Urbanization and the Bipolarization of Carbon Emission Efficiency Across Chinese Cities

Abstract

The commitment to peak carbon emissions before 2030 and the goal of achieving carbon neutrality by 2060 have placed carbon emission efficiency (CEE) at the center of China’s low-carbon development strategy. Although national policies have promoted energy conservation and technological upgrading, substantial heterogeneity in CEE persists across cities of different administrative and economic tiers. To examine this heterogeneity, we construct a city-level CEE index based on a stochastic frontier analysis (SFA) framework that explicitly treats CO₂ emissions as an undesirable output. Based on a panel dataset covering 274 prefecture-level cities grouped into five categories from 2006 to 2022, we find that first-tier cities such as Shanghai and Beijing, with advanced technological capacity and strong support by the emission trading scheme (ETS), have shown an upward trend in improving their CEE, while middle-range cities remain locked in carbon-intensive trajectories. In particular, the lowest-level fifth-tier cities show a decreasing trend, implying a bipolarization of urbanization among cities. To address this bipolarization by urbanization, we examine the governance mechanisms of CEE using a second-stage Tobit model and find that governance factors related to urbanization—such as high labor quality and intensive land development in larger cities—have contributed to the widening gap in CEE. This implies that mitigating the negative consequences of urbanization is essential for achieving a carbon-neutral economy.

1. Introduction

Achieving decarbonization while sustaining economic growth has emerged as one of the most pressing policy challenges of the twenty-first century. For emerging economies, and particularly for China, the world’s largest emitter of greenhouse gases and a central actor in global climate governance, the stakes are exceptionally high. In 2020, the Chinese government announced its dual-carbon strategy, pledging to peak national carbon emissions before 2030 and to achieve carbon neutrality by 2060. These ambitious targets require more than incremental reductions in absolute emissions, and at the same time call for fundamental improvements in the efficiency with which economic output is produced in environmentally sustainable ways. This dual challenge of carbon emission efficiency (CEE)—the ability to produce a given level of economic value while minimizing carbon dioxide emissions—has become a critical metric for evaluating the successful performance of low-carbon development policies [1].

The national mission to enhance CEE is reinforced by the pressure of rapidly growing urbanization and industrial transformation. Over the past two decades, Chinese cities have been the core engines of economic growth, accounting for the majority of national energy consumption and carbon emissions. Urban centers concentrate capital, labor, technology, and governance capacity, and thus play a decisive role in determining whether the country can achieve its dual-carbon targets. However, despite a national trend of improving environmental performance, significant heterogeneity persists across cities of different administrative and economic tiers [2]. Recent studies further show that a city’s position within the national urban hierarchy plays a fundamental role in shaping its carbon emission patterns, with higher-tier cities generally benefiting from stronger institutional and economic capacities [3]. Rapidly urbanizing metropolitan regions such as Beijing, Shanghai, Guangzhou, and Shenzhen have long been considered natural leaders in the green transition, given their advanced industrial bases, fiscal strength, and proactively environmentally friendly policy frameworks. The central government of China has promoted this selective concentration on these cities, even under the undesirable yet unavoidable pressure of urbanization. Nonetheless, empirical evidence reveals a striking paradox: the metropolitan cities have not consistently maintained superior carbon efficiency [2], while the small cities of the lowest level of economic development seem to remain locked in carbon-intensive development paths [4]. Such patterns challenge the conventional expectation of a simple positive relationship between development level and carbon performance, suggesting that institutional and market-based factors may jointly shape urban trajectories [2,4,5]. Some authors argue that this paradox may come from the non-linear relationship of urbanization [2], while others emphasize the role of intermediary mechanisms [6]. Therefore, our research question is whether the higher level of economic development results in better performance of the carbon emission abatement across diverse cities. If not, our focus may shift to the role of urbanization in the carbon emission mechanism.

To solve this research question, we shall focus on the governance factors of sustainable development. Among the potential drivers of these divergent outcomes of research, market-based environmental instruments, particularly a carbon emissions trading scheme (ETS), occupy a prominent position. In theory, carbon pricing should improve efficiency by internalizing the social cost of carbon, thereby creating continuous incentives for firms to adopt greener technologies, optimize energy-saving structures, and reallocate resources toward low-carbon production. In practice, however, the impact of carbon prices on city-level CEE is far from straightforward. Large metropolitan economies often face stricter carbon-accounting standards, higher compliance costs, and more intensive industrial restructuring pressures. These transitional frictions can temporarily suppress measured efficiency despite long-term benefits. In contrast, small and under-resourced cities frequently lack the administrative capacity, financial resources, and market-oriented support for meaningful efficiency improvements. ETS may not effectively work in these small cities, resulting in a practical bias when identifying the governance role of ETS in CEE. These complex dynamics indicate that the relationship between carbon prices and CEE is likely to be conditional rather than uniform, and thus key contextual variables—such as population scale, industrial composition, and governance capacity—may not moderate the effectiveness of market incentives [2,7]. Therefore, considering the heterogeneity at the city level, the ETS may not be a good governance factor of CEE.

In order to include this spatial heterogeneity into the research, a growing body of empirical research has examined spatial patterns and temporal trends in China’s urban carbon efficiency, but only a few existing studies showed that carbon market signals may not influence CEE well across heterogeneous urban contexts. Few studies have convincingly pointed out persistent spatial divergence in the industrial upgrading, technological progress, and environmental regulation in shaping efficiency outcomes [8]. Therefore, understanding whether the ETS price fails to interact well with city-specific features is critical for designing more effective carbon market mechanisms and for market-based environmental policies.

To address these gaps, this study develops a city-level carbon emission efficiency (CEE) index using a stochastic frontier analysis (SFA) framework based on the Shephard distance-functional type of environmental production technology. This methodological design explicitly treats CO₂ emissions as an undesirable output and accommodates the diverse production structures observed across Chinese cities. Building on these efficiency estimates, we construct a panel dataset spanning 2006 to 2022 to examine the categorical CEE based on economic development level and spatial size. A key innovation of our analysis is the differentiation of cities into categorical groups to identify the bipolarization of carbon emissions, with larger cities showing better performance over time and smaller cities showing lower performance. This paradox of bipolarization may come from the non-linear relationship between city development and carbon emissions, resulting in the paradox of urbanization. To examine the governance factors of CEE more precisely, we employ Tobit analysis on the carbon emission mechanism with urbanization at the second stage of the research.

The contributions of this research are threefold. First, by explicitly incorporating five categorical city groups into a city-level efficiency framework, the study provides direct evidence for the ever-increasing gap among the city groups in promoting low-carbon urban development. This approach goes beyond the conventional use of provincial or industrial-level proxies by capturing actual variations in the carbon price signals that individual cities face. The research shall show the increasing trend in the metropolitan city group, but their average is much lower than the nationwide average, implying that the ETS may not result in more workable solutions due to the intensive land use in the metropolitan cities coming from urbanization. Our efficiency measurement is based on the weighted average of economic growth and carbon emissions. If economic development is outperforming, then even with ETS policies, the average carbon efficiency may stay at a much higher level than in other city groups. If not, then it implies the lack of governance even with ETS in metropolitan cities.

Second, the introduction of the urbanization term sheds light on the moderating role of the governance mechanism on the carbon emission of each categorical city group, offering a new perspective on how urbanization factors result in the performance of the local economy. In larger cities, much higher human capital provided by higher education and research centers brings about better performance in the local economy, while dense industrial land development may decrease the carbon efficiency. Third, by focusing on tier-specific heterogeneity, the analysis generates policy-relevant insights for tailoring carbon market design to different types of cities, thereby improving the alignment between national climate goals and local development realities. Fourth, this research shall examine the effect of urbanization on environmentally friendly economic development. In theory, and in practice as well, it has been argued that urbanization is a critical contributor to the economic development of metropolitan cities with great networking economies of human capital and financial concentration. However, it is also criticized due to its extraction of the high-quality labor from the small cities, and it has resulted in the wider gap of bipolarization. Our research shall evaluate the role of urbanization on the carbon emission mechanism and find out the implications of urbanization for the low-carbon economy.

Beyond its empirical findings, this study also contributes to broader debates on sustainable urban transitions in emerging economies. Conceptually, the study advances the understanding of how urban hierarchy shapes the different evolution of carbon emission efficiency across cities, focusing on the role of urbanization. Building on this foundation, the paper makes three specific contributions. First, it introduces a tier-specific SFA framework that incorporates production heterogeneity with undesirable outputs, offering a more robust measurement of urban carbon emission efficiency. Second, by documenting the widening divergence between first-tier and fifth-tier cities, the study provides new empirical evidence of a structural bipolarization that has not been clearly identified in prior work. Third, the findings advance a differentiated governance perspective by showing that cities at different development stages require distinct policy tools, thereby helping to align national low-carbon goals with local institutional capacities.

2. Literature Review

The precise and appropriate measurement of CEE has increasingly become a critical research agenda in environmental economics, particularly under the dual imperatives of sustaining economic growth while addressing global climate change, especially with the utmost challenges of pressure from urbanization. Earlier cross-country analyses revealed substantial disparities in efficiency performance across different regions and stages of development, highlighting the complex interplay between economic progress and environmental sustainability. Jin and Kim (2019) showed that the carbon efficiency of emerging economies remained comparatively weak, emphasizing the necessity of distinguishing between economic and environmental outcomes [9]. Likewise, Herrala and Goel (2012), using a stochastic cost frontier for 170 countries, identified noticeable heterogeneity in CO₂ efficiency, demonstrating the analytical importance of frontier-based methods in assessing undesirable outputs [10]. Together, these studies provided the conceptual and methodological groundwork for the subsequent in-depth research model, because carbon efficiency cannot be adequately captured without explicitly integrating heterogenetic indicators into efficiency frameworks.

The evolution of methodologies has further enriched this body of literature. In its earlier stages, non-parametric approaches such as data envelopment analysis (DEA) were frequently employed for environmental efficiency evaluations due to their flexibility and minimal requirements for functional form specification. However, DEA-based analyses have been criticized for their vulnerability to noise, their inability to separate inefficiency from random shocks, and the biases generated in two-stage regression models. This has spurred an increase in the adoption of stochastic frontier analysis (SFA) and hybrid designs that combine parametric and non-parametric strengths. For instance, Sun et al. (2019) applied a single-stage SFA to estimate carbon emission efficiency in China and showed that modeling undesirable outputs within an SFA structure captures more reliable efficiency by separating random disturbances from true inefficiency, thereby avoiding the bias that often arises in multi-step DEA applications [1]. Yang (2024) took this approach by integrating super-efficiency SBM with SFA to distinguish random disturbances from true inefficiency, thereby producing more reliable estimates [5]. Rodríguez and Trujillo (2025), in their study of Spanish ports, highlighted that embedding CO₂ emissions directly into the efficiency frontier alters the production boundary fundamentally, offering a more realistic assessment of environmental performance [11]. Complementarily, Lu, Peng, and Lu (2022) employed a heterogeneous SFA framework to examine China’s energy industry chain, showing that distortions in factor markets and high carbon intensity were major obstacles to technical efficiency [6]. Yu and Zhang (2016) further demonstrated the strengths of SFA at the provincial level, incorporating urbanization and industrial indicators into their analysis and revealing substantial regional heterogeneity [12]. Collectively, these methodological advancements reinforce the rationale for employing SFA in the city-level studies of CEE, where random fluctuations, industrial diversity, and contextual heterogeneity must be carefully addressed.

Within the Chinese context, empirical research consistently documents substantial spatial and temporal heterogeneity in CEE. Liu et al. (2016) established a bidirectional causal relationship between urbanization and CO₂ emissions, demonstrating that more-developed provinces leveraged technological advances and policy frameworks to gradually improve efficiency, while less-developed provinces remained locked in energy-intensive industrial structures [13]. Similarly, Cai et al. (2019), using high-resolution gridded data across 280 cities, found persistent bipolarization, with coastal cities systematically outperforming inland and western counterparts [14]. More recently, Zhao, Li, and Duan (2023) documented a steady decline in average urban efficiency between 2006 and 2020, alongside strong spatial clustering of high- and low-efficiency cities [15]. Luo and Qu (2022) further employed spatial Markov chains and inequality decomposition to show that city-level trajectories are strongly shaped by spillover effects from neighboring regions [16]. Xing et al. (2024), integrating machine learning with an SBM-DDF model, confirmed that energy consumption, GDP, urban size, and population were decisive determinants of efficiency variation, while highlighting the regional heterogeneity of these effects [17]. Taken altogether, these findings emphasize the uneven landscape of urban carbon efficiency in China, both across space and over time, with coastal and higher-tier cities advancing more rapidly than their lower-tier counterparts. Nonetheless, most research did not examine the heterogeneous gap between cities, with the lack of governance on the regulatory policies. To fill this research gap, the role of urbanization should be utilized to differentiate the heterogenetic development level across the cities.

Urbanization and industrial restructuring emerge as recurring themes that shape these patterns. Tang and Hu (2023) demonstrated that land urbanization in its early phases significantly elevated emissions, but that over time an inverted U-shaped pattern emerged, with efficiency gains materializing once structural optimization and regulatory mechanisms became effective [18]. Similarly, Sun and Huang (2020) found a non-linear effect of urbanization on provincial efficiency, where rapid expansion initially depressed efficiency, but modernization eventually drove improvements once a certain developmental threshold was reached [2]. Zhang et al. (2024) classified Chinese cities into five categories with distinctive carbon-emission phases, showing that the first-tier cities are approaching or have already reached peak emissions, while many lower-tier cities remain trapped in emission-intensive growth trajectories [19].

Extending this perspective, Gu et al. (2025) analyzed the coordinated development of land and population urbanization, showing that their alignment can significantly enhance efficiency, though the magnitude and nature of the effect vary considerably by region [20]. Together, these studies suggest that urbanization level, industrial structure, and demographic trends constitute decisive factors shaping the long-run trajectories of CEE.

Policy frameworks and institutional interventions also exert a profound influence on efficiency outcomes. Yu et al. (2022) showed that the innovation city pilot program significantly improved carbon performance by stimulating green innovation, promoting modernization, and generating positive spillover effects in surrounding areas [21]. Jiang et al. (2022) emphasized that the impact of environmental regulation is conditional: while efficiency improves in provinces with low energy intensity and strong innovation capacity, regulations can exert disincentive effects in energy-intensive areas [22]. Kang et al. (2022) further analyzed the effects of regional integration, finding that integration can reduce efficiency gaps at early stages but, when exceeding certain thresholds, may exacerbate disparities [23]. Tu et al. (2024) explored the role of industrial intelligence, showing that advances in automation and digitalization substantially enhanced efficiency through green innovation and structural upgrading, while also producing positive spatial spillovers [24]. Collectively, these policy-oriented studies show that institutional interventions matter, but their effectiveness differs noticeably across cities because of variations in industrial structure, administrative capacity, and levels of technological development. Although this line of research has advanced our understanding, several issues remain insufficiently addressed.

First, much of the existing work relies on national or provincial comparisons, leaving limited evidence on how carbon efficiency differs across cities and between distinct urban tiers. Second, many studies employ single-stage efficiency measures, which are unable to reflect the simultaneous production of economic output and undesirable emissions. Third, while governance factors are frequently mentioned, their role is rarely examined within a unified analytical framework. These gaps point to the need for an approach that accounts for efficiency measurement together with urban heterogeneity and governance characteristics—an area to which the present study contributes.

Table 1. Summary of representative studies on carbon emission efficiency in China.

Author(s) (Year)	Field of Application/Region	First-Stage		Second-Stage
		Variables	Method	Variables	Method
Liu et al. (2016) [13]	114 resource-based cities in China (2006–2017)	Energy, labor, capital, and CO2 emissions	Super-SBM (undesirable outputs)	Urbanization rate, industrial structure	Tobit model
Zhang et al. (2024) [19]	Northern heavy-industry cities of China	Labor, capital, land, and CO2 emissions	DEA–Malmquist index	Urbanization, technological upgrading	Tobit model
Zhao, Li, and Duan (2023) [15]	30 Chinese provinces (2006–2020)	Energy, labor, and capital	SFA model	Energy intensity, industrial structure	Panel regression
Xing et al. (2024) [17]	286 prefecture-level cities (2003–2021)	Land, energy, and industrial output	Super-efficiency DEA	Land use intensity, technological progress	Regression model
Gu et al. (2025) [20]	30 provinces of China (2006–2020)	Energy, labor, capital, and CO2 emissions	Dynamic SFA	Economic growth, environmental regulation	Tobit model
This study	274 Chinese cities (2006–2022)	Energy, labor, capital, and undesirable CO2 emissions	SFA (translog frontier)	Carbon price × Population, land intensity, and universities	Tobit model

summarizes representative studies applying DEA and SFA models to measure CEE across different regions in China. Most of the previous works rely on single-stage efficiency estimation and/or Tobit or regression analyses at the second stage, focusing on urbanization or industrial structure. However, few studies explicitly consider heterogeneity across city tiers. By introducing urbanization as the moderating mechanism for heterogeneity at the second stage, this study extends the literature by examining how urban scale conditions the impact of carbon pricing on CEE. Based on comparisons of the literature, our research model shall use the capital, labor, and energy consumption of each decision-making unit (DMU) as inputs, and it will use GDP as desirable output and the carbon emission as undesirable output using the SFA approach in the first stage and the Tobit model in the second stage.

A review of the existing literature reveals three broad insights. First, China’s carbon-efficiency landscape shows persistent spatial divergence: eastern cities generally experience steady improvements, while cities in central and western regions continue to stagnate or decline. Second, recent methodological progress, particularly the expanded use of SFA, has improved the precision and interpretability of carbon-efficiency assessments compared with earlier DEA-based approaches. Third, institutional and policy factors, including urbanization patterns and land use reforms, play important roles, but their impacts differ considerably across regions and city tiers.

Building on this framework, the first stage of our analysis applies an SFA model to measure carbon emission efficiency across urban tiers. The second stage introduces urbanization as the main explanatory variable, with the number of universities as the proxy of high-quality manpower and the land use density as the proxy of urbanization intensity, capturing spatial heterogeneity, to examine why first-tier cities continue to improve while fifth-tier cities remain trapped in declining trajectories.

3. Methodology

The first stage of the analysis employs a stochastic frontier analysis (SFA) framework to measure city-level carbon emission efficiency. SFA is selected because it provides a more rigorous treatment of undesirable outputs and measurement uncertainty than non-parametric alternatives. Urban production activities often exhibit substantial random fluctuations arising from reporting errors, structural differences, and policy shocks. Unlike DEA, SFA explicitly separates random noise from inefficiency, producing efficiency estimates that are more statistically reliable.

Moreover, the translog specification adopted in this study allows the production frontier to capture non-linear substitution relationships among energy, labor, and capital inputs, which more accurately reflects the complexity of urban production systems. By embedding CO₂ emissions directly in the distance-function framework as an undesirable output, SFA offers a theoretically consistent and statistically grounded approach for constructing the carbon emission efficiency index.

To assess the environmental performance of the different city groups, we begin by constructing the distance function following the theoretical framework proposed by Zhang [19]. In line with Zhang [19], the 274 prefecture-level cities are grouped into five tiers that reflect their differences in economic scale, population size, industrial structure, and overall development level. In this study, we make a minor adjustment by placing several “new first-tier” cities—such as Nanjing, Wuhan, Chengdu, and Hangzhou—into the first-tier category. Although these cities are sometimes treated separately in commercial classifications, their economic capacity, functional roles, and industrial composition are broadly comparable to those of Beijing, Shanghai, Shenzhen, and Guangzhou. Under this classification, Tier 1 represents the major metropolitan hubs, Tiers 2–4 include provincial capitals, regional centers, and medium-sized developing cities, and Tier 5 consists of less-developed cities such as Ganzhou, Bazhong, and Dazhou.

The empirical analysis relies on city-level data obtained from the China Urban Construction Statistical Yearbook and the China City Statistical Yearbook. These official publications provide consistent information on population, land use, economic activity, and other key indicators across multiple years. Based on this tiered structure, we then applied the SFA model to measure the City Carbon Efficiency (CEE) index and to estimate the marginal abatement cost of CO₂ emissions.

3.1. Production Technology Hypothesis and Shephard Distance Function

For the implicit production technology, it is assumed that all tier cities in China achieve GDP (Y) output through inputs of labor (L), energy (E), and capital (K), accompanied by carbon dioxide (C) emissions. Production technology can be described as follows:

$$Q=\left\{\left.(E,L,C,K,Y)\right|\right(E,L,K\left) \mathrm{can} \mathrm{produce} \right(Y,C) \}$$

(1)

It is assumed that the production set Q satisfies the standard axiom that finite inputs can only lead to finite outputs [25]. Based on this, the environmental production technology needs to satisfy the following basic properties:

Weak disposability: $f(x,y,b)\in P(x) \mathrm{and} 0\le \theta \le 1,({x,\theta }_y,{\theta }_b)\in P(x)$.

Null-jointness: $f(x,y,b)\in P(x) \mathrm{and}\mathit{b}=0,y=0$.

Weak disposability means that the reduction in undesirable outputs must be achieved at the expense of some of the desired outputs. Null-jointness, on the other hand, means that the complete elimination of undesirable outputs cannot be achieved without completely suspending the production of desirable outputs. With this constraint, the Shepherd distance function for carbon emissions can be constructed as follows:

$$D(Y,L,K,C,E)=\mathit{sup}\{\alpha :(E,L,C/\alpha ,K,Y)\in T\}$$

(2)

where parameter α denotes the moderating factor and $C/\alpha$ is used to characterize the potential carbon emission level of each city in China. The City Carbon Emission Efficiency Index (CEE) is defined as the ratio of potential emissions to actual emissions, which measures the relative efficiency of cities in controlling carbon emissions:

$$0\le CEI={\displaystyle \frac{C/\alpha }{C}}={\displaystyle \frac{1}{\alpha }}={\displaystyle \frac{1}{D(E,L,C,K,Y)}}\le 1$$

(3)

3.2. Stochastic Frontier Analysis for CEE

Stochastic Frontier Analysis (SFA) models are structurally more similar to parametric linear programming (LP) methods, with the advantage of being able to derive the production function at any point and having a high sensitivity to outliers, which is not the case with non-parametric methods such as DEA. In addition, SFA possesses the statistical inference capability that LP lacks, and it is effective in identifying and eliminating measurement errors. The fitting accuracy and robustness of the model are further improved by introducing a random perturbation term. Based on the research framework of Aigner et al. [26], our SFA model is formulated as follows:

$$\mathit{lnD}(X,Y,C)=\mathit{lnC}+\mathit{lnD}(X,Y,1)$$

(4)

Compared with other types of functional forms, the Shephard distance function has more flexibility in measuring the efficiency boundaries of real environments [27]. Therefore, we further extend to the logarithmic transformation for the function to suit the research needs:

$$\begin{array}{cc}\hfill \mathit{lnD}(X,Y,C)& ={\beta }_0+{\beta }_K\mathit{lnK}+{\beta }_L\mathit{lnL}+{\beta }_E\mathit{lnE}+{\beta }_Y\mathit{lnY}+{\beta }_C\mathit{lnC}+{\displaystyle \frac{1}{2}}{\beta }_{\mathit{KK}}({lnK)}^2\hfill \\ & +{\beta }_{\mathit{KL}}\mathit{lnk}\cdot \mathit{lnL}+{\beta }_{\mathit{KE}}\mathit{lnK}\cdot \mathit{lnE}+{\displaystyle \frac{1}{2}}{\beta }_{\mu }({\mathit{lnL})}^2+{\beta }_{\mathit{LE}}\mathit{lnL}\cdot \mathit{lnE}+{\displaystyle \frac{1}{2}}{\beta }_{\mathit{EE}}({lnE)}^2\hfill \\ & +{\beta }_{\mathit{KY}}\mathit{lnK}\cdot \mathit{lnY}+{\beta }_{\mathit{LY}}\mathit{lnL}\cdot \mathit{lnY}+{\beta }_{\mathit{EY}}\mathit{lnE}\cdot \mathit{lnY}+{\beta }_{\mathit{KC}}\mathit{lnK}\cdot lnC\hfill \\ & +{\beta }_{\mathit{LC}}\mathit{lnL}\cdot \mathit{lnC}+{\beta }_{\mathit{EC}}\mathit{lnE}\cdot \mathit{lnC}+{\displaystyle \frac{1}{2}}{\beta }_{\mathit{YY}}({\mathit{lnY})}^2+{\beta }_{\mathit{YC}}\mathit{lnY}\cdot \mathit{lnC}+{\displaystyle \frac{1}{2}}{\beta }_{\mathit{CC}}({lnC)}^2\hfill \end{array}$$

(5)

In the SFA framework, the traditional perturbation term is divided into two components: a stochastic error term $\nu$ and a non-negative term $\mu$ reflecting the inefficiency, with a simplified parameter. $\mathit{lnD}(X,Y,C)=\mu \ge 0$ is introduced for a non-negative setting. Based on this, Equation (4) can be estimated in the following form:

$$\begin{array}{cc}\hfill \mathit{ln}({\displaystyle \frac{1}{C}})& =\mathit{lnD}(X,Y,1)-\mathit{lnD}(X,Y,C)\hfill \\ & ={\beta }_0+{\beta }_K\mathit{lnK}+{\beta }_L\mathit{lnL}+{\beta }_E\mathit{lnE}+{\beta }_Y\mathit{lnY}+{\displaystyle \frac{1}{2}}{\beta }_{\mathit{KK}}({lnK)}^2\hfill \\ & +{\beta }_{\mathit{KL}}\mathit{lnK}\cdot \mathit{lnL}+{\beta }_{\mathit{KE}}\mathit{lnK}\cdot \mathit{lnE}+{\displaystyle \frac{1}{2}}{\beta }_{\mu }({\mathit{lnL})}^2+{\beta }_{\mathit{LE}}\mathit{lnL}\cdot lnE\hfill \\ & +{\displaystyle \frac{1}{2}}{\beta }_{\mathit{EE}}({\mathit{lnE})}^2+{\beta }_{\mathit{KY}}\mathit{lnK}\cdot \mathit{lnY}+{\beta }_{\mathit{LY}}\mathit{lnL}\cdot \mathit{lnY}+{\beta }_{\mathit{EY}}\mathit{lnE}\cdot lnY\hfill \\ & +{\displaystyle \frac{1}{2}}{\beta }_{\mathit{YY}}({\mathit{lnY})}^2-\mu +\nu \hfill \end{array}$$

(6)

SFA is motivated by its ability to separate statistical noise from inefficiency and to incorporate CO₂ emissions as an undesirable output within a flexible distance-functional framework. Our data across cities often contain reporting fluctuations and structural differences that cannot be fully captured by non-parametric methods, making a parametric frontier more suitable for cross-city comparisons. Based on the resulting CEE, in the second stage, the Tobit model is used because the efficiency scores lie between 0 and 1, and a censored regression framework provides estimates that remain consistent with the bounded nature of the dependent variables.

4. Empirical Results

4.1. Overview of the Carbon Emission Efficiency of the Categorical City Groups

In order to evaluate the differences in carbon emission efficiency among the cities of different tiers in China, this paper includes 274 Chinese cities from 2006 to 2022. From the perspective of the bipolarization of urbanization, we categorized all the cities into five groups, from the first-tier metropolitan cities to the fifth-tier small cities in rural areas.

Table 2. Descriptive statistics of input and output variables from 2006 to 2022.

Variable	Unit	Mean	Std. Dev.	Min	Max	Obs.
k (capital)	CNY 10,000	70,784,721.24	85,681,804.67	3,307,030.00	993,605,800.00	4658
e (energy)	10,000 tons	2,044,030.70	3,563,578.03	12,321.95	41,648,440.00	4658
l (labor)	10,000 persons	52.33	82.67	4.21	1143.32	4658
y (GDP)	CNY 10,000	17,773,311.71	27,605,360.49	536,818.87	333,289,500.00	4658
c (CO2)	10,000 tons	4014.38	8126.63	552.24	113,834.70	4658

reports the descriptive statistics for the input and output variables employed in our SFA approach in the first stage. The dataset covers 274 prefecture-level cities in China from 2006 to 2022, forming a balanced panel dataset of 4658 observations. Among the input variables, capital shows the highest mean and deviation, reflecting significant differences in industrial scale across cities. Energy consumption also displays large dispersion, indicating heterogeneous energy-use patterns and industrial structures in urban economies. In contrast, labor exhibits relatively low variation, suggesting that workforce size differs less markedly among cities compared with other physical or energy inputs. Regarding outputs, GDP reveals substantial disparities in economic scale, which is consistent with the spatially uneven nature of the local economy, while CO₂ emissions present an even wider spread, implying persistent differences in carbon intensity and environmental performance.

Although the SFA dataset covers most prefecture-level cities in China between 2006 and 2022, minor data gaps remain. Specifically, one city (Puyang) was excluded because of incomplete base-year input data, while five cities (Bazhong, Fuzhou, Xiangyang, Ganzhou, and Dazhou) lack a single year of valid records due to inconsistencies between energy and output data sources. As a result, the final sample constitutes a slightly unbalanced panel of 273 cities and 4636 observations. These omissions are minimal in scale (representing less than 0.5% of all potential observations) and are unlikely to bias the SFA results, as the missing entries are randomly distributed across regions and years. The maintained panel thus retains sufficient coverage and reliability to support robust efficiency estimation.

Overall, the large standard deviations of capital, energy, and emissions highlight pronounced heterogeneity in both economic and environmental dimensions, which provides a solid empirical foundation for the subsequent SFA-based efficiency estimation.

Table 3. Correlation matrix of input and output variables.

Variable	k (Capital)	e (Energy)	l (Labor)	y (GDP)	c (CO2)
k (capital)	1.000
e (energy)	0.800	1.000
l (labor)	0.776	0.793	1.000
y (GDP)	0.901	0.875	0.858	1.000
c (CO2)	0.597	0.704	0.695	0.672	1.000

summarizes the correlation coefficients among the input and output variables used in the efficiency model. The results show that all variables are positively correlated, which is consistent with the notion that higher levels of capital, energy consumption, and labor are associated with greater economic output and CO₂ emissions. The correlation between GDP (y) and energy consumption (e) is particularly strong, reflecting the continued dominance of energy-intensive growth patterns across Chinese cities. None of the coefficients exceed the conventional multicollinearity threshold (|r| > 0.95), indicating that the variables can be jointly included in the SFA model without significant risk of collinearity bias.

Based on the variables in Table 2, we run the SFA model using Equation (6) to evaluate the carbon efficiency of all cities in the five categorical perspectives.

Table 4. Carbon emission efficiency of the five groups of cities (2006–2022).

City Tier	Average Carbon Emission Efficiency
First-tier group	0.914
Second-tier group	0.932
Third-tier group	0.932
Fourth-tier group	0.930
Fifth-tier group	0.923
Nationwide average	0.926

reports the average carbon emission efficiency across five city tiers in China during 2006–2022. The results show a gradual improvement in efficiency from the first-tier to the third-tier cities, followed by a modest decline in lower-tier cities (the fourth- and fifth-tier cities). This pattern suggests that the higher-tier cities have achieved greater efficiency gains through technological advancement and industrial upgrading, whereas smaller cities still face structural constraints, resulting in the bipolarization of carbon efficiency. The nationwide average carbon emission efficiency was 0.926, indicating quite competitive performance overall (see Figure 1).

In order to find out the heterogenetic character of the categorical groups of the cities, our research analyzes the average carbon emission efficiency for cities of Tiers 1 to 5 and the overall average in China in Figure 1. The carbon emission efficiency of the first-tier group of cities and the fifth-tier cities is relatively low, at 0.914 and 0.923, respectively. The carbon emission efficiency of the middle range of cities exceeds 0.93, which is much higher than the other two extreme groups of cities. It is noteworthy that the first-tier group of cities, such as Shanghai and Beijing, shows the lowest efficiency even with the strong regulation and market-oriented ETS support. Does this really mean that any policy measure and/or market-oriented promotion is not workable in the metropolitan cities? If so, most developing countries may not pursue the sustainable optimal path toward a low-carbon economy. In order to figure out the governance of the policies and ETS arrangements, dynamic trends for each group of cities are examined in Figure 2.

Figure 2 shows the very unique, dynamic character of the city group depending on the level of economic development. First, most of the first-tier group cities face temporary pressures as pilot zones for low-carbon policies, subject to stricter carbon standards and more rigorous regulatory requirements. This may result in their carbon emission efficiency appearing to comparatively increase over time. Compared with the economically well-developed first-tier group of cities, the fifth-tier group of cities shows a decreasing trend of its efficiency over time, showing another extreme of bipolarization with the first-tier group of cities. Between these two extremes, all the middle range of cities remained relatively stable and thus locked in carbon-intensive trajectories. This strange phenomenon in the carbon emission efficiency may result from the effect of urbanization, because the first-tier group shows intensive urbanization, compared with the middle range of the cities, while the fifth-tier group of cities is far behind in the urbanization trend. In order to figure out these phenomena in more detail, we shall examine two extreme groups of cities with the bipolarization perspective of urbanization in the following section.

4.2. Analysis of Carbon Emission Efficiency in the First- and Fifth-Tier Cities

Most of the previous empirical studies found that the metropolitan cities in eastern China are more carbon emission efficient than other tiers of cities. This assessment was based on the fact that the first-tier cities have significant advantages in terms of economic development, technological capabilities, management systems, and environmental awareness. For example, cities such as Beijing, Shanghai, and Shenzhen are already actively exploring the potential for green transition and have established stable political foundations and technical support in areas such as green energy utilization, renewable energy integration, green buildings, and smart transportation. Therefore, they should demonstrate better performance in terms of carbon emission efficiency per unit of production.

In terms of infrastructure and energy use, the first-tier cities are leading the green energy transition nationwide. Take Shenzhen as an example: by 2017, the city had already achieved 100% electrification of buses and taxis. Other cities such as Beijing and Shanghai have also achieved carbon neutrality in some areas and are actively promoting green initiatives such as distributed solar power generation, centralized heating, and the popularization of electric vehicles. These measures have strengthened the low-carbon capabilities of the first-tier cities from both market-supportive and political perspectives. As shown in Figure 3, the carbon dioxide emission efficiency of metropolitan cities has generally been improved intermittently from 2006 to 2021, except for the year 2022 due to COVID-19. This reflects that the cities have enhanced the gradual effectiveness of green transition and energy conservation policies. Nonetheless, the average carbon emission efficiency of the first-tier cities was 0.9136, the lowest among all city levels. This phenomenon demonstrates that despite the dynamic advances of the first-tier cities in the field of green growth, their carbon dioxide emission efficiency is too slow and thus still far behind the sustainable carbon-zero economy due to the lack of governance by the various structural factors.

Figure 4 shows changes in CO₂ emission efficiency in the fifth-tier cities from 2006 to 2022. Overall, CEE in the fifth-tier cities stayed relatively low compared to other city tiers during the analysis period, and it has shown a clear downward trajectory since reaching a peak (approximately 0.942) in 2007. This continuously decreasing trend highlights the persistent structural constraints faced by the small cities in their green transition process. There are not many strong regulatory policies and/or market-supportive incentives, such as ETS, in these small cities, implying no critical motivation to reduce carbon emissions. Moreover, there is no momentum or willingness for the better performance of local businesses. Outstanding, highly qualified labor leaves these small cities for better opportunities in the metropolitan cities. There is not much of a good spill-over effect from the leading businesses from the first-tier cities, resulting in ever-aggravated carbon emission abatement. It is worthwhile to note that the first- and fifth-tier cities are much lower than the nationwide average, while the first-tier cities show an increasing trend and the fifth-tier cities show a decreasing trend over time; hence, the so-called bipolarization. To address this issue and capture the underlying mechanism, the following section examines the governance factors of the carbon emission efficiency using the Tobit model.

5. Governance Evaluation with the Tobit Model

5.1. Tobit Regression Model

The SFA results in the first stage reveal a clear bipolarization between the first- and fifth-tier cities: the first-tier cities show a gradual improvement in carbon emission efficiency, while the fifth-tier cities exhibit a persistent downward trend over time, resulting in enlarging gaps in CEE. Previous studies, such as Sun et al. [2], attribute such divergence to the non-linear effects of economic and policy conditions. However, the patterns identified in our analysis require a more explicit examination of the underlying governance mechanisms. Given the substantial heterogeneity across cities, treating all cities as a single homogeneous group would obscure important differences and fail to capture the distinct structural and institutional contexts of each tier.

Since the efficiency values produced by the SFA model are bounded between 0 and 1, ordinary least squares (OLS) is not appropriate for estimating the determinants of these scores [28]. The Tobit model is therefore employed, as it is specifically designed for censored dependent variables and is widely used in efficiency studies to examine governance and structural drivers [13]. This choice ensures that the second-stage estimation is statistically consistent with the nature of the efficiency measurements.

In the two-stage analysis procedure, the Tobit model is generally applied in efficiency literature [8,28]. Using this Tobit model, this research shall evaluate the role of urbanization on the CEE of categorical cities. To find out this categorical difference, the urbanization rate is used as the governance factor to determine the performance of each categorical city. Here, urbanization rate is defined as the degree or capacity to integrate human capital, financial facilities, and intensive land development as the advanced inputs of economic performance in metropolitan cities. Most urbanization-related research emphasizes the role of migration of the workforce from a rural area to a metropolitan city. Specifically, from the perspective of environmental efficiency, the human capital, as the well-trained people with a high educational background, may have a more crucial effect on CEE coming from the urbanization, and thus we shall use the number of universities in the city as the proxy variable for not only manpower migration, but also for the technological innovation performance of the urbanization. Moreover, as urbanization results in a more selective concentration on space, the increasing facility and infrastructure demand makes intensive land use expand outward from the metropolitan city center. As shown in Figure 5, the first-tier cities are rapidly increasing their intensive land-use development over time, while the fifth-tier cities are almost stable over the test period, implying that urbanization has more intensively promoted the intensive land use in the metropolitan cities. The ever-increasing gap in Figure 5 shows the urbanization rate for intensive land use as the potential reason for bipolarization between the city groups. Therefore, the intensive land development ratio should be used as an explanatory variable of the Tobit model in the second stage.

To evaluate the effect of urbanization on the carbon efficiency of the cities, the number of universities in the city is used as a proxy of the highly qualitative human capital in the city, and the ratio of development area divided by the total city area is used as a proxy of the intensive land use.

Based on the Tobit regression approach by Debbarma et al. [28], we can define the econometric model in Equation (7):

$$Y_{\mathit{np}}={\{}_{0 otherwise}^{Y_{\mathit{np}}={\beta }^T{x}_{\mathit{np}}+{\epsilon }_{\mathit{np}}{\beta }^T{x}_{\mathit{np}}+{\epsilon }_{np}}\}$$

(7)

• $Y_{np}$ = the explained variable (CEE);
• $x_{np}$ = the explanatory variables;
• ${\beta }^T$ = the vector of the regression coefficient of the explanatory variable;
• ${\epsilon }_{np}$ = the stochastic error assumed to follow the distribution of N(0, ${\sigma }^2$), respectively.

To assess the factors influencing inefficiency in agriculture and its related sectors, the Tobit model can be defined in Equation (8) as follows [29]:

$$Y_{np}={\beta }_0+{\beta }_1{Z^1}_{np}+{\beta }_2{Z^2}_{np}+{\beta }_3{Z^3}_{np}+\cdots {\beta }_x{Z^x}_{np}+{\epsilon }_{np}$$

(8)

• Y = the efficiency measure (or environmental efficiency of sample firms);
• np = the $n^{th}$ city of sample study and the year or period of study;
• ${\beta }_x$ = the coefficient;
• ${Z^x}_{np}$ = the explanatory variables (number of universities, intensive land use ratio);
• ${\epsilon }_{np}$ = the stochastic error, respectively.

5.2. Tobit Regression Results and Their Implications

Based on the theoretical model in Equation (4), Tobit regression analysis is conducted to identify the underlying determinants as the governance factors of the CEE in the different city groups.

Table 5. Tobit regression results.

Variable Category	Variable	Baseline	With Fixed Effect
		Coefficient	Coefficient
Explanatory Variables (X)	Number of universities	0.000146	0.000146 **
	Land Development ratio	−0.23765 **	−0.237650 **
Control Variables (Z)	Domestic industrial output value	Included
	Population density	Included
Model Tests	Wald χ2 (X jointly)	7.751 **
Fixed Effects	City and year FE	Included

Notes: City and year-fixed effects are included. ** p < 0.05.

presents the results of the Tobit model estimation, in which the dependent variable is carbon emission efficiency (carbon_eff) and the explanatory variables include the number of universities and the intensive land development ratio. Two control variables of local industrial output value and population density are included to account for the intensity of industrial activity and the impact of population, while city and year-fixed effects are incorporated to reflect unobserved factors. The results of the Tobit regression provide important information to explain the contrasting trends in the carbon efficiency observed at all urban scales.

As shown in Table 5, the coefficient of the number of universities is positive, indicating that cities with stronger higher-education and human-capital resources tend to exhibit better carbon efficiency. This aligns with the expectation that universities facilitate technology diffusion, support green innovation, and enhance environmental awareness. However, the effect is not statistically significant in the baseline estimation, suggesting that the influence of human-capital concentration may be relatively modest or masked by broader structural differences across cities. To verify that the Tobit regression results are not driven by temporary shocks, the model was re-estimated after excluding observations from the year 2020 (year-fixed model), which was heavily affected by the COVID-19 pandemic. Removing this anomalous year helps assess whether short-lived disruptions distort the structural relationships between higher education, urbanization, and carbon efficiency.

As reported in Table 5, the results remain consistent with the baseline Tobit model, but the statistical significance has significantly improved. The coefficient of the number of universities continues to be positive and statistically significant. These patterns indicate that the efficiency-enhancing role of higher-education resources is robust to the exclusion of 2020. This consistency suggests that the estimated relationships are temporally stable and not driven by short-term shocks.

Compared with the number of universities, the coefficient for intensive land development is negative and statistically significant both at baseline as well as in the fixed effect model, implying that rapid land expansion and concentrated industrial construction tend to suppress carbon efficiency. Congestion effects, higher energy use, and the clustering of carbon-intensive activities may help explain this relationship. Due to the opposite signs of urbanization rate between labor capital represented by universities and the intensive land use ratio, urbanization may take on the non-linear patterns of the CEE over time, but in our research, it is clear that both made a wider gap of bipolarization in CEE.

During the early phase of rapid urbanization in the 2000s, metropolitan areas such as Beijing, Shanghai, and Shenzhen underwent extensive industrialization and large-scale infrastructure expansion. This period was marked by high carbon emissions and relatively low efficiency, as economic growth was prioritized over environmental performance. The Tobit regression results show that the coefficient of the intensive land development ratio is significantly negative, indicating that rapid urban growth over time may raise carbon emissions faster than output, leading to lower efficiency levels. Liu and Zhang (2022) also demonstrated that urban expansion significantly increased CO₂ emissions in the major metropolitan cities of China from 2000 to 2018, as rapid land urbanization outpaced population and industrial transitions, thereby exacerbating carbon intensity and undermining efficiency [30].

As these metropolitan economies matured, the accumulation of human capital and the expansion of higher education institutions gradually shifted the efficiency trajectory upward. The positive and significant coefficient of the number of universities highlights the essential role of education-driven innovation in improving carbon efficiency. Cities with a higher concentration of universities benefit from stronger technology diffusion and research collaboration, and a larger pool of skilled labor, which together accelerate the transition toward low-carbon development through the positive role of urbanization only in the metropolitan cities. This pattern helps explain why the first-tier cities, despite early inefficiencies, have achieved steady improvements in carbon efficiency over time.

By contrast, the fifth-tier cities, although initially more efficient due to the limited industrial activities, eventually experienced declining efficiency in recent years. The lack of higher education and technological infrastructure constrains their capacity to adopt advanced green technologies. Moreover, their economic expansion often depends on resource-intensive sectors and uncontrolled urban sprawl, which amplifies the negative effect captured by the intensive land development ratio. Consequently, while these smaller cities once held a relative efficiency advantage, they have gradually lost it as their growth remains tied to high-carbon, low-innovation pathways. This trend becomes worse when the labor capital emigrates to the metropolitan cities for better job opportunities.

The control variables of industrial output, as well as population density of the city, showed that its effective inclusion is necessary, implying that urbanization had a great effect not only on the qualitative human capital and intensive land development ratio, but also on the quantitative support by the industrial outputs and population density. More people and businesses resulted in higher CEE as well, implying that urbanization is the core governance factor to enhance the CEE.

Overall, the Tobit regression results complement the earlier SFA findings by revealing a structural mechanism: education-led innovation enhances efficiency, whereas land-driven urban expansion suppresses it. This dual pattern emphasizes the need for differentiated carbon reduction strategies across city tiers, promoting innovation-based growth in large cities and stronger momentum in the small cities.

6. Discussion

This study provides new insights into the feasibility of a carbon-zero economy across the diverse cities of China. Many previous studies have shown that large and coastal cities often achieve higher levels of environmental sustainability, while many inland and small cities lag behind. In this research, we confirmed that this pattern over time is correct, but the overall performance of the top-tier metropolitan cities is much, much lower than any other city. What is worse is that the fifth-tier small cities are much lower in their average performance of environmental sustainability, and also they have been on a decreasing trend, resulting in a larger gap between the first- and fifth-tier cities. These tier-level differences highlight structural differences that are not apparent when cities are simply grouped by region or province.

This analysis is also relevant in the context of the current debate on measurement methods. The use of a stochastic frontier model that includes CO₂ as a by-product is consistent with studies that emphasize the importance of distinguishing between random fluctuations and inefficiencies when analyzing different cities. Our empirical results support this choice: SFA-based estimates remain stable across cities, while revealing systematic differences between levels. Compared with uncontrolled single-stage methods, this approach seems more suitable for capturing the co-production of economic performance and environmental outcomes, where the mix of sectors and policy changes varies by location.

The tier-based research also sheds light on the mechanism behind the “urban growth puzzle” described in this study. In the first-tier cities, urban growth is closely linked to diverse industrial structures, strong technological capabilities, and more effective governance institutions. These factors allow for cost savings and foster innovation that improves carbon efficiency. In the fifth-tier cities, urban growth still relies on land-intensive development activities that generate high carbon emissions. Limited financial resources and weak manpower capacity further hinder the ability of these cities to introduce greener production technologies or enforce environmental standards.

The research is unique in that it examined the role of urbanization from an environmental governance perspective. Tools such as ETS seem to work well in megacities with diverse economies and advanced administrative capacities. Nonetheless, the regulated ETS market in metropolitan cities could not enhance the quality of life. Moreover, the small cities are not strongly regulated for carbon emissions, and thus their average CEE is much lower than the middle range of cities. All these difficulties can be found to be deeply rooted in urbanization. The undesirable performance of urbanization promoted energy-intensive, concentrated land development with much more complicated and complex carbon emissions. However, due to the strong market-oriented ETS system and regulation, the metropolitan cities have shown an increasing CEE trend, implying that the policy direction is right, but with more market-oriented incentives instead of the regulatory regime of ETS. In order to enhance and complement the regulatory ETS market, this research proposed voluntary reduction measures as complementary momentum to the ETS system.

In contrast, smaller cities often lack the institutional and market conditions necessary for full participation in emissions trading systems. A uniform national policy that ignores tier-specific constraints risks exacerbating existing inequalities in carbon efficiency. Since there is no regulation and no market-oriented support of the ETS, these small cities have no willingness or potential capacity. The only solution for these small cities to participate in carbon emission abatement may come from voluntary reduction by the companies and organizations in the small cities, implying that the strong promotional policies of the voluntary reduction system are necessary for small cities in China, as well as in other developing countries.

7. Conclusions

In this research, we analyzed the CEE of each categorical city group from the 274-city data for the period of 2006 to 2022. In the first stage, we estimated the CEE of cities in the five categorical perspectives and found that the first- and fifth-tier cities are extremely different from the middle range of groups in that the first-tier metropolitan cities showed the lowest CEE, increasing over time, while the fifth-tier cities are still lower than the middle range of cities, and worse, it is decreasing over time. It implies that the policies for the low-carbon economy should be differentiated depending on the categorically distinct group characteristics. Specifically, based on the trend of the first- and fifth-tier cities, the bipolarization of carbon emissions is too serious to take measures because the CEE gap becomes larger and larger over time, and urbanization may aggravate this paradox of CEE.

Many developing countries, such as China, have promoted urbanization to speed up the economic growth of the country. This selective concentration by the central government made strong regulations on the metropolitan cities, including ETS for the leading businesses in the cities. The sponge effect through urbanization may enhance the CEE of metropolitan cities, and urbanization seems to be working well in the metropolitan cities, at least. However, in the urbanization process, the small cities experienced a deterioration in their CEE because the highly qualified labor force goes to the metropolitan cities, resulting in the aggravation of CEE in the small cities. Moreover, urbanization makes this bipolarization in the CEE over time become larger and larger, resulting in an unsustainable future for the other parts of the cities. Clearly, our research shows that, except for the first-tier cities, CEEs of all the other cities are locked in a stable range or are even decreasing over time. It suggests that the selective concentration of urbanization on the metropolitan cities is not sustainable, and thus, the central government should provide more sustainable governance for small cities in rural areas.

Overall, all the cities require not only political support for environmental issues, but a strong momentum for the companies to voluntarily reduce their carbon emissions more proactively. Therefore, our research proposes a public–private partnership (PPP) for voluntary reduction initiatives, as the Paris regime is based on the voluntary bottom-up approach to solving the global climate crisis. The only way to promote sustainable development under the Paris regime is not regulated ETS policies, but promotional governance with strong incentives for voluntary participation by the local private powers.

Although this study provides a comprehensive assessment of tier-specific carbon emission efficiency, several limitations should be acknowledged. The empirical analysis relies on reliable and affordable datasets and methodologies. In our research, the tier-based classification may not fully capture variations within each group of cities. The better proxy variables for urbanization, as well as the enhanced research methodology, shall certainly result in more precise and appropriate implications and suggestions, not only for China but for all developing countries.