role of environmental decentralization in moderating the effectiveness of environmental regulation, suggesting that simply strengthening environmental regulation may not be sufficient to achieve emission reduction targets. Particularly at high levels of environmental decentralization, local government policy responses may weaken or produce counterproductive effects [21,46,47].

Simultaneously, the threshold value of environmental decentralization is identified at 0.811, with only 40% of provincial samples distributed on the left side of the threshold. This structural imbalance reveals that for the majority of provinces, the current excessive level of environmental decentralization has undermined the effectiveness of environmental regulations in driving carbon mitigation. These findings carry significant policy implications: Central environmental authorities should strategically implement a moderate contraction of environmental governance authority while strengthening institutional constraints on local governments’ environmental responsibilities. Therefore, policymakers must carefully consider the actual level of environmental decentralization and the execution capacity of local governments when formulating and implementing environmental regulations, avoiding blindly increasing regulatory intensity to achieve effective carbon emission control [27]. Rather than indiscriminately intensifying regulatory pressures, a calibrated approach that aligns governance structures with jurisdictional capabilities proves essential for achieving optimal carbon emission control outcomes.

Additionally, after performing the PSTR regression, this study further conducted a nonlinear test and a residual nonlinearity test to verify the reliability and robustness of the PSTR application [48]. The linearity test was conducted using Taylor expansions from first to fourth order, progressively examining the model for nonlinearity. A rejection of the null hypothesis up to the fourth-order expansion indicates the model’s nonlinearity, justifying the application of the PSTR. The results of the tests show that all statistics indicate significant nonlinearity in the model (

Table 4. The PSTR regression results.

Variables	Linear Part	After Threshold
ER	−0.150 **	0.356 ***
	(0.0598)	(0.0656)
ED	−0.0975
	(0.115)
R&D	−0.0467 **
	(0.0233)
GDP	1.30 × 10−5 ***
	(1.86 × 10−6)
FDI	−1.041 *
	(0.610)
Str	−0.565
	(0.346)
Infra	0.467 ***
	(0.0708)
Threshold	0.811 ***
	(0.00941)
lngamma	4.751 ***
	(0.966)

Note: *, ** and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively.

).

The Escribano–Jorda linearity test distinguishes whether the smooth transition function should use the LSTR (Logistic Smooth Transition Regression) or the ESTR (Exponential Smooth Transition Regression) model. The p-value for the LSTR model is 0.0001052, while the p-value for the ESTR model is 0.0007652. Since the p-value for the LSTR model is smaller, the LSTR model should be selected.

As for the residual nonlinearity test, the p-values for the first-order to fourth-order Taylor expansions are all greater than 0.05, meaning the null hypothesis of the linear model cannot be rejected. It proves that the nonlinear aspects of the model have been adequately addressed, and there are no residual nonlinear factors in the remaining portion of the model. Therefore, the number of thresholds is robust (

Table 5. Nonlinearity test for the regression results.

Hypothesis	${\mathbf{H}}_0$	p-Value
Nonlinearity Test	b1 = 0	0.001613
	b1 = b2 = 0	0.00002431
	b1 = b2 = b3 = 0	0.00002691
	b1 = b2 = b3 = b4 = 0	5.106 × 10−7
Remaining Linearity Test	b1 = 0	0.7804
	b1 = b2 = 0	0.6069
	b1 = b2 = b3 = 0	0.6926
	b1 = b2 = b3 = b4 = 0	0.1158

).

5.2. Analysis of the Threshold Effect of Three Subdivided Decentralizations

First, we examine the case of environmental administrative decentralization. All three test mechanisms (LM, LMF, LRT) suggest that, with environmental administrative decentralization as the threshold variable, the PSTR reveals a single transition mechanism, where γ = 1 (

Table 6. Nonlinearity test for environmental administrative decentralization.

Hypothesis	Wald Tests (LMs)	Fisher Tests (LMFs)	LRT Tests (LRTs)
$H_0:\gamma =0,H_1:\gamma =1$	W = 116.319	F = 162.763	LRT = 143.421
	p = 0.000	p = 0.000	p = 0.000
$H_0:\gamma =1,H_1:\gamma =2$	W = 0.000	F = 0.000	LRT = 0.000
	p = 1.000	p = 1.000	p = 1.000

).

In determining the number of thresholds, if we consider the threshold number m = 1, the AIC = −1.200 and BIC = −1.154. If the threshold number m = 2, the AIC = −1.194 and BIC = −1.137. According to the principle of information criterion minimization, the number of thresholds should be considered as 1.

The regression results indicate that environmental regulation maintains a nonlinear effect on carbon emissions. Before the level of environmental administrative decentralization reaches the threshold of 2.114, the regression coefficient for environmental regulation has a notable negative impact, showing that increasing the intensity of environmental regulation can significantly reduce carbon emissions. However, once the threshold is exceeded, the overall regression coefficient for environmental regulation intensity becomes 0.1855 (i.e., −0.0875 + 0.273), suggesting that after surpassing the threshold, further increases in environmental regulation intensity will significantly increase carbon emissions. This suggests that the level of environmental decentralization moderates the strength of environmental regulation. Additionally, the regression coefficient for environmental administrative decentralization is also significantly negative, showing that an increase in administrative decentralization can significantly reduce carbon emissions, though the regression coefficient is relatively small. This suggests that policymakers need to consider the appropriate level of environmental administrative decentralization when enhancing environmental regulation intensity to ensure the smooth implementation and maximization of policy effectiveness (

Table 7. PSTR regression results for three subdivided decentralizations.

Variables	EDA	EDS	EDM
ER	−0.219 ***	−0.0875 *	−0.136 ***
	(0.0489)	(0.0447)	(0.0470)
EDA	−0.0626 ***
	(0.0184)
EDS		0.0119
		(0.0122)
EDM			−0.109 ***
			(0.0166)
Control	Yes	Yes	Yes
After Threshold	0.356 ***	0.273 ***	0.475 ***
	(0.0656)	(0.0579)	(0.0609)
Threshold	2.114 ***	1.362 ***	3.733 ***
	(0.0708)	(0.0425)	(0.429)
lngamma	2.340 ***	3.693 **	5.234 **
	(0.735)	(1.564)	(1.9)

Note: *, ** and *** indicate statistical significance at 10%, 5%, and 1% levels, respectively.

).

The three tests for environmental supervision decentralization also support the existence of a nonlinear transition regime in the model, with one transition mechanism. In determining the number of thresholds, when m = 1, AIC = −1.128 and BIC = −1.082. When m = 2, AIC = −1.119, and BIC = −1.061. The threshold number should be considered as 1 (

Table 8. Nonlinearity test for environmental supervision decentralization.

Hypothesis	Wald Tests (LMs)	Fisher Tests (LMFs)	LRT Tests (LRTs)
$H_0:\gamma =0,H_1:\gamma =1$	W = 69.114	F = 79.211	LRT = 77.553
	p = 0.000	p = 0.000	p = 0.000
$H_0:\gamma =1,H_1:\gamma =2$	W = 0.002	F = 0.002	LRT = 0.002
	p = 0.963	p = 0.965	p = 0.963

).

The regression results in column (2) reveal a notable nonlinear relationship between environmental regulation and carbon emissions. Below a threshold value of 1.362 for environmental supervision decentralization, increasing the intensity of environmental regulation significantly reduces emissions. However, once decentralization exceeds this threshold, the influence of stronger regulation on emissions turns positive. Moreover, while environmental supervision decentralization does not directly impact carbon emissions, it indirectly moderates the effect of environmental regulation by influencing its intensity.

This finding implies that the relationship between environmental supervision decentralization and carbon emissions is complex and does not follow a straightforward linear pattern. At lower levels of environmental supervision decentralization, increasing the intensity of environmental regulation effectively reduces emissions. This indicates that local governments have strong enforcement and response capabilities in environmental supervision, allowing for the effective implementation of emission reduction policies. However, once environmental supervision decentralization surpasses the threshold, local governments’ enforcement capacity may weaken, causing the increase in environmental regulation intensity to fail to effectively reduce carbon emissions, and potentially even leading to negative effects that increase carbon emissions.

As for monitoring decentralization, the three tests confirm the existence of one transition mechanism. When m = 1, AIC = −1.042 and BIC = −0.996. When m = 2, AIC = −1.033 and BIC = −1.976. Therefore, the threshold number should be considered as 1 (

Table 9. Nonlinearity test for environmental monitoring decentralization.

Hypothesis	Wald Tests (LMs)	Fisher Tests (LMFs)	LRT Tests (LRTs)
$H_0:\gamma =0,H_1:\gamma =1$	W = 76.237	F = 89.828	LRT = 86.689
	p = 0.000	p = 0.000	p = 0.000
$H_0:\gamma =1,H_1:\gamma =2$	W = 0.009	F = 0.008	LRT = 0.009
	p = 0.923	p = 0.927	p = 0.923

).

Column (3) presents the results based on monitoring decentralization (EDM). The relationship between environmental regulation and carbon emissions shows clear nonlinear patterns. When environmental monitoring decentralization is below the 3.733 threshold, the regression coefficient for environmental regulation shows a significant negative value, implying that increasing regulatory intensity results in a substantial decrease in carbon emissions. However, once decentralization surpasses this threshold, the coefficient rises to 0.339 (i.e., −0.136 + 0.475), indicating that stronger regulatory intensity leads to an increase in carbon emissions. Moreover, the regression coefficient for environmental monitoring decentralization is notably negative, suggesting that greater decentralization in environmental monitoring contributes to a reduction in carbon emissions.

From the perspective of the responsibilities of environmental monitoring decentralization, these results reveal the complex role of environmental monitoring decentralization in carbon emission management. At lower levels of environmental monitoring decentralization, local governments are able to effectively execute environmental monitoring and enforcement tasks, ensuring that environmental regulations are strictly implemented, which in turn reduces carbon emissions. However, after surpassing the threshold, an increase in environmental monitoring decentralization may weaken local governments’ capabilities in environmental monitoring and enforcement. Excessive decentralization may lead to lax environmental monitoring and enforcement in local governments due to a lack of effective supervision and incentive mechanisms, even resulting in implementation deviations, which would lead to poor regulatory effectiveness and potentially increase carbon emissions. This scenario suggests that excessive environmental monitoring decentralization may cause local governments to lack unified standards and coordination when implementing policies, leading to increased carbon emissions.

Additionally, this section conducts linearity tests on the regression results of the three subdivided decentralizations to ensure the accuracy and robustness of the PSTR. The linearity test uses first to fourth-order Taylor expansions to gradually examine whether nonlinearity exists. If all null hypotheses can still be rejected in the fourth-order Taylor expansion, it indicates that the model indeed exhibits nonlinearity, thereby confirming that the application of the PSTR is valid. The test results show that all statistics indicate significant nonlinearity in the model.

Furthermore, the Escribano–Jorda (E–J) test for LSTR and ESTR also indicates that LSTR is more suitable for the current smooth transition model (

Table 10. Nonlinearity test for three subdivided decentralizations.

Subdivided Decentralizations	${\mathbf{H}}_0$	p-Value
Environmental Administrative Decentralization	b1 = 0	0.001613
	b1 = b2 = 0	0.00002431
	b1 = b2 = b3 = 0	0.00002691
	b1 = b2 = b3 = b4 = 0	5.106 × 10−7
Environmental Supervision Decentralization	b1 = 0	0.0001212
	b1 = b2 = 0	0.0000501
	b1 = b2 = b3 = 0	1.281 × 10−6
	b1 = b2 = b3 = b4 = 0	7.079 × 10−7
Environmental Monitoring Decentralization	b1 = 0	4.763 × 10−12
	b1 = b2 = 0	3.346 × 10−11
	b1 = b2 = b3 = 0	6.848 × 10−11
	b1 = b2 = b3 = b4 = 0	2.218 × 10−13

).

6. Robustness Test and Heterogeneity Analysis

6.1. Changing the Transition Function

In this study, to assess the robustness of the PSTR, we substituted the transition function from LSTR (logistic function form) to NSTR (normal function form). By using NSTR instead of LSTR, we can verify whether the nonlinear effects are consistent under different transition function assumptions, thereby enhancing the robustness of the model. LSTR is suitable for capturing sharp nonlinear relationships, while NSTR can accommodate a broader range of smooth transitions. By comparing the results of both models, we can assess the stability of the model’s nonlinear effects. If the results are consistent, it indicates that the conclusions are more reliable.

Therefore, using the NSTR model for robustness testing can further verify the stability of the nonlinear effects, thus increasing the credibility of the study’s conclusions. The PSTR with NSTR as the transition function is constructed as follows:

$${{CO}_2}_{it}={\beta }_0+{\beta }_1{ER}_{it}+{\beta }_2{ED}_{it}+{\beta }_3{ER}_{it}G\left(\gamma ,\delta ,{ED}_{it}\right)+\beta X_{it}+{\epsilon }_{it}$$

(12)

In this model, the smooth transition function is replaced by $G\left(\gamma ,\delta ,{ED}_{it}\right)=\Phi \left(\gamma \right({ED}_{it}-\delta \left)\right)$, where $\Phi$ denotes the cumulative distribution function (CDF) of the normal distribution. Here, $\gamma$ still controls the transition rate, indicating the sensitivity of environmental regulation to changes at different levels of decentralization, and $\delta$ denotes the threshold value of environmental decentralization.

After changing the smooth transition function, we continue to use the PSTR for regression. Whether considering the overall decentralization regression results or the results for the three subdivided decentralizations, the signs and significance of the regression coefficients remain consistent with the previous analysis. Both the overall decentralization level and the three subdivided decentralization levels continue to moderate the carbon emission effects of environmental regulation intensity. When environmental decentralization is below the threshold, strengthening environmental regulation leads to a significant reduction in carbon emissions. However, once the threshold is surpassed, additional increases in regulation intensity result in higher emissions, aligning with the earlier findings. Furthermore, the direct effect of environmental decentralization on carbon emissions remains consistent. Neither the overall degree of decentralization nor environmental supervision decentralization significantly impacts emissions. However, environmental administrative decentralization and environmental monitoring decentralization continue to directly reduce emissions.

Through these robustness tests, we confirm the stability of the research conclusions, indicating that the policy implications and nonlinear relationships revealed in this study hold true under different model assumptions (

Table 11. Robustness test—changing the transition function.

Variables	ED	EDA	EDS	EDM
ER	−0.149 **	−0.218 ***	−0.0874 *	−0.137 ***
	(0.0598)	(0.0488)	(0.0447)	(0.0470)
ED	−0.0965
	(0.115)
EDA		−0.0627 ***
		(0.0184)
EDS			0.0119
			(0.0122)
EDM				−0.110 ***
				(0.0166)
Control	Yes	Yes	Yes	Yes
After Threshold	0.355 ***	0.610 ***	0.272 ***	0.477 ***
	(0.0654)	(0.0686)	(0.0579)	(0.0609)
Threshold	0.811 ***	2.124 ***	1.362 ***	3.642 ***
	(0.00917)	(0.0701)	(0.0436)	(0.0195)
lngamma	4.270 ***	1.856 **	3.174	7.238
	(1.038)	(0.803)	(1.945)	(22.02)

Note: *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively.

).

6.2. Grouped Linear Regression

In this robustness check, we follow the methods of Liu et al. [49]. By introducing a dummy variable, we divide the sample into high and low environmental decentralization intervals and then use panel regression to re-evaluate how varying levels of environmental regulation intensity influence carbon emissions across distinct decentralization regimes (

Table 12. Robustness test—grouped linear regression.

Variables	Sub-Samples	$\mathbf{\beta }$	t-Statistic
ED	Below	−0.211 ***	−0.0308
	Above	0.253 **	−0.0931
EDA	Below	−0.217 ***	−0.0609
	Above	0.395 ***	−0.111
EDS	Below	−0.208 ***	−0.0658
	Above	0.462 **	−0.147
EDM	Below	−0.160 **	−0.0726
	Above	0.138 **	−0.048

Note: ** and *** indicate statistical significance at 5% and 1% levels, respectively.

).

The specific procedure is as follows. Based on the threshold value obtained in the previous PSTR, we divide the sample into two sub-samples: one above the threshold and one below the threshold. We then conduct separate panel linear regressions for these two sub-samples to examine if the relationship between environmental regulation intensity and carbon emissions at different levels of decentralization aligns with the findings from the nonlinear model.

The regression outcomes reveal that when environmental decentralization is low (i.e., below the threshold), a rise in environmental regulation leads to a substantial decrease in carbon emissions, which is consistent with the regression results from the PSTR. However, at high levels of environmental decentralization (i.e., above the threshold), a rise in environmental regulation intensity results in higher carbon emissions, which also aligns with the results obtained from the nonlinear model. Specifically, in the low decentralization regime, the negative impact on carbon emissions remains significant, while in the high decentralization regime, the impact on carbon emissions shifts to positive, with a significantly positive regression coefficient. This indicates that the degree of environmental decentralization moderates the effect of environmental regulation.

This finding reinforces the nonlinear connection between environmental regulation and carbon emissions, highlighting the significant moderating effect that the level of environmental decentralization has on the effectiveness of regulation. In low decentralization regimes, environmental regulation effectively reduces carbon emissions, but in high decentralization regimes, the effect of environmental regulation reverses. This reversal may be related to factors such as local government enforcement capacity, regional economic development levels, and differences in policy implementation. By introducing dummy variables for sample grouping and applying panel regression for robustness testing, this analysis further affirms the role of environmental decentralization in shaping the influence of environmental regulation.

6.3. Heterogeneity Analysis

In order to investigate how the effect of environmental regulation intensity on carbon emissions varies by region, this study categorizes the sample into eastern, central, and western regions and analyzes the regional differences in the impact of environmental regulation. There are notable variations in economic development, industrial composition, and the implementation of environmental policies across China’s eastern, central, and western regions. Therefore, analyzing the impact of environmental regulation on carbon emissions in different regions is of great significance [31,50,51]. The data are separated into three groups: the eastern, central, and western parts of China. Regression analysis is then performed for each sub-sample to investigate whether the effect of environmental regulation varies significantly across these regions (

Table 13. Heterogeneity analysis for total decentralization.

Variables	East	Middle	West
ER	−0.245 ***	0.826 ***	0.0726
	(0.0539)	(0.195)	(0.0919)
ED	0.206	0.458 **	−1.601 ***
	(0.156)	(0.229)	(0.352)
Control	Yes	Yes	Yes
After Threshold	−0.451 ***	−0.713 ***	1.208 **
	(0.138)	(0.208)	(0.584)
Threshold	1.247 ***	1.213 ***	1.139 ***
	(0.0172)	(0.119)	(0.0399)
lngamma	5.113 **	4.995	3.227 ***
	(2.262)	(10.26)	(0.475)

Note: ** and *** indicate statistical significance at 5% and 1% levels, respectively.

).

The results of the heterogeneity analysis indicate that the effects of environmental regulation and decentralization on carbon emissions differ notably between regions.

First, in the eastern region, increasing regulation intensity consistently has a significant suppressive effect on carbon emissions. Whether before or after the threshold level of environmental decentralization, raising regulation intensity leads to a significant reduction in carbon emissions, with environmental decentralization enhancing this emission reduction effect. This is likely linked to the strong economic foundation, technological level, and the well-developed environmental regulatory system in the eastern region. The eastern region is more economically developed, has a more advanced industrial structure, and stronger environmental governance capacity, allowing it to effectively implement and enforce environmental regulations [25,32,51]. Additionally, technological innovation and green industry development in the eastern region are relatively mature, which facilitates industrial upgrading in the context of strengthened regulation, thus leading to a reduction in carbon emissions.

In contrast, the central and western regions show distinct outcomes. In the central region, initially, higher environmental regulation intensity leads to an increase in carbon emissions. However, once environmental decentralization surpasses a certain threshold, the impact of regulation becomes negative, although the magnitude of this effect significantly diminishes. Additionally, the regression coefficient for environmental decentralization is significantly negative, suggesting that in the central region, decentralization not only directly reduces carbon emissions but also helps lessen the adverse impact of environmental regulation on emissions to a certain degree.

In the western region, the intensity of environmental regulation leads to an increase in carbon emissions only after the threshold. Moreover, environmental decentralization not only directly promotes the increase in carbon emissions but also indirectly raises carbon emissions through the intensity of environmental regulation. This may be attributed to the economic structure, energy consumption patterns, and the challenges in enforcing environmental policies in these regions. These areas are still dependent on energy-intensive, high-pollution industries, and local governments often lack the capacity for effective environmental governance and enforcement of policies [22,36,49]. In such circumstances, the strengthening of environmental regulation may increase production costs for enterprises, particularly in resource-intensive and high-energy industries, leading to a “cost-shifting effect” and subsequently increasing carbon emissions. Furthermore, in these regions, environmental decentralization may lead local governments to prioritize immediate economic growth over sustainable environmental development, possibly even relaxing environmental regulation enforcement, thus exacerbating the increase in carbon emissions.

This paper continues to examine the heterogeneous effects of three subtypes of decentralization across different regions. In this section, the sample is divided into eastern regions and non-eastern regions to investigate the threshold effects of the three subtypes of decentralization in the impact of environmental regulation on carbon emissions. The results are shown in

Table 14. Heterogeneity analysis for subdivided decentralization.

Variables	EDA		EDS		EDM
	East	Other	East	Other	East	Other
ER	−0.608 **	−0.113	−0.234 ***	−0.0901	0.122	−0.209 **
	(0.270)	(0.0798)	(0.0467)	(0.0789)	(0.0755)	(0.0866)
EDA	−0.203 ***	0.00333
	(0.0363)	(0.0294)
EDS			0.000411	−0.00223
			(0.0201)	(0.0213)
EDM					−0.0339 ***	−0.0920 *
					(0.0118)	(0.0520)
Control	Yes	Yes	Yes	Yes	Yes	Yes
After Threshold	1.745 ***	0.501 ***	−0.747 ***	0.448 ***	−0.425 ***	0.533 ***
	(0.505)	(0.0989)	(0.194)	(0.0916)	(0.0708)	(0.105)
Threshold	2.097 ***	2.278 ***	4.010 ***	1.358 ***	0.223 ***	1.790 ***
	(0.454)	(0.142)	(0.0627)	(0.0209)	(0.0690)	(0.0426)
lngamma	−0.0964	3.428	2.402 *	4.742 ***	3.058 ***	3.220 ***
	(0.307)	(3.398)	(1.264)	(1.091)	(0.782)	(0.758)

Note: *, ** and *** indicate statistical significance at 10%, 5% and 1% levels, respectively.

.

In terms of environmental administrative decentralization, before the decentralization level reaches the threshold, the improvement of environmental regulation significantly reduces carbon emissions, but after crossing the threshold, the improvement of environmental regulation increases carbon emissions. At the same time, the increase in environmental administrative decentralization levels can produce significant emission reduction effects. In non-eastern regions, only after the administrative decentralization level crosses the threshold does the improvement of environmental regulation increase carbon emissions, while the impact on carbon emissions before the threshold is not significant, and administrative decentralization also shows no significant direct effects.

The effects of environmental supervision decentralization show substantial differences across regions. Supervision decentralization demonstrates no significant direct effects but exhibits different threshold effects. In eastern regions, the improvement of environmental regulation promotes carbon emission reduction regardless of threshold phases, and supervision decentralization further amplifies this emission reduction effect. In non-eastern regions, the carbon reduction effect of environmental regulation is not significant when the degree of decentralization is below the threshold, but excessive supervision decentralization leads to a promoting effect of environmental regulation on carbon emissions.

Environmental monitoring decentralization consistently shows significant carbon reduction effects but displays different threshold effects on environmental regulation across regions. In eastern regions, when monitoring decentralization is below the threshold, environmental regulation has no significant impact on carbon emissions, but above the threshold, the improvement of environmental regulation shows significant carbon reduction effects. In other regions, the threshold effect of environmental monitoring resembles the inverted U-shape mentioned earlier. When decentralization levels are below the threshold, higher-intensity environmental regulation promotes carbon reduction, but exceeding the threshold increases local carbon emissions.

These findings provide important insights for the formulation of differentiated regional carbon reduction policies, especially on how to adjust environmental regulations and decentralization arrangements according to the specific conditions of each region.

7. Conclusions

Effective control of carbon emissions is a critical safeguard for ensuring sustainability. This paper reviews the relevant literature on the relationship between environmental regulation, environmental decentralization, and carbon emissions. Based on this review, a provincial panel dataset from 2005 to 2015 in China is constructed to examine the impact of environmental regulation on provincial carbon emissions, as well as the threshold effects of environmental decentralization.

The intensity of environmental regulation has a significant impact on carbon emissions, but this effect is nonlinear. The level of environmental decentralization acts as a threshold variable that explains this nonlinearity, even though overall environmental decentralization does not show a significant direct effect. Before the threshold level of decentralization is reached, increasing the degree of environmental decentralization enhances local governance autonomy, which allows stronger environmental regulation to significantly reduce carbon emissions. However, once environmental decentralization exceeds the threshold, excessive local autonomy leads to conflicts of interest between central and local governments, and stronger environmental regulations may actually lead to an increase in carbon emissions.

When environmental decentralization is categorized into three specific types, their impacts on carbon emissions differ significantly. Environmental administrative decentralization and environmental monitoring decentralization both have significant direct effects on carbon emissions, while the direct effect of environmental inspection decentralization remains insignificant. However, when these three subcategories are incorporated into the analysis, the nonlinear relationship between environmental regulation and carbon emissions remains, and the threshold effects of the three types of decentralization are still clearly observable.

Additionally, robustness checks are conducted using alternative transformation functions and the inclusion of dummy variables for grouped regression, confirming the reliability of the findings. A regional heterogeneity analysis further shows that the effects of environmental regulation and environmental decentralization vary significantly across different regions.

Based on these findings, to ensure carbon control and sustainability of development, this paper offers the following policy recommendations:

For the sustainability of economic growth, due to the threshold effect of environmental decentralization, the division of environmental rights and responsibilities between the central and local governments should be carefully reconsidered. Both excessively low and excessively high levels of local autonomy hinder the carbon reduction effects of environmental regulation, suggesting the need for a balanced approach to decentralization. Decentralization should be maintained within an optimal range. By streamlining the scale of local environmental agencies and centralizing regulatory authority, a more robust oversight system should be established. In key functional areas such as environmental monitoring and inspection, recentralizing jurisdictional authority can strengthen the central government’s regulatory capacity. Enhanced supervision over local governance practices must be implemented while preserving local autonomy and regulatory capabilities, utilizing mechanisms like central environmental inspection protocols to ensure vertical accountability. Environmental governance outcomes, including carbon emission performance, should be formally incorporated into officials’ performance evaluations and promotion criteria to align central–local objectives and mitigate principal-agent risks.

In terms of the specific dimensions of decentralization, enhancing local environmental administrative decentralization and environmental monitoring decentralization can help improve the carbon reduction effects of environmental regulation. However, the delegation of environmental inspection authority should be approached with caution to avoid a race-to-the-bottom phenomenon [52]. Given the regional heterogeneity in the effects of environmental regulation and decentralization, differentiated environmental policies and institutional arrangements should be adopted. Local policies should be tailored to regional environmental conditions and economic structures, making use of local governments’ informational advantages [53]. When assessing officials, both economic and environmental performance should be integrated into the evaluation criteria to minimize the risk of a race to the bottom.