Does Dual-Pilot Policy of Broadband China and Low-Carbon City Enhance Carbon Emission Efficiency?

Abstract

As global climate change intensifies, achieving the dual goals of economic efficiency and low-carbon development has become a pressing challenge. Using panel data for 269 Chinese cities from 2010 to 2021, based on their carbon emission efficiencies (CEEs) measured using the DEA-SBM model, a staggered difference-in-differences (SDID) model is employed to identify the policy impacts, which is further extended into a triple-difference (DDD) framework to examine the causal impact of the dual-pilot policy. The results show that (1) China’s CEE has improved gradually but remains relatively low, with significant regional disparities. (2) Empirical results indicate that the dual-pilot policy leads to a significant improvement in CEE, raising it by approximately 4.06%. The positive impact is particularly pronounced in cities characterized by more advanced industrial structures and stricter environmental regulatory frameworks. (3) Industrial upgrading and green technological innovation serve as key mediating channels, contributing 2%, 7%, and 10.7% to the total mediation effect. (4) The positive impacts are particularly evident in eastern, large-scale cities. These results underscore that the integration of digitalization and low-carbon initiatives serves as an effective pathway to improving CEE. Therefore, policymakers are encouraged to further advance the dual-pilot programs, foster green technological innovation, and accelerate industrial upgrading toward a digitally empowered and low-carbon development model.

1. Introduction

Balancing economic growth with environmental integrity has become a defining test of sustainable development in the twenty-first century. According to the Global Energy Review 2025 published by the International Energy Agency (IEA), global carbon emissions rose by 0.8% in 2024, which is slower than the 3.2% increase in global GDP, but still reached a record high of 37.8 billion tons of CO₂ from energy-related sources. This trend reveals that economic expansion continues to depend heavily on energy-intensive and resource-dependent production patterns. In response, the Chinese government has implemented a comprehensive set of policies under its carbon peaking and carbon neutrality framework, positioning emission reduction and resource efficiency as central pillars of national development. The fundamental goal is to decouple economic growth from environmental degradation by advancing resource optimization, improving energy efficiency, and fostering green innovation, ultimately guiding China toward a sustainable and resilient development path [1].

Within China’s broader sustainability agenda, two landmark national programs include the LC pilot and the BC strategy; both of them have become central instruments for advancing environmental and digital transformation. Initiated in 2010, the LC program encourages cities to experiment with context-specific approaches to emission reduction and urban economic restructuring. Introduced three years later, the BC strategy redefined broadband networks as critical national infrastructure, positioning digital connectivity as a foundation for economic competitiveness and an accelerator of ecological transition. Empirical studies demonstrate that each initiative independently contributes to emission reduction. Specifically, broadband infrastructure—an essential component of the digital economy has been shown to reduce urban carbon emissions by approximately 12% to 17%, underscoring its significant role in promoting greener urban development [2]. Empirical evidence indicate that the LC pilot initiative can achieve a 7.3% annual reduction in residents’ carbon emissions [3], and contribute significantly to the reduction of overall urban carbon emissions [4]. Comparable environmental outcomes are also observed in relation to broadband infrastructure: as mobile broadband coverage expands, carbon emissions exhibit a marked decline [5]. Yang et al. (2022) pointed out that the development of digital cities, driven by digital technologies, plays a vital role in reducing carbon emission intensity [6]. Likewise, Wang et al. (2023) highlighted the significant potential of the digital economy in enhancing CEE [7]. Nevertheless, despite these encouraging prospects, the on-the-ground realization of such policies faces persistent obstacles. Regional disparities in development stages, industrial composition, and resource endowment lead to considerable variation in policy effectiveness across different urban contexts.

However, despite extensive attention to both policies, two fundamental gaps persist in the current literature. First, previous research has largely investigated the BC and the LC program as independent policy experiments, without considering how their concurrent rollout may generate interactive or compounded effects. As a result, the empirical understanding of whether digital infrastructure expansion and environmental governance can jointly amplify improvements in CEE remains limited. The absence of evidence on policy interdependence leaves the question of whether these strategies act as complements or substitutes in shaping sustainable urban transitions. Second, while industrial structure upgrading and green technological innovation are frequently identified as theoretical channels linking digitalization and carbon efficiency, prior studies often treat them as parallel or independent pathways. This fragmented approach fails to clarify their relative contributions, causal directions, or empirical significance under the joint BC–LC policy framework, leaving the internal mechanisms of policy synergy insufficiently verified. To address these gaps, this study conceptualizes the overlap between the BC and LC programs as a quasi-natural experiment that enables causal identification of policy synergy under staggered adoption. Using panel data from 269 Chinese cities between 2010 and 2021 and applying a SDID approach, the study systematically evaluates whether the joint implementation of DC and LC produces greater efficiency gains than either policy in isolation. Beyond average treatment effects, the analysis explores two mediating channels and heterogeneity across city types, extending into a difference-in-difference-in-differences (DDD) framework based on industrial structure and environmental regulation. This multi-dimensional framework allows for a comprehensive assessment of how institutional and technological transformations interact to influence CEE within the broader context of China’s sustainable development agenda.

This study makes several significant contributions to the existing literature. From a conceptual standpoint, it deepens the understanding of policy synergy as a critical catalyst for sustainable urban transformation by integrating digital development and environmental governance within a cohesive analytical framework. Methodologically, it refines causal identification through the SDID approach, mitigating the biases inherent in staggered policy implementation and offering more credible evidence on the joint effects of digital and low-carbon initiatives. It also extends into a DDD model to strengthen causal identification and capture differential policy effects across development contexts. Empirically, it develops a comprehensive evaluation index system for CEE that incorporates resource inputs, expected outputs, and undesirable emissions, and employs this framework to reveal the mechanisms and boundary conditions of policy effectiveness. By examining the combined impacts of this dual policy, the research extends previous studies and provides valuable insights for future policy formulation, emphasizing the importance of policy synergies in enhancing CEE.

2. Theoretical Analysis and Hypothesis

2.1. The Environmental Effects of the Pilot Policy

Broadband infrastructure has emerged as a key driver of sustainable urban development. Studies from OECD countries indicate that widespread broadband adoption accelerates digital integration and promotes industrial upgrading [8]. In China, the BC initiative not only enhances digital infrastructure but also stimulates entrepreneurship, expands employment, fosters innovation, and mitigates environmental harms such as air pollution [9,10,11,12]. The construction of broadband infrastructure not only enhances urban ecological efficiency and occupational health but also influences the consumption patterns of rural households [13,14] and reduces geographical barriers to communication [15]. In addition to this analysis of the BC initiative, the LC pilot policy serves as a more immediate mechanism for reducing carbon emissions. In their study, Song et al. (2019) revealed that the implementation of the LC plays a significant role in mitigating urban air pollution and fostering the transition toward low-carbon urban development [16]. Utilizing a DID model, Wolff (2014) investigated the impact of low-carbon zone policies in Europe on local air quality, concluding that such policies substantially improve air quality in designated pilot areas [17]. Similar findings have been corroborated by other scholars, particularly in the context of Germany, where low-carbon zone policies have demonstrably enhanced air quality in pilot regions [18]. Tang et al. (2020) assert that command-and-control policy instruments are effective in enhancing environmental efficiency within the Chinese context [19], with an emphasis on strengthening technological innovation as a critical facilitator of this outcome [20]. From a macroeconomic perspective, low-carbon construction significantly alleviates air pollution, and the implementation of corresponding governance measures effectively curtails carbon emissions [21]. The LC pilot policy is characterized by its relatively weak constraints, industry specificity, and the combination of various policy instruments [22]. Compared to more rigorous regulatory frameworks, this policy may produce limited environmental benefits. The BC pilot policy initiative provides substantial technical support for the advancement of low-carbon cities [23]. Specifically, it establishes a digital platform that facilitates the production and distribution of goods, enabling not only regional information sharing but also oversight of the low-carbon construction process and assessment of its efficacy [24]. Moreover, the expansion of digital infrastructure enables residents to embrace more sustainable consumption patterns and low-carbon lifestyles, effectively reducing their overall carbon footprint [25]. The advancement of a green digital economy not only expedites China’s progress toward its “dual carbon” objectives but also strengthens the development of the digital sector. The emergence of digital technologies has not only enhanced daily living but has also led to decreased energy consumption and a reduced personal carbon footprint [26]. The effective implementation of LC necessitates substantial support from information technology, including intelligent energy management systems and smart transportation solutions. The establishment and widespread adoption of broadband networks provide the essential foundation and assurance for the deployment of these technologies.

Both the BC and LC pilot policies exert notable effects on carbon dioxide emissions and the broader governance landscape. Nevertheless, the path toward low-carbon urban transformation is inherently uncertain. The BC initiative functions as an engine for economic growth, alleviating the high consumption and emission pressures associated with traditional industrial structures. This alignment with the LC pilot program’s objectives indicates that the two policies may work synergistically to enhance CEE. For LC pilot cities, the integration of digital technologies and knowledge is essential to move beyond conventional growth models and to navigate the complex governance challenges involved. Meanwhile, cities selected for broadband deployment already possess foundational digital infrastructure, which can facilitate the advancement of low-carbon governance efforts. Such infrastructure not only supports the steady advancement of LC initiatives but also promotes urban sustainability transitions through policy-oriented mechanisms. Fundamentally, the synergy between the BC and LC programs is anticipated to improve urban CEE by integrating technological progress with institutional innovation. The BC policy improves digital connectivity and information flows, enabling real-time monitoring, industrial upgrading, and innovation dissemination. At the same time, the LC policy strengthens regulatory oversight and environmental management. Their complementarity lies in the fact that digital infrastructure boosts the efficiency of environmental regulations, while low-carbon mandates further stimulate digital innovation. Based on this reasoning, we formulate the following hypothesis.

Hypothesis 1.

The dual-pilot policy of BC and LC will help improve urban CEE.

2.2. The Mechanism Effects of the Pilot Policy

The enhancement of digital infrastructure under the BC pilot policy improves CEE through two interrelated channels: industrial structure upgrading and green technological innovation. Rather than treating these effects in isolation, they are integrated into a coherent framework of technological–institutional co-evolution. The BC initiative promotes digital industry clustering, enhances information efficiency, and reallocates resources toward high-value, low-emission sectors, thereby optimizing urban industrial structures and boosting production efficiency [27]. In resource-based cities, broadband development has been shown to improve CEE through not only structural adjustments and innovation but also through the accumulation of human capital, reflecting broader productivity gains [28]. Similarly, the LC pilot policy drives transformations in traditional development patterns by encouraging technological upgrading and structural change, contributing to the growth of urban green GDP [29]. At the enterprise level, the LC framework’s environmental regulations serve as critical catalysts for green innovation. These regulations incentivize firms to implement eco-friendly production practices and improve environmental management, thereby boosting both environmental performance and CEE [30]. Additionally, LC pilot policies stimulate corporate innovation in green technologies, including energy conservation and renewable energy production [22], Xiao et al. (2023) further demonstrate that low-carbon pilot policies primarily influence green invention patents, particularly among large, state-owned, and high-carbon enterprises [31], highlighting the role of targeted regulation in promoting innovation quality.

The BC pilot policy, as a quasi-natural experiment in information infrastructure construction, enhances greenhouse gas emission efficiency by improving technological capabilities, optimizing industrial structure, and facilitating efficient factor allocation [32]. Guo and Liang (2022) confirm that LC pilot policies improve CEE through both direct effects and indirect mechanisms, such as technological advancement and structural optimization, with significant regional variation [33]. Furthermore, Zhang and Liu (2022) emphasize that the synergistic interaction between digitalization and green innovation effectively enhances CEE and generates spatial spillover effects [34]. Ning et al. (2021) find that government intervention, energy structure, and technological progress are key determinants of CEE, underscoring the multifactor nature of carbon efficiency improvement [35]. Consistent with these findings, Wang and Chu (2023) identify the BC pilot policy as an exogenous technological shock that significantly improves CEE, with green technology innovation serving as a complete mediating mechanism, thereby realizing both emission reduction and efficiency gains [36].

Scholars have applied diverse approaches to examine the determinants of CEE, with a primary focus on technological innovation and industrial structure optimization. Technological innovation and industrial restructuring exhibit both direct and intermediary effects in this regard. Industrial structure optimization, at the regional level, typically involves the reallocation of resources to high-value-added, low-emission industries, thereby reducing carbon emissions during production. Technological innovation improves production efficiency and optimizes energy use, reducing energy waste and lowering carbon emissions, thereby enhancing CEE. Moreover, policies that support digitalization and low-carbon development further strengthen carbon performance [23]. Based on the above analysis, we propose hypothesis 2 and hypothesis 3:

Hypothesis 2.

The dual-pilot policy of BC and LC will improve CEE by adjusting industrial structure.

Hypothesis 3.

The dual-pilot policy of BC and LC will improve CEE through technological innovation.

Based on the analysis above, the research framework for this study is illustrated in Figure 1.

Figure 1. Research framework diagram.

3. Research Design

3.1. Model Setting

The approach of DID is a commonly employed method in policy assessment, valued for its ability to capture changes between treated and control groups before and after the introduction of a policy [37]. Additionally, the DID approach effectively mitigates bias arising from unobservable factors, such as variations in economic development [16]. This study employs a SDID model, integrating both regional and temporal controls. The dual-pilot policy can create notable differences within a city over time and among pilot and non-pilot cities. The SDID method is particularly adept at controlling for the influence of other confounding factors, thereby isolating the net effect of the policy [37]. Since the BC and LC pilot policy was implemented in stages, this study applies an SDID model to investigate their progressive impact on CEE [23]. Based on this framework, we construct the following baseline evaluation model to capture the effects of the BC and LC dual-pilot policy on CEE.

$${\mathrm{CEE}}_{it}={\alpha }_0+{\alpha }_{1}Trea{t}_{it}Pos{t}_{it}+{\alpha }_{2}Contro{l}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{it}$$

(1)

$i$and t are the proxy variables of city and time, respectively, and the explanatory variable CEE is derived from the DEA-SBM mode, Core explanatory variable ${Treat}_{it}{Post}_{it}$ denotes the interaction between the individual and time dummy variables; if the sample city has implemented both the BC and the LC pilot policy, individual dummy variable ${Treat}_{it}$ is 1, otherwise, it is 0. Temporal dummy variable ${Post}_{it}$ is 0 before the implementation of the dual-pilot policy and 1 after its implementation. ${Control}_{it}$ is the selected control variable that affects the efficiency of CEE, ${\sigma }_t$ is a fixed effect of time, ${\mu }_1$ is an individual fixed effect, ${\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$ is a random perturbation term that affects the explanatory variable, and the coefficient ${\alpha }_1$ that we focus on in this paper represents the direction and degree of impact of the dual-pilot policy on CEE.

3.2. Variables

3.2.1. Explanatory Variables

Regarding the measurement of CEE, prevailing methods generally rely on inputs such as capital stock, labor, and electricity consumption, with outputs commonly represented by regional GDP and CO₂ emissions [34,38,39]. Wang (2016) constructed a comprehensive evaluation index system by selecting indicators across energy consumption [40], major pollution emissions, pollution treatment and utilization, and carbon sink capacity, establishing a comprehensive evaluation index system. Using a beneficiary DEA model, Wang assessed the carbon governance efficiency of 79 cities across six central provinces. Liu et al. (2023) measured urban carbon performance from three aspects: efficiency, effectiveness, and efficacy, finding that the digital economy promotes improvements in carbon performance by enhancing carbon productivity, reducing carbon intensity, and decreasing fossil energy consumption [41]. Additionally, Shao et al. (2022) combined technological frontier analysis with non-angular, non-radial directional distance functions within the DEA framework to assess CEE across 30 Chinese provinces, considering both input and output dimensions [42]. Low-carbon development is the outcome of a multifaceted effort that encompasses economic, scientific, and technological advancements. Therefore, relying on a single indicator to measure carbon performance is insufficient; a comprehensive set of indicators, integrating both inputs and outputs from the perspective of governance, is essential for a scientific evaluation [43]. Based on the aforementioned definition of CEE, this paper constructs a CEE index system, which incorporates both input and output dimensions (Table 1). Environmental governance relies heavily on the essential roles of capital and labor [44], and the capital stock is determined using the perpetual inventory method, based on the formula provided by [45].

$$K_{i,t}=K_{i,t-1}(1-{\delta }_{i,t})+I_{i,t}$$

(2)

Table 1. Evaluation system for carbon emission efficiency.

Dimension	Indicator	Explanation	Unit
Input	Personnel input	Number of employees in water conservancy, environment, and public facilities management	10,000 people
	Funding investment	Capital stock	10,000 yuan
	Energy inputs	Electricity consumption	10,000 kWh
	Investment in scientific and technological innovation	Number of patents granted	item
Undesired outputs	Carbon emissions	CO2 emissions	10,000 tons
Expected output	Economic growth	GDP	100 million yuan

$K_{it}$represents the capital stock of city i in the year t, ${\delta }_{it}$ is the depreciation rate, which is calculated at 9.6%, and $I_{i,t}$ represents capital flows. The effectiveness of low-carbon governance is significantly shaped by the role of urban environmental practitioners, with the optimal allocation of human resources playing a crucial role in enhancing low-carbon development outcomes [46].

Power consumption intensity, reflecting the balance between electricity usage and carbon dioxide emissions, serves as a robust indicator of both technological progress and governance effectiveness, making it a suitable proxy for energy input [47]. Electricity consumption, which encapsulates key dimensions such as urban energy efficiency, technological innovation, policy prioritization, and carbon reduction potential, is commonly employed as an input metric for assessing CEE [48]. Furthermore, technological innovation, as a fundamental driver of energy efficiency and carbon mitigation, reflects the strategic direction of government policies on sustainable development. The number of patents granted serves as an indicator of urban innovation capacity and the level of technological development [49]. In the context of efficiency evaluation, outputs are categorized into economic and environmental dimensions. In line with existing literature, GDP is used as an indicator of expected output, whereas carbon dioxide emissions are considered to be an undesirable output in the measurement of CEE [50].

3.2.2. Key Explanatory Variable

When both pilot policies are simultaneously implemented in the same city during the sample period, the policy dummy variable Treat as the key explanatory variable is set to 1; otherwise, it is 0. This research utilizes the SDID model to examine the link between the dual-pilot policy and CEE. A time dummy variable (Post) is also included. In pilot cities, post is assigned a value of 0 before the implementation of the BC and LC pilot policies in the same city, and for each subsequent year thereafter, it takes a value of 1.

3.2.3. Mediation Variables

In selecting intermediary variables, this study is grounded in prior theoretical analyses and relevant scholarly contributions. The selection of industrial structure optimization and green technological innovation as mediating variables is based on their essential functions in transmitting the effects of policy interventions to enhancements in CEE. Industrial structure optimization serves not only as a core component in transforming economic development models but also directly shapes resource allocation and energy consumption patterns, thereby impacting the intensity of carbon emissions [51]. It is typically quantified by the ratio of the tertiary industry to the secondary industry, as this ratio reflects the extent of an economy’s transition from traditional manufacturing to modern services. As the proportion of the service sector increases, overall energy consumption efficiency and carbon emission intensity tend to improve, a relationship widely validated in economic and environmental economic studies.

Green technological innovation plays a pivotal role in driving low-carbon development and facilitating the broader transition toward sustainable urbanization. It includes both substantial advances in new green technologies and incremental improvements in existing ones. In this study, two categories of green patents are employed as coarse proxies for different types of green innovation. Specifically, the number of green invention patents (lsfm) reflects higher-threshold, invention-type innovation activity, while the number of green utility-model patents (lssyxx) represents utility-type innovation activity emphasizing practical application and diffusion. These indicators capture the overall intensity and direction of regional green innovation rather than the precise degree of technological breakthrough, providing a useful approximation of innovation dynamics within cities [23].

3.2.4. Control Variables

To enhance the accuracy of the empirical analysis, this study incorporates a series of control variables into the model to minimize the impact of non-policy external factors on CEE. These variables reflect economic development, governance capacity, population density, and technological support, capturing city-level differences and aiding in accurately identifying policy effects. Table 2 delineates the definitions and descriptions of the associated variables.

Table 2. Definition and description of variables.

Indicator	Name of the Variable	Variable	Calculation Method
Explanatory variables	Carbon emission efficiency	CEE	Calculated based on DEA-SBM
Core explanatory variables	Dual-pilot policy	Treat*Post	BC
			LC
Mediation variables	Optimization of industrial structure	cyjg	Tertiary Sector/Secondary Sector
	Green technology innovation (GTI)	lsfm	Ln (Inventive GTI + 1)
		lssyxx	Ln (Improved GTI + 1)
Control variables	level of economic development	rjsr	Ln (GDP per capita)
	Educational attention	jy	Education expenditure/Local Government Expenditure
	Opening	dwkf	Actual utilization of foreign investment/GDP
	Decentralization of government	zffd	Fiscal expenditure/fiscal revenue
	Government intervention	zfgy	Fiscal spending/GDP
	Population agglomeration	rkjj	Population/total urban area
	Technology support	kyttr	Science and Technology Expenditure/Fiscal Expenditure

Per capita GDP reflects the overall economic capacity of a city, and improvements in economic development are typically accompanied by increased demand for energy-intensive and carbon-emitting resources [23]. The Chinese government has long emphasized the strategy of “rejuvenating the nation through science and education,” thereby strengthening the foundational support for high-quality development. Within this framework, education expenditure plays a crucial role in cultivating talent that supports low-carbon initiatives, measured by the ratio of education spending to total urban fiscal expenditure [52]. Furthermore, foreign investment exerts a substantial influence on the economic structure of cities and promotes their economic advancement [53]. To evaluate fiscal decentralization, this study employs the ratio of fiscal expenditure to fiscal revenue [54]. Additionally, government intervention is measured through the ratio of government expenditure to GDP within each city [55,56]. The extent of population concentration is represented by the total population density per unit area [57], while technological support is quantified as the proportion of technology expenditure relative to fiscal spending [51].

3.3. Data Description and Descriptive Statistics

Table 3 presents the descriptive statistics of the variables. The average CEE stands at 0.2936, with a standard deviation of 0.1376, ranging from a minimum of 0.0611 to a maximum of 1 over the sample period. These results reveal considerable regional differences in CEE. Furthermore, significant variation is observed in industrial structure, green technological innovation, and economic development levels, providing a robust basis for examining the effects of the dual-pilot policy on CEE.

Table 3. Descriptive statistics of key variables.

Variable	Obs	Mean	Std. Dev.	Min	Max
CEE	3228	0.2936	0.1376	0.0611	1.0000
Treat*Post	3228	0.0994	0.2993	0.0000	1.0000
cyjg	3228	1.0553	0.5892	0.1087	5.3500
lsfm	3228	1.2569	0.5032	0.0000	2.3235
lssyxx	3228	1.6796	0.3201	0.0000	2.3609
rjsr	3228	10.7353	0.5662	8.8809	12.2928
jy	3228	0.1762	0.0452	0.0010	0.8762
dwkf	3228	0.0243	0.0253	0.0000	0.2822
zffd	3228	2.8265	1.7645	0.8756	37.0761
zfgy	3228	0.2509	0.1129	0.0398	0.9186
rkjj	3228	0.0450	0.0353	0.0005	0.3160
kyttr	3228	0.0139	0.0156	0.0000	0.3226

3.4. Data Sources and Summary

To ensure data reliability and minimize the impact of missing values on the analysis, this study employed rigorous data cleaning and screening processes. City samples with a missing rate exceeding 20% were excluded, and outliers were systematically identified and corrected. To enhance the accuracy of manually collected data, multiple rounds of verification were conducted, cross-referencing policy implementation timelines and city distribution details with multiple official sources to mitigate potential subjective biases. Additionally, the distribution characteristics of the remaining data were thoroughly evaluated following the cleaning process, ensuring that the exclusion of certain samples did not significantly affect the overall analytical results. For the few remaining missing values, linear prediction and interpolation methods were employed, effectively maximizing data completeness and ensuring the robustness of the analysis. The study utilizes a dataset comprising 269 prefecture-level and higher cities, covering the period from 2010 to 2021. Information on the implementation timing and city coverage of the BC policy was obtained from the Ministry of Industry and Information Technology of China, whereas the details for the LC pilot policy were sourced from the National Development and Reform Commission. All city-level and policy timeline data were systematically compiled and cross-verified using official documents issued by the respective authorities. Macroeconomic indicators were primarily derived from the China City Statistical Yearbook, the EPS (Easy Professional Superior) database, and the CSMAR (China Stock Market and Accounting Research) database, while city-level CO₂ emissions data were collected from the China Carbon Accounting Database.

4. Results and Discussion

4.1. Characteristics of CEE

Data Envelopment Analysis (DEA) is widely recognized for its capability to evaluate efficiency by systematically analyzing input–output relationships [58], and it is frequently applied to assess both resource utilization and environmental performance [59,60]. Conventional DEA models, however, often overlook slack variables in efficiency measurement, making it difficult to differentiate decision-making units (DMUs) that achieve an efficiency score of 1. To overcome this limitation, Tone proposed the DEA-SBM (Slack-Based Measure) model, which reduces the impact of measurement errors on efficiency outcomes [61]. To correct for biases caused by radial and angular inconsistencies in efficiency evaluation, this study employs the DEA-SBM model incorporating undesired outputs to measure CEE. Consider a production system composed of n DMUs, each characterized by a set of inputs, desired outputs, and undesired outputs. The corresponding calculation formula is provided below.

$${\rho }^{\ast }=\min {\displaystyle \frac{1-\frac{1}{m}{\displaystyle \sum _{i=1}^m\frac{S_i^{-}}{X_i}}}{1+\frac{1}{S_1+S_2}\left[{\displaystyle \sum _{r=1}^{S_1}\frac{S_r^g}{y_{r0}^g}}+{\displaystyle \sum _{r=1}^{S_2}\frac{S_r^b}{y_{ro}^b}}\right]}}$$

(3)

$$\begin{array}{l}s.t\left\{\begin{array}{l}{x}_0=X\lambda +S^{-}\hfill \\ y_0^g=Y^g\lambda -S_{\omega }^g\hfill \\ y_0^b=Y^b\lambda +S^b\hfill \end{array}\right.\\ S^{-}\ge 0,S^g\ge 0,S^b\ge 0,\lambda \ge 0,0\le {\rho }^{\ast }\le 1\end{array}$$

(4)

${\rho }^{\ast }\in [0, 1]$ denotes CEE, a higher value indicates greater CEE, and a lower value reflects reduced CEE. S denotes the relaxation variables of inputs and outputs, and $\lambda$ serves as a representation of the weight vector. When $S^{-}=0, S^g=0, S^b=0$ meet, the decision-making unit achieves absolute efficiency in terms of input and output.

Based on the classification standards for CO₂ emission efficiency established by Ning et al. (2021) [35], the measured values are categorized as follows: values in the range [0, 0.3] are deemed ineffective, (0.3, 0.6] are classified as weakly effective, and (0.6, 1) indicate that a city is strongly effective in CEE. The calculated CEE is thus divided into three distinct categories: ineffective, weakly effective, and strongly effective. Subsequently, ArcGIS 10.8 was employed to visualize the CEE of 269 cities for the years 2010, 2014, 2017, and 2021 (Figure 2), facilitating an assessment of the trends in CEE across these cities.

Figure 2. Carbon emission efficiency distribution maps for different years.

Overall, most cities have progressed from the ineffective stage to the weakly effective stage of CEE. This improvement is reflected in the declining share of cities classified as ineffective and the rising proportion of those considered weakly effective. Marked regional disparities in CEE remain evident, with cities in eastern China consistently achieving higher efficiency compared to their central and western counterparts. At the provincial level, capital cities demonstrate superior CEE relative to non-capital cities. In summary, the distribution of cities across the ineffective and strongly effective stages remains highly uneven, highlighting substantial differences in CEE among urban areas nationwide.

Kernel density estimation is a widely used non-parametric method for describing the dynamic distribution of data, which is especially useful when examining datasets with uneven or skewed distributions [62]. The positioning of the kernel density curve represents CEE levels, while its height and breadth provide insights into regional coordination and variability. Specifically, the peak’s elevation signifies the degree of data polarization, whereas the curve’s spread illustrates the extent of regional differences [63]. Let f(w) represent the density function of CEE; thus, we can express it as follows:

$$\mathrm{f}(\mathrm{w})={\displaystyle \frac{1}{nh}}{\displaystyle \sum _{i=1}^{n}K}({\displaystyle \frac{w_i-\overline{w})}{h}})$$

(5)

K(w) denotes the kernel density function, n represents the observed value, h represents bandwidth, and $w_i$ refers to independently distributed observations with $\overline{w}$ as the mean. This study primarily employs the Gaussian kernel density to generate the CEE kernel density curve. By analyzing the curve’s position and shape, insights into the dynamic distribution and evolution of regional disparities are revealed.

To further analyze the distribution of CEE over the years, the kernel density estimation was conducted using the default Epanechnikov kernel function in the Stata 18.0 software. This analysis generated a CEE density map for 269 cities spanning from 2010 to 2021 (Figure 3). The dynamic shifts in CEE levels among these cities were observed during the study period. Notably, the kernel density curve exhibited a progressive shift to the right, indicating a year-on-year increase in CEE across Chinese cities. Concurrently, the peak value of the curve decreased while its width increased, suggesting a reduction in the polarization characteristics of CEE, although these characteristics remain evident. Additionally, regional disparities and spatial inequities in urban CEE have gradually expanded. While the kernel density curve has not displayed multi-peaked characteristics, the right tail has extended, indicating a decrease in the proportion of cities exhibiting low CEE and an improvement in the aggregation of low CEE. Nevertheless, disparities in CEE persist among cities, with a limited proportion demonstrating strong efficiency, highlighting the continued presence of significant spatial differences in CEE across the urban landscape.

Figure 3. Kernel density estimation of urban carbon emission efficiency.

4.2. Baseline Regression

Table 4 reports the estimated direct effect of the BC and LC dual-pilot policy on CEE using an SDID model that accounts for both individual and temporal effects. In column (1), after including control variables, the regression coefficient rises to 0.505, exceeding the coefficient in column (2), and remains statistically significant at the 1% level. The coefficient of the dual-pilot policy on CEE is 0.0406, also significant at the 1% level, suggesting that the policy effectively increases CEE by 4.06%, thus supporting Hypothesis 1. To evaluate the synergistic effects of the dual-pilot policy, this study follows the testing procedures proposed by Guo and Ma (2023) [55]. In column (3), we exclude the sample of cities that have implemented the BC pilot policy, retaining only those that have adopted the LC pilot policy. In column (4), we focus solely on the experimental samples associated with the BC pilot policy, which serve as the control group in (2) for the regression coefficient analysis. The regression coefficient for the LC pilot policy on CEE in column (3) is 0.0267, which is statistically significant at the 1% level. In column (4), the regression coefficient for the BC pilot policy on CEE is 0.013; however, this does not achieve statistical significance and is smaller than the coefficients observed under the dual-pilot policy.

Table 4. Regression results of dual pilot policy on carbon emission efficiency.

Variable	Dual-Pilot Policy	Dual-Pilot Policy	LC	BC
	CEE	CEE	CEE	CEE
	(1)	(2)	(3)	(4)
Treat*Post	0.0505 ***	0.0406 ***	0.0267 ***	0.013
	(7.93)	(3.130)	(2.700)	(1.210)
rjsr	-	0.0643 ***	0.0639 ***	0.0638 ***
		(2.640)	(2.670)	(2.620)
jy	-	−0.216 ***	−0.221 ***	−0.208 **
		(−2.71)	(−2.72)	(−2.59)
dwkf	-	−0.256 **	−0.272 **	−0.298 **
		(-2.05)	(−2.12)	(−2.31)
zffd	-	0.003	0.003	0.003
		(1.430)	(1.320)	(1.440)
zfgy	-	−0.386 ***	−0.391 ***	−0.392 ***
		(-5.74)	(−5.79)	(−5.81)
rkjj	-	0.904 ***	0.995 ***	1.054 ***
		(2.860)	(2.950)	(3.010)
kyttr	-	0.085	0.118	0.133
		(0.550)	(0.730)	(0.800)
Constant	0.289 ***	0.310	0.312	0.311
	(209.72)	(−1.14)	(−1.16)	(−1.14)
Year	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes
Control	No	Yes	Yes	Yes
R2	0.768	0.791	0.789	0.768
N	3228	3228	3228	3228

Note: t statistics in parentheses, *, **, *** means that they passed the significance test at the 10%, 5%, and 1% levels, respectively, and the same below.

These findings indicate that the dual-pilot policy has a more pronounced effect on enhancing CEE compared to the individual policies. Consequently, it can be concluded that the BC and LC policies exhibit a synergistic effect in promoting improvements in CEE. The regression results align with the conclusions drawn in previous research conducted by Yan et al. (2023) [23].

4.3. Robustness Tests

4.3.1. Parallel Trend Test

The application of the SDID necessitates adherence to the assumption that both the control and experimental groups exhibit parallel trends [64]. This assumption is critical for ensuring the unbiased nature of the model, implying that, prior to the implementation of the dual-pilot policy, the CEE of each city must demonstrate a similar and stable trend. This condition is essential to mitigate the influence of confounding factors on the experimental results. Currently, the event study approach is increasingly employed to validate this model. Following the parallel trend testing methodology outlined by Jacobson et al. (1992) [65], the model can be structured as follows:

$${\mathrm{CEE}}_{it}={\alpha }_3+{\displaystyle \sum _{t=5}^{11}{\delta }_t}{D}_{it}+{\displaystyle \sum _{k=1}^n{\lambda }_3}Contro{l}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{it}$$

(6)

$D_{it}$ represents dummy variables; if the experimental group of cities has implemented both the BC pilot policy and the LC pilot policy within the year, the value is 1, and if not, it is 0. The parameter ${\sigma }_t$ denotes the variation in CEE between the experimental group and the control group during year t of policy implementation, with other variables aligning with those in the baseline model.

Figure 4 illustrates the relationship between the relative timing of policy implementation (horizontal axis) and the estimated coefficients of the main explanatory variable (vertical axis). The BC initiative began in 2014, while the LC pilot program was launched in July 2010. Consequently, the dual-pilot policy is considered to have been implemented in 2014 (relative time 0), establishing this year as the reference point. Prior to implementation (relative time −4 to 0), trends in CEE for both the experimental and control groups remained parallel, with regression coefficients not statistically significant, confirming the validity of the parallel trends assumption and supporting the causal identification in the SDID framework. Following policy implementation (relative time 1 to 7), the regression coefficients gradually increased, becoming statistically significant in the fifth-year post-implementation (relative time 5). This indicates that the full synergistic effects of the BC and LC policies emerged over time, reflecting the accumulation and release of policy interactions within a complex economic and technological context. The BC policy enhanced information infrastructure, fostering digitalization and informatization, which created favorable conditions for the LC policy’s success. For example, widespread adoption of information technology can improve energy management, optimize carbon emission monitoring in transport and industry, and accelerate the deployment of smart energy-saving technologies. However, this synergy did not appear immediately. The expansion of digital infrastructure and technological absorption requires a gradual process, while adjustments to energy structures and the promotion of energy-efficient technologies under the LC policy also need time to overcome institutional and economic constraints. As a result, the significance of the combined policy effects emerged in the fifth year, marking the transition from isolated effects to synergistic impacts. Overall, the estimation results satisfy the parallel trends assumption, supporting the robustness of the SDID findings.

Figure 4. Parallel trends test.

4.3.2. Placebo Testing

When examining the relationship between the dual-pilot policy and CEE, it is crucial to mitigate the influence of random factors on policy evaluation [66]. Specifically, it is essential to demonstrate that improvements in CEE are attributable to the dual pilot policy, while any effects on control group cities arise randomly. This approach ensures the authenticity and objectivity of the experimental findings. Building on the methodological framework proposed by Cai X, Lu Y, Wu M, et al. (2016) [67], a placebo test was conducted involving a fictitious treatment group and policy timeline. For this test, 49 cities were randomly assigned to the pseudo-treatment group, while 120 cities were designated as the pseudo-control group. The regression estimation coefficients, standard biases, and p-values were subsequently analyzed, with the entire experimental process repeated 500 times to generate Figure 5.

Figure 5. Placebo test result.

In the placebo test depicted in Figure 5, the horizontal axis represents the experimental estimation coefficients, while the vertical axis indicates the p-values. The results of the repeated experiments demonstrate that the coefficients approximate a normal distribution, with the regression coefficients clustering around zero. It is worth noting that most of the regression coefficients have p-values greater than 0.1, indicating that these results are not statistically significant. Moreover, the peak position of the coefficients is lower than the benchmark regression coefficient of 0.0406. This finding suggests that the observed relationship between the dual pilot policy and CEE is a low-probability event within the context of the placebo test, thereby indicating that the regression results derived from the SDID approach are minimally influenced by random factors. Consequently, the findings from the SDID regression analysis of CEE under the dual-pilot policy can be considered robust.

4.3.3. PSM-DID

The SDID method, often referred to as a “quasi-natural experiment,” builds upon the framework of natural experiments. However, sample selection bias may arise during the experimental process, potentially compromising the robustness of the results. Specifically, the selection of pilot cities is not random, leading to inherent differences between the control and experimental groups both prior to and following the implementation of the policy. As an explanatory variable, CEE may not significantly influence the implementation of the dual-pilot policy, thereby mitigating the endogenous bias associated with reverse causality. This study examines a sample comprising 269 cities, which provides a substantial dataset characterized by significant variations in economic and environmental conditions across different locations. To address potential endogenous selection bias, the study employs a Propensity Score Matching Difference-in-Difference (PSM-DID) approach [68].

Given the absence of randomization in quasi-natural experiments, self-selection bias may inevitably arise. To address this issue, the Propensity Score Matching (PSM) technique is employed to construct an appropriate control group for each treated unit, effectively converting the quasi-natural setting into one that approximates random assignment. This approach enhances the comparability between the treatment and control samples, allowing the analysis to more accurately isolate policy effects. Given that PSM requires a substantial sample size, this study incorporates 3228 samples, meeting the necessary criteria for PSM. By utilizing the PSM-DID technique, this research aims to rigorously assess the relationship between the dual-pilot policy and CEE, thereby alleviating potential selection biases in the classification of cities under the dual-pilot policy. In addressing the endogeneity problem associated with sample selection bias, the PSM-DID method was employed to regress Model (1). In this regression analysis, the radius matching method, kernel matching method, and nearest neighbor matching method were specifically applied [20]. The regression results are presented in Table 5. Across all model specifications, the coefficients of the interaction term “Treat*Post” remain significantly positive, consistent with the benchmark regression findings. This alignment underscores the robustness and reliability of the baseline results, demonstrating that the dual-pilot policy has a meaningful and positive impact on CEE.

Table 5. PSM-DID Test Results.

Variable	Radius Matching	Nuclear Matching	Nearest Neighbor Matching
	CEE	CEE	CEE
Treat*Post	0.0406 ***	0.0406 ***	0.0375 ***
	(6.61)	(6.61)	(5.99)
Constant	−0.310 ***	−0.310 ***	−0.295 ***
	(−2.85)	(−2.85)	(−2.63)
Year	Yes	Yes	Yes
City	Yes	Yes	Yes
Control	Yes	Yes	Yes
F	50.08	50.08	45
R2	0.791	0.791	0.794
N	3228	3228	3228

4.3.4. Endogeneity Test

While the SDID approach is effective in mitigating endogeneity concerns, it assumes that pilot cities are selected entirely at random. In reality, this assumption rarely holds, as the designation of pilot areas typically involves a comprehensive evaluation of local socioeconomic and developmental conditions to ensure policy feasibility. Such pragmatic selection criteria, however, may compromise the objectivity of experimental inference. While the instrumental variable (IV) approach can help mitigate endogeneity concerns, selecting a valid IV requires meeting two stringent criteria: it must be correlated with the endogenous regressor and simultaneously uncorrelated with the error term [69]. Building on the findings of Guo & Ma (2023) [55], regions characterized by relatively flat terrain generally incur lower construction and maintenance costs for broadband infrastructure. Improved broadband accessibility and network quality tend to attract population inflows and stimulate economic activity, both of which elevate carbon emissions and strengthen local demand for low-carbon development. As a result, regions with advantageous topographic conditions are more likely to be designated as dual-pilot cities [70,71]. Unlike regulatory policies, the BC and the LC policies aim to stimulate intrinsic industry motivation and enhance regional enthusiasm for digital and low-carbon transformation. Consequently, this study utilizes topographic relief as an instrumental variable to analyze the impact of dual-pilot policy on CEE. Column (1) and column (2) in Table 6 present the estimation results from a two-stage instrumental variables analysis employing the two-stage least squares method. In the first stage, the regression coefficient for regional fluctuation, utilized as an instrumental variable, is 0.0193, which passes the significance test at the 1% level. Additionally, the F-statistic is 21.71, exceeding the threshold of 10, indicating that weak identification is not a concern, and thus satisfying the relevance criteria for instrumental variable selection. Consequently, the choice of instrumental variable is deemed appropriate. After addressing the endogeneity issue, the findings indicate that the implementation of the dual-pilot policy continues to facilitate improvements in CEE. Furthermore, the sample selection bias does not compromise the experimental results of this study, thereby reinforcing the robustness of the conclusion that the dual-pilot policy positively influences CEE.

Table 6. Endogeneity treatment results.

Variable	Tool Variables		Reject the Sample
	(1)	(2)	(3)
	Phase 1	Phase 2	CEE
	Treat*Post	CEE
qfd	0.0193 ***	-	-
	(2.63)
Treat*Post	-	0.023 ***	0.0223 ***
		(3.00)	(3.41)
Constant	−1.349 ***	3.886 ***	−0.322 ***
	−3.19	−1.84	(−3.06)
F	21.71	25.51	40.59
Year	Yes	Yes	Yes
City	Yes	Yes	Yes
Control	Yes	Yes	Yes
N	3228	3228	3000
Adj R2	0.194	0.891	0.799

The designation of pilot cities follows a deliberate selection process rather than random assignment, reflecting a comprehensive evaluation of each city’s socioeconomic and developmental conditions. Municipalities decide whether to submit pilot applications to the National Development and Reform Commission (NDRC) based on their local development priorities and institutional readiness [72]. To minimize potential endogeneity arising from reverse causality, this study adopts the approach of Yu (2017) by excluding central cities [73]—such as municipalities, provincial capitals, and sub-provincial administrative units—from the sample, focusing instead on prefecture-level cities. As shown in column (3) of Table 6, the estimated coefficient for the dual-pilot policy is 0.0223 and remains statistically significant at the 1% level. These results align with the baseline regression findings, offering additional support that the dual-pilot initiative positively and consistently enhances CEE in peripheral cities. This consistency further strengthens the reliability and validity of the benchmark estimates.

4.4. Difference in Difference in Difference

Nonetheless, the potential for reverse causality within the DID framework may lead to endogeneity concerns. In the context of the SDID model, the estimated impact of the dual-pilot policy on CEE could be affected by factors such as industrial composition or the intensity of local environmental regulations. Even so, the DID approach helps to account for unobservable heterogeneity between the treatment and control groups, thereby enhancing the reliability of the estimation results [74]. To tackle these issues, and in line with the method proposed by Ren et al. (2019) [75], this study introduces a grouping variable to construct a DDD model, designed to examine the effects of the dual-pilot policy on CEE in various cities.

$${\mathrm{CEE}}_{it}={\beta }_0+{\beta }_{1}Trea{t}_{it}\ast Pos{t}_{it}\ast Group+{\beta }_{2}Trea{t}_{it}\ast Pos{t}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{it}$$

(7)

In model (7), as a grouping variable, Group represents the type of industry and the intensity of environmental regulation to be selected by SDDD below. If a sample city meets the specified conditions, the variable takes a value of 1; otherwise, it is 0, and all other variables remain consistent with those in model (1). In this case, the double coefficient ${\beta }_2$ represents the average effect of the dual-pilot policy on the CEE of all sample cities, and the triple term coefficient ${\beta }_1$ represents the difference between the average promotion effect of the dual-pilot policy on the CEE of the cities in the group and the average effect of all cities in the sample, and (${\beta }_1+{\beta }_2$) represents the net impact of the dual-pilot policy on the CEE of sample cities within the group.

4.4.1. SDDD in Industry Type

To further examine how the dual-pilot policy influences CEE across cities with distinct industrial structures, this study employs industry type as a grouping variable. This classification helps to narrow the heterogeneity between the treatment and control groups. The results of the stepwise triple-difference regression based on industrial categories are summarized in Table 7.

Table 7. Industry-led triple difference.

Variable	The Primary Sector Is Dominant		The Secondary Industry Is Dominant		The Tertiary Industry Is Dominant
	(1)	(2)	(3)	(4)	(5)	(6)
Treat*Post	0.0505 ***	0.0582 ***	0.0318 ***	0.0343 ***	0.0599 ***	0.0673 ***
	(7.08)	(7.80)	(3.63)	(3.62)	(7.55)	(8.12)
Treat*Post*Group	0.0026	0.0037	0.0304 ***	0.0373 ***	0.0309 ***	0.0303 **
	(0.14)	(0.19)	(3.38)	(4.08)	(2.69)	(−2.51)
Constant	−0.320 ***	0.289 ***	−0.322 ***	0.289 ***	−0.346 ***	0.290 ***
	(−2.93)	(214.99)	(−2.96)	(215.60)	(−3.17)	(217.85)
Control	Yes	No	Yes	No	Yes	No
Year	Yes	Yes	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes	Yes	Yes
F	45.30	30.42	46.97	38.89	46.22	33.61
R2	0.790	0.770	0.793	0.769	0.792	0.768
N	3228	3228	3228	3228	3228	3228

In Table 7, the estimated coefficients of the interaction term Treat*Post in the first row are significantly positive, suggesting that the implementation of the dual-pilot policy substantially enhances CEE across all industrial sectors, with improvements ranging from 3% to 7%. These findings are consistent with the benchmark regression results. The coefficient of the triple interaction term Treat*Post*Group captures the difference in the policy’s effect on CEE between specific industry groups and the overall industrial average. The sum of both coefficients represents the total marginal impact of the dual-pilot policy on targeted sectors. The results reveal that the dual-pilot initiatives significantly strengthen CEE in both the secondary and tertiary industries, with the highest enhancement reaching 8.99%. This effect exceeds that of models without control variables and remains consistent with the benchmark estimation outcomes.

4.4.2. SDDD in Environmental Regulation

Environmental regulation primarily aims to govern actions that cause public environmental harm by converting the societal costs of pollution into private costs. The stronger the environmental regulation, the more effective the incentives for pollution control, thereby enhancing the competitive advantages of environmentally conscious practices. However, excessive regulation may also impose substantial costs on firms, leading to challenges in balancing regulatory costs with economic growth. This tension creates a significant hurdle for companies striving for sustainable development while maintaining stable operations [76]. This study further examines whether the impact of the dual-pilot policy on improving CEE varies with differences in the intensity of environmental regulation across cities. This paper conducts SDDD analysis, stratifying sample cities based on the strength of their environmental regulations. This approach allows us to determine whether cities with varying levels of regulatory stringency experience differential effects from the dual pilot policy.

To assess the intensity of environmental regulation, this study adopts the methodological framework established by Zhang and Chen (2021) [76] and Chen and Chen (2018) [77]. Specifically, the frequency of “environmental protection”-related expressions in municipal government work reports is quantified, and their proportion relative to the total word count is used as a proxy indicator of regulatory intensity. The search terms encompass “green”, “low-carbon”, “pollutant”, “ecological”, “chemical oxygen demand”, and “environmental protection”, among others. Government work reports provide a comprehensive reflection of a city’s emphasis on environmental protection, its governance intensity, and the overarching framework of its environmental policies. Based on the environmental regulation intensity data from 269 cities between 2010 and 2021, we use the average intensity of environmental regulation as the threshold to categorize cities. These categories form the basis for the SDDD grouping variables, which are then incorporated into the model (7). The findings from this analysis are detailed in Table 8.

Table 8. Environmental regulation triple difference.

Variable	Strong Intensity of Environmental Regulation		Weak Intensity of Environmental Regulation
	(1)	(2)	(3)	(4)
Treat*Post	0.0495 ***	0.0566 ***	0.0604 ***	0.0681 ***
	(6.94)	(7.59)	(6.95)	(7.49)
Treat*Post*Group	0.00804 ***	0.0112 ***	−0.0194 **	−0.0195 *
	(2.86)	(3.82)	(−1.99)	(−1.90)
Constant	−0.307 ***	0.283 ***	0.0660 ***	0.289 ***
	(−2.82)	(135.95)	(6.85)	(217.75)
Control	Yes	No	Yes	No
Year	Yes	Yes	Yes	Yes
City	Yes	Yes	Yes	Yes
F	46.34	37.85	45.80	32.25
R2	0.792	0.769	0.792	0.768
N	3228	3228	3228	3228

The results in Table 8 indicate that the dual-pilot policy has a significant positive effect on CEE in cities with varying intensities of environmental regulation, consistent with the baseline regression findings. However, in cities with stronger environmental regulation, the dual-pilot policy improves CEE more effectively, achieving a promotion rate of 5.754%. In contrast, the promotion effect in cities with weaker environmental regulation reaches 4.1%, which is lower than the average policy effect observed across all pilot cities. Columns (2) and (4) present the regression results without control variables, and their coefficients are higher than those in columns (1) and (3), where control variables are included. This outcome aligns with the benchmark regression, further supporting the robustness of the baseline results. Scholars have drawn various conclusions about the impact of environmental regulation on ecological outcomes. Shao et al. (2023) found an inverted U-shaped relationship between environmental regulation and provincial ecological efficiency, with notable spatial heterogeneity [78]. Other research suggests that environmental regulation can accelerate advancements in energy efficiency [56], and the effectiveness of different types of environmental regulations varies [79]; these findings align with the conclusions of this paper.

4.5. Mechanism Analysis

Following the mainstream methods for testing mediating effects, and drawing on the approach outlined by Wen et al. (2004) [80] and the testing steps proposed by Shi et al., (2018) [81], a three-step model was constructed to examine the mechanism discussed in this paper.

$${\mathrm{CEE}}_{it}={\beta }_0+{\beta }_{1}Trea{t}_{it}Pos{t}_{it}+{\beta }_{2}Contro{l}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{it}$$

(8)

$$M_{it}={\lambda }_0+{\lambda }_{1}Trea{t}_{it}Pos{t}_{it}+{\lambda }_{2}Contro{l}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{it}$$

(9)

$${\mathrm{CEE}}_{it}={\gamma }_0+{\gamma }_{1}Trea{t}_{it}Pos{t}_{it}+{\gamma }_2{M}_{it}+{\gamma }_{3}Contro{l}_{it}+{\mu }_i+{\sigma }_t+{\epsilon }_{ii}$$

(10)

The implementation of the dual-pilot policy was taken as a quasi-natural experiment. The mediating variables were included in the above model at the same time, and the existence of the mediation effect was judged by observation of ${\beta }_1$, ${\lambda }_1$, ${\gamma }_1$, and ${\gamma }_2$, where $M_{it}$ was the mediator variable, the rest of the variables like the explanatory variables, explanatory variables, and control variables were the same as those of the benchmark regression model (1) described above.

Under the dual-carbon objectives, the rollout of the dual-pilot policy is anticipated to expedite the phase-out of enterprises characterized by high energy intensity, pollution, and emissions, sectors that are largely concentrated in secondary industries. To sustain their market competitiveness, firms are compelled to pursue low-carbon development and digital transformation strategies. Simultaneously, the expansion of the digital economy, indicated by the rising value added in the tertiary sector, reinforces this transformation in industrial structure. This study follows the methodology adopted in previous research [51] and measures industrial structure by calculating the ratio of the tertiary sector’s value added to that of the secondary sector in each city. As shown in Table 9, the regression coefficient of the dual-pilot policy on industrial structure in column (2) is 0.0528, and it passes the 1% significance level test. This finding implies that the dual-pilot initiative significantly fosters the upgrading and transformation of urban industrial structures. In column (3), both the coefficients of the dual-pilot policy and industrial structure are positive and statistically significant, indicating a partial mediating effect with a mediation ratio of 0.02. This result suggests that the dual-pilot policy enhances CEE indirectly through industrial restructuring, thereby confirming Hypothesis 2. Overall, the upgrading of industrial structure not only acts as a statistical mediator but also represents a tangible economic mechanism through which the policy improves carbon emission efficiency. Specifically, the transition toward service-oriented and knowledge-intensive sectors contributes to higher total factor productivity, improved energy allocation among industries, and reduced marginal abatement costs, collectively advancing both energy conservation and emission reduction objectives.

Table 9. Mediation effect examination.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	CEE	cyjg	CEE	lsfm	CEE	lssyxx	CEE
Treat*Post	0.0406 ***	0.0528 ***	0.0398 ***	0.0835 ***	0.0379 ***	0.047 ***	0.0363 ***
	(3.130)	(1.500)	(6.470)	(0.220)	(6.190)	(4.390)	(5.960)
cyjg	-	-	0.0154 **	-	-	-	-
			(2.05)
lsfm	-	-	-	-	0.0318 ***	-	-
					(5.59)
lssyxx	-	-	-	-	-	-	0.0925 ***
							(8.85)
rjsr	0.0643 ***	−0.328 ***	0.0592 ***	0.216 ***	0.0711 ***	0.124 ***	0.0757 ***
	(2.64)	(−13.85)	(5.96)	(6.94)	(7.36)	(7.38)	(7.89)
jy	−0.216 ***	0.08	−0.215 ***	0.23	−0.209 ***	0.226 ***	−0.195 ***
	(−2.71)	(0.77)	(−4.89)	(1.59)	(−4.77)	(2.95)	(−4.49)
dwkf	−0.256 **	0.4280 **	−0.249 ***	0.20	−0.249 ***	0.03	−0.253 ***
	(−2.05)	(2.04)	(−2.92)	(0.73)	(−2.94)	(0.21)	(−3.00)
zffd	0.00	0.0085 **	0.0033 **	−0.0113 **	0.00281 *	0.00	0.00321 **
	(1.43)	(2.28)	(2.17)	(−2.3)	(1.85)	(0.15)	(2.13)
zfgy	−0.386 ***	0.28 ***	−0.382 ***	0.414 ***	−0.373 ***	0.465 ***	−0.343 ***
	(−5.74)	(3.40)	(−11.39)	(3.83)	(−11.17)	(7.97)	(−10.27)
rkjj	0.904 ***	−0.841 *	0.891 ***	(0.85)	0.877 ***	−0.847 **	0.825 ***
	(2.86)	(−1.81)	(4.71)	(−1.39)	(4.66)	(−2.57)	(4.42)
kyttr	0.09	(0.04)	0.08	0.702 *	0.11	(0.08)	0.08
	(0.55)	(−0.13)	(0.66)	(1.71)	(0.85)	(−0.36)	(0.62)
Constant	(0.31)	3.379 ***	−0.258 **	−1.137 ***	−0.346 ***	0.24	−0.288 ***
	(−1.14)	(12.63)	(−2.31)	(−3.24)	(−3.19)	(1.24)	(−2.68)
R2	0.79	0.91	0.79	0.84	0.79	0.88	0.80
F	50.08	96.14	44.55	12.49	47.91	19.39	54.76
N	3228.00	3228.00	3228.00	3228.00	3228.00	3228.00	3228.00

Technological innovation is widely recognized as a fundamental driver of high-quality economic growth, with green technological innovation playing a particularly critical role in restructuring energy systems and improving energy efficiency [22]. Green technology emphasizes the conservation of resources and energy, and the LC pilot policy promotes such innovation through two primary mechanisms: cost reduction and innovation compensation [82]. Furthermore, the expansion of broadband infrastructure facilitates the spatial diffusion of technology, strengthening regional innovation capabilities. The combined influence of green technologies and digital infrastructure significantly accelerates the transition toward green and low-carbon urban development. In this study, green technological innovation is categorized into two types: inventive green technology innovation and improved green technology innovation [54]. The regression outcomes for these patent categories are presented in Table 9. As shown in columns (4) and (6), the coefficients for green invention patents and green utility model patents under the dual-pilot policy are 0.0835 and 0.047, respectively, both statistically significant at the 1% level. Columns (5) and (7) further show that the interaction terms between green technological innovation and the dual-pilot policy are positive and statistically significant, confirming a mediating effect. The mediation ratios are 0.07 and 0.107, suggesting that the dual-pilot policy enhances CEE not only through inventive green patents but also via improvements in green technological capabilities. These findings imply that cities implementing the BC and LC dual-pilot initiatives can effectively improve CEE and facilitate low-carbon transformation by strengthening green technological innovation. This evidence supports both Hypothesis 3 and, indirectly, Hypothesis 1. Importantly, the mediating role of green technology innovation extends beyond patent activity to tangible outcomes in energy efficiency and emissions performance. Cities with higher levels of green innovation exhibit lower carbon intensity, demonstrating that technological progress translates into measurable improvements in CEE.

In summary, industrial upgrading and green technological innovation serve as complementary yet distinct channels through which the dual-pilot policy enhances CEE. Industrial upgrading improves the allocation and utilization of resources, thereby boosting CEE, whereas green technological innovation drives efficiency gains via technological development and dissemination. Together, these mechanisms demonstrate that the dual-pilot policy has sustained, multidimensional effects on CEE.

4.6. Heterogeneity Test

4.6.1. Geographical Location

China exhibits substantial regional disparities in terms of land area, population distribution, economic development, and industrialization, which are accompanied by notable differences in energy consumption patterns and the advancement of the digital economy across regions [49]. Following the framework of the Seventh Five-Year Plan for National Economic and Social Development of the People’s Republic of China, the 269 cities in the sample are classified into the eastern, central, and western regions based on their geographic location. These regional classifications are incorporated into Model (1) to investigate whether geographic heterogeneity affects the effectiveness of the dual-pilot policy in enhancing CEE.

As shown in Table 10, the estimated effect of the dual-pilot policy in the central region under Model (1) is 0.0409, which is statistically significant at the 1% level. In contrast, Model (2) indicates that the coefficients for western regions are not statistically significant. For the eastern region, Model (3) reports a regression coefficient of 0.0492, also significant at the 1% level. This coefficient exceeds the 0.0409 observed for the central region, implying that the BC and LC dual-pilot initiative has a stronger positive effect on CEE in the eastern region.

Table 10. Regression results for different geographic locations.

Variable	(1)	(2)	(3)
	Central	Western	Eastern
	CEE	CEE	CEE
Treat*Post	0.0409 ***	0.007	0.0492 ***
	(4.890)	(0.470)	(5.200)
Constant	−0.547 ***	−1.900 ***	0.278 *
	(−2.96)	(−7.63)	(1.810)
Controls	Yes	Yes	Yes
R2	0.830	0.777	0.806
N	1176	900	1152

The eastern region has made significant strides in fostering a resource-efficient and environmentally sustainable society, with green coverage rates in first-tier and new first-tier cities surpassing 40%. This region has reached the “late stage of industrialization,” characterized by a relatively advanced industrial structure and higher energy consumption patterns, attracting substantial foreign investment [83]. Economic expansion in the east is primarily driven by the tertiary sector and high-end manufacturing industries, leveraging its advantages in capital, skilled labor, and technological resources built through sustained development. The relatively mature industrial structure allows the eastern region to better balance economic growth with environmental protection, thereby facilitating the effective implementation of BC and LC dual-pilot policy [23].

4.6.2. City Size

Following the guidelines outlined in the 2014 Notice on Adjusting the Classification Standards for City Sizes issued by the State Council, this study categorized 169 cities in the sample based on population size. Cities with fewer than 1 million residents were defined as small- and medium-sized cities, those with populations between 1 million and 5 million as large cities, and those exceeding 5 million as megacities.

The regression results are shown in Table 11. For small and medium-sized cities, the estimated coefficient of the dual-pilot policy is −0.149, significant at the 1% level. In large cities, the coefficient is 0.0167, passing the 5% significance threshold, while for megacities, the coefficient is 0.0519, significant at the 1% level. Notably, the positive effect in megacities (0.0519) surpasses that in large cities (0.0167), indicating that the impact of the BC and LC dual-pilot policy on CEE strengthens with city size. Larger cities typically possess more advanced digital infrastructure, stronger technological capabilities, and higher levels of capital and skilled labor, all of which facilitate the effective implementation of both BC and LC. Furthermore, megacities typically possess more diversified and higher value-added industrial structures, which facilitates the integration of low-carbon innovations and green technologies. In contrast, small- and medium-sized cities often encounter constraints in technology adoption, financial capacity, and industrial upgrading, limiting the overall effectiveness of the dual-pilot policy.

Table 11. Regression results for different city sizes.

Variable	(1)	(2)	(3)
	Small- and Medium-Sized Cities	Metropolis	Megacities
	CEE	CEE	CEE
Treat*Post	−0.149 ***	0.0167 **	0.0519 ***
	(−4.05)	(2.440)	(4.420)
Constant	0.472	−0.882 ***	1.319 ***
	(0.680)	(−7.63)	(3.690)
Controls	Yes	Yes	Yes
R2	0.872	0.833	0.776
N	79	1649	844

5. Conclusions, Recommendations, and Limitations

This study examines the impact of the dual-pilot policy on CEE, with the main findings summarized as follows: (1) Overall, China’s CEE exhibits a slow but steady improvement, yet the absolute level remains relatively low. Significant regional disparities persist, and spatial imbalances appear to be widening. Although the proportion of cities with low CEE is gradually declining, the majority of cities still face substantial challenges in achieving higher carbon efficiency. (2) The dual-pilot policy (BC and LC) significantly enhances CEE, with a comparison of individual policy regressions showing that the dual-pilot approach has a more substantial positive effect on CEE, and the positive effects of the dual-pilot policy on CEE are significantly stronger in cities characterized by more advanced industrial structures and higher levels of environmental regulation. (3) This study confirms two mediating pathways through which the dual-pilot policy influences CEE: by driving industrial structure upgrades and by fostering green technology innovation. (4) The heterogeneity analysis explores the impact of the dual-pilot policy on CEE across natural resource endowments and city scales, highlighting stronger effects in eastern regions and larger cities.

Based on the above findings, this paper proposes the following policy recommendations: (1) Deepening the implementation of the dual-pilot policy. The study demonstrates that the dual-pilot policy of BC and LC has substantially contributed to CEE improvements in pilot cities, with the synergistic effects of the two policies proving more effective together than when implemented individually. To sustain and expand the impact of this policy, it is crucial for the government to increase support and encourage wider adoption, particularly among small- and medium-sized cities. This could be achieved through financial support, tax incentives, and policy guidance aimed at promoting active participation. Furthermore, establishing cross-regional collaboration mechanisms will allow for experience sharing and knowledge transfer among cities, thereby facilitating the replication of successful practices and ensuring the broader application of this policy. (2) Promoting green technological innovation. The BC policy strengthens urban digital infrastructure, which, in turn, supports green technological innovation. The LC policy incentivizes enterprises to adopt eco-friendly innovations through tax benefits and other supportive measures. Given that green innovation plays a pivotal role in enhancing CEE, it is imperative that the government fosters an environment conducive to technological advancements. This could include supporting research and development in key sectors such as energy, transportation, and manufacturing, and encouraging partnerships between academia, industry, and government. Additionally, enhancing financial incentives for green technology development and expanding market mechanisms to promote these technologies are critical steps to drive innovation and improve CEE across the nation. (3) Accelerating industrial structure optimization. This study highlights that optimizing industrial structures is central to improving CEE, particularly through the development of low-carbon industries and digital economies. The government should implement region-specific industrial policies that align with local characteristics. In more developed eastern regions, the focus should be on enhancing green industry clusters and optimizing industry chains, leveraging advanced infrastructure and technology. In contrast, central and western regions should prioritize the green transformation of traditional high-carbon industries by adopting digital technologies to drive decarbonization. A comprehensive, coordinated approach to industrial policy will be essential for facilitating the green transition across regions, supporting the simultaneous development of both high-carbon and low-carbon industries. (4) Strengthen dual-pilot policy implementation by targeting industry and regional characteristics. The analysis shows that the effectiveness of BC and LC policies differs across urban contexts. To maximize CEE improvements, policymakers should prioritize support for resource-intensive and industrially less-developed cities, focusing on enhancing digital infrastructure, facilitating industrial upgrading, and providing technical assistance to promote low-carbon and green transformation. In more developed cities and service-oriented industries, the emphasis should shift toward reinforcing innovation incentives, encouraging cross-sector collaboration, and scaling up best practices. By aligning policy efforts with local industrial and regional conditions, cities can better leverage the complementary effects of digitalization and low-carbon initiatives to achieve sustainable urban transitions.

Although this study offers an examination of the impacts of the BC and LC dual-pilot policy on CEE, several limitations remain. Firstly, due to constraints in data coverage and consistency, the study period is limited to 2010–2021. Future research should seek to incorporate more updated datasets to extend the temporal scope and better reflect the evolving effects of this policy. Secondly, the analysis primarily relies on city-level panel data and does not fully examine the differentiated responses of specific economic entities; thus, future directions should involve incorporating more precise industry-level or micro-enterprise-level data to identify the policy’s specific impacts on different sectors. Furthermore, this study only focused on industrial structure optimization and green technological innovation as mediating mechanisms. It is suggested that future research should broaden its scope to investigate other potential mediating factors, such as green finance, market-driven incentives, or consumer behavior.