A Study on the Improvement Pathways of Carbon Emission Efficiency in China from a Configurational Perspective Based on the Dynamic Qualitative Comparative Analysis

Abstract

Improving carbon emission efficiency (CEE) is crucial for coordinating economic development and reducing carbon emissions. Drawing on panel data for 30 provinces in China from 2013 to 2022, this paper selects six key antecedent conditions guided by the Technology–Organization–Environment (TOE) framework. Then the dynamic qualitative comparative analysis (DQCA) is employed to explore CEE improvement pathways from a configurational perspective, and regression analysis is used to compare the driving effects of different pathways. The findings reveal that (1) single factors cannot independently achieve high CEE; instead, multiple factors must work synergistically to form various improvement pathways, including “technology–organization dual-driven”, “environment-dominated”, and “multi-equilibrium” pathways, with industrial structure upgrading as the core factor for improving CEE; (2) temporally, these improvement pathways demonstrate universality, while, spatially, they exhibit significant provincial heterogeneity; and (3) in terms of marginal effects, the “multi-equilibrium” pathway has the strongest driving effect on CEE. The findings provide valuable policy implications for developing targeted CEE enhancement strategies across provinces at different developmental stages.

1. Introduction

With the rapid advancement of global industrialization and urbanization, carbon emissions have become one of the primary factors contributing to global warming. The resulting series of impacts, such as frequent extreme weather events, rising sea levels, and sharp declines in biodiversity, are manifesting globally [1]. As a populous country, China also faces severe environmental problems. Its long-term development model, which is highly dependent on fossil fuels, has placed China at the forefront of global carbon emissions, as it accounts for approximately 28% of the world’s total [2]. Faced with enormous pressure to reduce carbon emissions, China pledged at the 75th United Nations General Assembly to reach the “carbon peak” by 2030 and achieve “carbon neutrality” by 2060. Moreover, the problem of unbalanced and inadequate economic development in China remains prominent, and improving carbon emission efficiency (CEE) is a feasible and effective approach for coordinating energy, the economy, and the environment [3].

Although scholars have studied the factors influencing CEE, the traditional linear regressions adopted focus on the “net effect” analysis of single factors, which is not applicable to complex causal relationships involving multiple concurrent factors. Given that the CEE results from a holistic system of interacting variables, it calls for a methodological approach capable of analyzing the “configurational effects” of variables [4,5]. Therefore, this study employs dynamic qualitative comparative analysis (DQCA) to explore the joint effects and interactive relationships of different influencing factors on CEE from a configurational perspective. Then the multivariate linear regression is subsequently used to explore the driving effects of each configuration pathway. This study aims to answer the following questions: Is there a single factor that is necessary for influencing CEE? What are the pathways for improving CEE? Are there temporal or case effects? What are the driving effects of different pathways? The remainder of this paper is organized as follows: The second section provides a detailed review of research related to CEE. The third section establishes the model and indicator system and describes the data sources. The fourth section presents the empirical results. The fifth section analyzes the rationality of the results and proposes policy implications and limitations. The sixth section concludes this research.

2. Literature Review

Globally, the pursuit of enhanced CEE is a central theme in climate policy research, though the research priorities vary significantly across economies with different development levels. Regarding developed economies, research primarily centers on cross-national comparative analysis. Wang et al. (2022) [6] found through cross-national comparative research on Organization for Economic Co-operation and Development (OECD) high-income countries that urbanization generally promotes carbon emission reduction, with all member countries achieving decoupling between urbanization and carbon emissions. In contrast, research on emerging economies increasingly focuses on domestic regional differences or industry characteristics. At the microlevel, Bagchi et al. (2022) [7] conducted enterprise-level analysis by constructing carbon productivity indicators and identified that technological advancement, export-oriented strategies, and other factors significantly influence CEE in Indian manufacturing enterprises. At the industry level, Modise et al. (2021) [8] focused on South Africa’s automotive parts manufacturing industry, employing ARIMA models to achieve precise predictions of energy consumption and carbon emission trends in this sector.

In China, numerous scholars have used total-factor indicators to assess CEE at the provincial [9,10], city [11,12], and industry levels [13,14]. Currently, the most studied influencing factors include technological progress [15,16], industrial structure [17,18], energy efficiency [19], and carbon emission trading systems [20]. In previous studies on the factors influencing CEE, various econometric models and estimation methods have been adopted, such as threshold regression [21,22], the Tobit model [23,24], and spatial autocorrelation analysis [25,26]. Empirical studies in different regions and periods have yielded diverse conclusions, including positive correlations, negative correlations, bidirectional correlations, nonlinear correlations, and no significant correlations.

Most studies on influencing factors have focused primarily on the impact of a single factor on CEE and neglected the complex interactions between multiple factors. When variables are interrelated, the unique effects of individual variables may be overshadowed by the effects of correlated variables, which is one of the reasons for the inconsistent research conclusions. Recognizing this, scholars have gradually shifted toward adopting a configurational perspective to study CEE influencing factors from a holistic approach. Yan et al. (2023) [27], on the basis of the Technology–Organization–Environment (TOE) framework, used the fuzzy set qualitative comparative analysis (fsQCA) method to identify three driving modes for low carbon emission intensity. Similarly, adopting a configurational perspective, Zhang et al. (2024) [28] combined fsQCA with necessary condition analysis (NCA) to identify multiple path combinations supporting low-carbon transformation by applying the superefficiency slack-based measure (SBM) model to measure CEE.

In summary, scholars have actively explored CEE and laid a solid foundation for this study. However, there are still areas that need further development: (1) Existing research often focuses on analyzing the “net effect” of a single factor on CEE and insufficiently explains the intrinsic mechanisms underlying the formation of high (non-high) CEE under the synergistic action of multiple factors. Although scholars have explored this issue from a configurational perspective, existing qualitative comparative analyses (QCAs) are limited by cross-sectional data and lack adequate consideration of the time factor, which may lead to nonrobust configurational conclusions. (2) Existing configurational studies often stop at path identification and lack in-depth quantitative analysis of the driving effects of different configurational pathways. In view of this, further investigation is needed to explain the changing trends of CEE over time from a configurational perspective and explore effective pathways for improving CEE.

The potential marginal contributions of this paper are as follows: (1) This paper shifts the perspective from isolated factors to complex causal combinations, exploring the driving pathways of the CEE under the influence of multiple interconnected factors. The time dimension is introduced into the configuration for the first time, breaking down the barriers between panel data and QCA methods, making the research more realistic and forward-looking. (2) By integrating configurational antecedent conditions into comprehensive indicators and incorporating them into econometric models along with other control variables, the marginal effects of various driving pathways are comparatively analyzed using multiple regression methods, which allows us to better verify the configurational results.

3. Materials and Methods

3.1. Main Methods

3.1.1. Refined Three-Stage Data Envelopment Analysis (DEA) Model

Traditional or improved DEA models often overlook the impact of environmental factors and random disturbances on the efficiency evaluation of decision-making units (DMUs), which may lead to biases in efficiency values. To address this issue, Fried et al. (2002) [29] proposed a three-stage DEA model, which introduces the stochastic frontier analysis (SFA) method to separate environmental factors and random disturbances and thereby obtain more objective efficiency evaluation results.

This study proposes a refined three-stage DEA model by introducing a superefficiency epsilon-based measure (Super-EBM) model that considers undesirable outputs and integrates radial and nonradial methods. The specific steps are as follows: (1) Stage 1: Apply the Super-EBM model to evaluate the initial efficiency and obtain the initial efficiency value and input slack variables. (2) Stage 2: Utilize the SFA model to eliminate the effects of environmental factors and random disturbances. Via construction of the SFA model, the input slack variables obtained in the first stage are regressed, analyzed, and decomposed into three effects: environmental factors, management inefficiency, and statistical noise. Then, the adjusted input variables are obtained. (3) Stage 3: Replace the original input data with the adjusted input variables, rerun the Super-EBM model, and obtain the final efficiency evaluation results.

3.1.2. Dynamic Qualitative Comparative Analysis

Complex social phenomena are typically the result of dynamic interactions among multiple factors. Traditional QCA methods have a “time blind spot” in theoretical construction, as cross-sectional data tend to be used, such that the impact of time on condition configurations is ignored and the dynamic trajectory of multiple factors is not identified [30]. To address these issues, Garcia-Castro et al. (2016) [31] innovatively developed a panel data QCA method. This method comprehensively handles panel data from three aspects, namely, pooled results, between results, and within results, enabling the observation of changes in a particular configuration across time and geographic dimensions and effectively overcoming the problem of ignoring time and space in current QCA studies.

3.2. Variable Selection

3.2.1. Outcome Variable

This study employs the refined three-stage DEA model to assess CEE as the outcome variable. The reasonable selection of input and output variables can improve the accuracy of DEA, and multiple dimensions, such as the environment, economy, society, and resources, should be comprehensively considered. In general, the consumption of capital, resources, or energy represents input, whereas a product or service represents output. Notably, the more variables there are in the model, the more difficult it is to distinguish the DMU [32]. Therefore, the number of variables should be minimized while retaining the necessary factors of production.

This paper refers to the variables selected in the CEE assessment [33,34] and considers the applicability of the indicators in the selected model. The employment population, fixed asset investment, and total electricity consumption are used as the proxy variables for labor, capital, and energy, respectively. The gross regional product is the desired output, and the total carbon emission is the undesired output.

3.2.2. Conditional Variables

Research and practice indicate that the factors influencing CEE are mostly contained within the technological, organizational, and environmental levels. However, the variables selected in relevant studies are relatively scattered, and no integrated analytical framework has been theoretically constructed to promote carbon emission reduction [27]. In this study, the TOE theoretical framework is introduced into CEE research to facilitate the integration and analysis of existing factors.

The TOE framework, first proposed by Tornatizky and Fleischer in 1991 [35], plays an important role in identifying factors and explaining complex causal relationships. The technology dimension was chosen first and includes two secondary indicators: technological progress and energy efficiency. (1) Technological progress, as explained by endogenous growth theory, is the core of economic growth [36]. In the context of the intelligent era, its importance in promoting carbon emission reduction and improving environmental quality has become increasingly prominent. Numerous scholars, such as Xie et al. (2021) [37] and Li et al. (2023) [38], have confirmed that technological progress is a key pathway for enhancing CEE. (2) Existing studies show that improving energy efficiency means reducing the energy consumption at the same output level, promoting the optimization of the energy structure, further reducing dependence on fossil fuels, and helping countries achieve environmental benefits at lower costs, thereby effectively enhancing CEE [39].

Next, the organizational dimension is selected, which includes two secondary indicators: industrial structure and government support. (1) Industrial structure upgrading usually drives the transformation from traditional high-energy-consuming and high-emission industries to low-energy-consuming and low-emission high-tech and service industries. This transformation not only directly reduces carbon emission intensity but is also often accompanied by technological progress and innovation capacity enhancement, which results in optimized resource allocation, reduced energy waste, and thereby enhanced CEE [40]. (2) Government support is indispensable for urban development, with financial inputs and policies inclined to strongly support the achievement of low-carbon transformation. Government fiscal expenditures can be directed toward green and low-carbon fields and promote the progress and application of low-carbon technologies. By constructing a panel effect threshold model, Yang et al. (2019) [41] concluded that government subsidies have a positive threshold effect on economic development and the energy transition.

Finally, the environmental dimension is selected, which includes two secondary indicators: economic development and environmental regulation. (1) The impact of economic development on CEE is complex and dual-sided [42]. On the one hand, during the initial stage of economic development, especially during the industrialization phase, the expansion of high-energy-consuming industries leads to increased energy consumption and decreased CEE. On the other hand, with further economic development, technological progress, industrial structure optimization, and enhanced environmental awareness can significantly improve energy efficiency, which promotes CEE improvement. (2) Research indicates that environmental regulation effectively promotes carbon emission reduction [43]. By setting strict emission standards, environmental regulation prompts enterprises to adopt clean and efficient production technologies while driving the industrial structure in a low-pollution and low-energy-consumption direction, thereby effectively increasing the overall economic CEE. In summary, on the basis of the TOE framework, this study selects six conditional variables—technological progress, energy efficiency, industrial structure, government support, economic development, and environmental regulation, to explore the influence mechanisms of antecedent conditions on CEE enhancement in three dimensions. The results are shown in Figure 1.

3.3. Data Sources and Calibration

This study selects samples from provinces, autonomous regions, and municipalities of China (excluding Tibet) from 2013 to 2022. The data come mainly from the China Carbon Emissions Accounts and Datasets (CEADS), the China Statistical Yearbook, the China Energy Statistical Yearbook, the China Industrial Statistical Yearbook, the China Labor Statistical Yearbook, and provincial and municipal statistical yearbooks. Following the approach of Ou et al. (2020) [44], this study uses interpolation methods to handle minor missing data points in individual years. The main variable definitions are shown in

Table 1. Variable definition.

Primary Indices	Secondary-Class Indices	Third-Class Indices
Input	Labor	Employment population
	Capital	Total fixed asset investment
	Energy	Total electricity consumption
Output	Desirable	Gross regional product
	Undesirable	Carbon emission
Outcome	Calculated from the refined three-stage DEA model
Conditions	Technological progress	Number of patent grants
	Energy efficiency	The inverse of total energy consumption in per capita GDP
	Industrial structure	The ratio of labor in the tertiary industry to labor in the secondary industry
	Government support	Fiscal expenditure
	Economic development	Per capita GDP
	Environmental regulation	General industrial solid waste comprehensive utilization rate

.

Before conducting DQCA, it is necessary to calibrate the antecedent condition and outcome variables. On the basis of the actual case, following the approach of Du et al. (2017) [4], the data was converted into fuzzy set membership scores using the direct calibration method, with the 5th percentile, 50th percentile, and 95th percentile set as the full nonmembership point, crossover point, and full membership point, respectively. The calibration and descriptive statistical results are presented in

Table 2. Variable calibration and descriptive statistics.

	Calibration			Descriptive Statistics
	Full Nonmembership	Crossover	Full Membership	Mean	Std. Dev.	Min	Max
CEE	1.001	0.746	0.538	0.754	0.148	0.434	1.009
Technological progress	307,365.95	37,886.50	3027.55	80,692.35	123,449.8	502	872,209
Energy efficiency	18.253	3.871	1.706	5.931	5.341	1.297	27.594
Industrial structure	4.177	1.728	1.017	1.964	0.931	0.879	6.134
Government support	11,723.235	5178.495	1478.541	5843.86	3152.129	922.48	18,533.08
Economic development	123,758.350	55,638	31,766.650	64,505.61	30,924.54	22,922	190,313
Environmental regulation	0.968	0.615	0.374	0.649	0.198	0.269	1.004

.

4. Results

4.1. Necessary Condition Analysis

This study employs the DQCA model in R studio 4.2.1 to explore the improving pathways of CEE. In traditional QCA, necessary condition analysis is used to examine whether a single condition variable is a necessary condition for the outcome variable. When the consistency level of the condition variable exceeds 0.9 and the coverage exceeds 0.5, the variable can be considered a necessary but not sufficient condition for the outcome variable [45]. In DQCA, it is necessary to further analyze the between consistency (BECONS) adjusted distance and within consistency (WICONS) adjusted distance. Garcia-Castro et al. (2016) [31] found that, when the adjusted distance does not exceed 0.2, pooled consistency (POCONS) has high precision and can serve as a basis for judgment. If the adjusted distance exceeds 0.2, further investigation of the necessity of the condition variable is needed, as there may be time or case effects. As shown in

Table 3. Necessary condition analysis results.

Conditional Variables	High CEE				Non-High CEE
	POCONS	Pooled Coverage	BECONS Adjusted Distance	WICONS Adjusted Distance	POCONS	Pooled Coverage	BECONS Adjusted Distance	WICONS Adjusted Distance
Technological progress	0.593	0.683	0.229	0.495	0.565	0.664	0.207	0.460
~ Technological progress	0.708	0.615	0.193	0.391	0.730	0.647	0.120	0.345
Energy efficiency	0.735	0.809	0.033	0.385	0.488	0.548	0.211	0.529
~ Energy efficiency	0.589	0.530	0.040	0.472	0.829	0.761	0.084	0.247
Industrial structure	0.672	0.734	0.098	0.403	0.550	0.613	0.229	0.495
~ Industrial structure	0.646	0.585	0.098	0.385	0.762	0.703	0.124	0.322
Government support	0.603	0.604	0.193	0.454	0.691	0.706	0.189	0.282
~ Government support	0.706	0.692	0.211	0.420	0.612	0.611	0.211	0.408
Economic development	0.693	0.734	0.251	0.391	0.533	0.576	0.294	0.495
~ Economic development	0.599	0.557	0.254	0.449	0.754	0.715	0.145	0.322
Environmental regulation	0.689	0.684	0.156	0.443	0.560	0.567	0.160	0.523
~ Environmental regulation	0.563	0.557	0.138	0.518	0.687	0.693	0.098	0.437

Note: “~” indicates the logical operation of “not”, that is, the condition is missing.

, the POCONS coefficients for the six conditional variables are all less than 0.9, indicating that they are not necessary conditions for the outcome variables. However, the BECONS and WICONS adjusted distances of some variables exceed 0.2, indicating that they require further analysis.

This can be explained by the following aspects. First, the 30 provinces (municipalities) selected in this study are vast and have different natural resource endowments and development environments, such that internal and external conditions affect CEE development to different degrees; this results in large WICONS-adjusted distances. Second, for cases where the BECONS adjustment distance is greater than 0.2, it is necessary to further investigate the changes in BECONS and between coverage (BECOVS) of the condition variable on the outcome variable.

Table 4. BECONS adjustment distance data greater than 0.2.

Case	Causal Combinations		Year
			2013	2014	2015	2016	2017	2018	2019	2020	2021	2022
1	Technological progress and high CEE	BECONS	0.392	0.433	0.542	0.55	0.541	0.576	0.633	0.699	0.756	0.744
		BECOVS	0.653	0.678	0.65	0.664	0.701	0.665	0.728	0.668	0.684	0.714
2	Technological progress and non-high CEE	BECONS	0.442	0.422	0.49	0.508	0.538	0.616	0.59	0.671	0.738	0.696
		BECOVS	0.792	0.775	0.753	0.737	0.705	0.697	0.631	0.632	0.569	0.542
3	Energy efficiency and non-high CEE	BECONS	0.36	0.387	0.392	0.437	0.485	0.539	0.558	0.557	0.598	0.626
		BECOVS	0.503	0.532	0.547	0.564	0.558	0.571	0.562	0.57	0.538	0.527
4	Industrial structure and non-high CEE	BECONS	0.377	0.409	0.449	0.491	0.556	0.604	0.671	0.666	0.673	0.666
		BECOVS	0.623	0.637	0.652	0.635	0.607	0.597	0.606	0.631	0.593	0.575
5	~ Government support and high CEE	BECONS	0.785	0.764	0.695	0.672	0.604	0.548	0.517	0.484	0.515	0.471
		BECOVS	0.582	0.624	0.665	0.665	0.601	0.592	0.599	0.604	0.583	0.569
6	Economic development and high CEE	BECONS	0.42	0.507	0.571	0.628	0.655	0.735	0.765	0.802	0.869	0.897
		BECOVS	0.719	0.73	0.725	0.744	0.754	0.762	0.747	0.721	0.724	0.719
7	Economic development and non-high CEE	BECONS	0.369	0.389	0.396	0.427	0.499	0.574	0.618	0.65	0.732	0.767
		BECOVS	0.681	0.656	0.645	0.607	0.582	0.583	0.56	0.577	0.52	0.498
8	~Economic development and high CEE	BECONS	0.813	0.761	0.721	0.669	0.637	0.598	0.549	0.53	0.424	0.373
		BECOVS	0.545	0.515	0.482	0.493	0.557	0.589	0.607	0.606	0.65	0.664
9	~ Government support and non-high CEE	BECONS	0.925	0.881	0.819	0.769	0.74	0.703	0.621	0.602	0.567	0.514
		BECOVS	0.636	0.613	0.611	0.634	0.728	0.774	0.776	0.761	0.753	0.765

shows the data for a BECONS adjustment distance greater than 0.2. The specific findings are as follows: In cases 1–8, the BECONS for each year is less than 0.9, indicating no necessary relationship. In case 9, although the BECONS in 2013 exceeds 0.9 and the BECOVS exceeds 0.5, a scatter plot of the condition variable and the outcome variable shows that more than one-third of the case points are distributed above the diagonal. This finding indicates that, despite passing the consistency test, the condition variable cannot be considered a necessary condition for non-high CEE [46]. Therefore, in the condition configuration sufficiency analysis, this study does not set a single condition variable as a necessary condition.

4.2. Condition Configuration Sufficiency Analysis

In QCA, conditional configuration analysis is crucial for studying how different combinations of conditions lead to outcomes. In accordance with the research of Navarro et al. (2016) [47] and Hu et al. (2024) [48], this study adopts a consistency threshold of 0.8, a frequency threshold of 3, and a proportional reduction in inconsistency (PRI) threshold of 0.6 as criteria for truth table construction. A literature review reveals that there is no consensus on the relationships between the six conditional variables and CEE. Therefore, no directional expectations are set, and all conditions are selected as “present or absent”. Regarding solution formulation, the intermediate solution is used as the primary basis, supplemented by the simple solution, to identify the core and peripheral conditions. Specifically, the variables present in both solutions are identified as core conditions, whereas those present only in intermediate solutions are considered marginal conditions. The results of the conditional configuration sufficiency analysis and the typical cases corresponding to the configurations are presented in

Table 5. Conditional configuration sufficiency analysis results.

Condition	High CEE				Non-High CEE
	S1a	S1b	S2	S3	N1a	N1b	N2	N3a	N3b
Technological progress
Energy efficiency
Industrial structure
Government support
Economic development
Environmental regulation
Consistency	0.842	0.842	0.939	0.895	0.891	0.877	0.941	0.929	0.954
PRI	0.647	0.643	0.843	0.681	0.753	0.734	0.853	0.807	0.865
Coverage	0.383	0.375	0.313	0.261	0.518	0.479	0.498	0.39	0.362
Unique coverage	0.021	0.018	0.092	0.029	0.022	0.037	0.037	0.004	0.002
Overall consistency	0.868				0.858
Overall PRI	0.751				0.735
Overall coverage	0.57				0.699

Note: Black circles () indicate the presence of a condition; crossed-out circles () indicate its absence. Large circles represent core conditions, while small circles represent marginal conditions. Blank spaces indicate “don’t care” [49].

and

Table 6. Typical cases for configuration pathways.

Configuration	Configuration Name	Typical Cases
S1a	Technology–Organization Dual-Driven	Gansu (2018–2022), Jilin (2013–2022), and Hainan (2014–2019)
S1b
S2	Multi-Equilibrium	Beijing (2013–2022) and Shanghai (2014–2022)
S3	Environment-Dominated	Fujian(2013–2015) and Tianjin(2013–2015)

, respectively.

4.2.1. Pooled Results

As shown in Table 5, the overall consistencies of high and non-high CEE are 0.868 and 0.858, respectively, both exceeding 0.8, which indicate that the condition configurations can be considered sufficient for high (non-high) CEE. The overall coverage rates are 0.57 and 0.699, respectively, suggesting that the configurations have high explanatory power. There are four configurational pathways for high CEE: configurations S1a and S1b can be termed “Technology–Organization Dual-Driven”; configuration S2 can be named “Multi-Equilibrium”; and configuration S3 can be named “Environment-Dominated”. There are five configurational pathways for non-high CEE. This study focuses on discussing the configurations for high CEE and analyzing the corresponding cases.

Configuration S1a indicates the core conditions of high energy efficiency, high industrial structure, and non-high government support, with peripheral conditions of non-high technological progress and non-high economic development. Configuration S1b shares the same core conditions but has non-high environmental regulation as a peripheral condition instead of non-high economic development. A comparison of S1a and S1b reveals a substitution relationship under peripheral conditions (non-high economic development in S1a versus non-high environmental regulation in S1b), both of which are related to the external environment. This reflects that, under circumstances of inadequate government support and low technological progress, high CEE can be achieved primarily through improving energy efficiency and actively promoting industrial structure upgrading. Typical provinces in this configuration include Gansu (2018–2022), Jilin (2013–2022), and Hainan (2014–2019), suggesting that, for less developed regions with weak economic foundations, excessive reliance on government support and technological innovation is not the only path. Instead, they can achieve leapfrog development through “asset-light” development methods that avoid the old path of “pollute first, treat later”. Hainan Province is a typical example, as it has proposed the “eco-provincial” strategy and leveraged its natural resources to vigorously develop tourism, and it consistently ranks among the top in CEE.

Configuration S2 includes the core conditions of high industrial structure and high environmental regulation, with peripheral conditions of high technological progress, high energy efficiency, and high economic development. This path demonstrates that the synergy of technology, organization, and the environment can provide strong support for improving CEE. Typical cases of this configuration are primarily Beijing (2013–2022) and Shanghai (2014–2022). As China’s most developed first-tier cities, Beijing and Shanghai possess strong economic foundations, advanced technological levels, and sophisticated industrial systems. As key national cities, they have consistently led in environmental regulation and implemented strict environmental standards and carbon emission reduction policies early on. They have completed the transition to service-oriented economies and have high proportions of tertiary industry, developed high-tech industries, and modern service sectors. This indicates that, in regions with adequate economic foundations and innovative capabilities, significant improvements in CEE can be achieved through multidimensional coordinated development. The effect of this multifactor synergy far exceeds the simple addition of single factors to form a powerful force for enhancing CEE.

Configuration S3 includes the core conditions of high economic development, high environmental regulation, and non-high technological progress, with peripheral conditions of high energy efficiency, non-high industrial structure, and non-high government support. Typical cases are Fujian and Tianjin from 2013 to 2015. This period coincided with the beginning of China’s intensified environmental governance and green transformation, and these regions had good economic foundations and faced significant environmental governance pressures. Supported by their strong economies, they could bear strict environmental regulations, respond to environmental constraints by improving energy efficiency, and achieve high CEE even without significant technological progress. Although government support was not high, the strict environmental regulations indicated that the government played more of a “regulatory” role than the role of a “promoter”. Under this model, enterprises and market entities were forced to improve their efficiency to cope with environmental constraints, which formed a “reverse forcing mechanism”. This suggests that regions may adopt different emission reduction pathways at different stages of economic transition. Even in provinces with relatively lagging technological innovation and industrial upgrading, as long as there is a solid economic foundation, excellent CEE can be achieved through strict environmental regulation and efficiency improvements. This finding provides a reference for regions in similar development stages.

4.2.2. Between Results

The BECONS adjustment distances for the four configurations are 0.109, 0.109, 0.047, and 0.062, respectively, which are all less than 0.2 and indicate no significant time effect. An examination of the consistency changes in each configuration during the study period (in Figure 2) reveals a noticeable decline in consistency for all configurations from 2020 to 2022. This decline shows a certain concentration trend rather than a random distribution and thus does not constitute a benign deviation [31]. The decline may be due to the outbreak of the COVID-19 pandemic in 2020, where the impact of the sudden and sustained public event inevitably reduced the explanatory power of regular antecedent conditions. However, as the BECONS-adjusted distances for all configurations are less than 0.2, this does not affect the overall explanatory power, and the research results still have strong universality for CEE improvement under normal conditions [50].

4.2.3. Within Results

The WICONS adjustment distances for the above four configurations are 0.259, 0.259, 0.201, and 0.230, respectively, all exceeding the threshold of 0.2 and indicating a certain degree of case effect; i.e., there is some heterogeneity in CEE improvement pathways across regions. As shown in Figure 3, an in-depth study of the WICONS of each configuration reveals that many cases cluster at a consistency equal to 1, forming data clusters. This finding indicates that most provinces have strong consistency, and some provinces may have more than one path to achieve high CEE. However, in less developed regions, such as Yunnan, Guizhou, and Neimenggu, the configurational pathways have limited explanatory power for CEE improvement, which may be related to their regional industrial structure and economic development level. On the one hand, in these regions, the energy consumption structure is dominated by fossil fuels, such as coal, and there is a relatively low proportion of clean energy, which leads to low CEE. Although Yunnan and Guizhou have abundant hydropower resources and Neimenggu has wind and solar energy resources, their development and utilization are insufficient to effectively replace fossil fuels. On the other hand, owing to the relatively low economic development levels of these provinces, local governments and enterprises have limited funds for environmental protection and energy conservation, which constrains the updating of related technologies and equipment.

4.3. Robustness Tests

To verify the robustness of the research findings, this study conducted robustness checks using various parameter adjustment schemes, following the principle proposed by Zhang et al. (2019) [51] that “results can be considered robust if they do not change substantially after minor adjustments to QCA operational parameters”. Specifically, this paper adjusted the following parameters: (1) increased the frequency threshold from 3 to 5 and (2) increased the consistency threshold from 0.8 to 0.9. The test results show that, although increasing the consistency threshold leads to a reduction in the number of configurations, the new configurations clearly exhibit a subset relationship with the original configurations. Other parameter adjustments also yield results consistent with those in Table 5, without causing substantial differences. These findings indicate that the QCA results of this study are highly robust. Owing to space limitations, these results are not presented, but they are available upon request.

4.4. Driving Effect Analysis on CEE

Current configurational studies are mostly limited to path recognition and rarely delve into the driving effects of different configurational pathways. Following the configuration valuation method of Du et al. (2024) [52], this study integrates the antecedent conditions of each configuration into a single variable and incorporates them into econometric models along with other control variables. Through multiple regression analysis, the marginal effects of each driving pathway are compared to achieve a deep validation of the configurational outcomes. To prevent multicollinearity issues caused by configurational correlations, independent regression analyses are conducted sequentially for each configuration. The results are shown in

Table 7. Configuration driving effects regression results.

Configuration	Dependent Variable CEE
	Model 1	Model 2	Model 3	Model 4
S1a	0.297 *** (0.096)
S1b		0.302 *** (0.095)
S2			0.663 *** (0.099)
S3				0.449 *** (0.140)
Controlled variable	control	control	control	control
Constant	0.886 ** (0.368)	1.044 ** (0.454)	1.044 *** (0.344)	1.136 *** (0.364)
Observed	300	300	300	300

Note: ***, ** are significant at the 1% and 5% levels, respectively, and the values in parentheses represent standard errors.

.

As shown in Table 7, all four configuration variables have a positive effect on CEE improvement at the 1% significance level. A comparison of the marginal effects of these four paths reveals that configuration S2 has the strongest driving effect, followed by S3, with S1 having the weakest effect. This finding indicates that the multifactor synergistic mechanism of technology, organization, and the environment is the most effective pathway for CEE improvement, far surpassing the simple addition of single factors. This aligns with the pooled analysis results and provides a basis for selecting CEE improvement paths. Subsequently, this study conducts robustness tests by substituting the dependent variable (replacing CEE with carbon intensity) and removing extreme values (winsorizing at the 1% level). As shown in

Table 8. Robustness tests.

Configuration	Replace the Dependent Variable				Remove Extreme Values
	Dependent Variable Carbon Intensity				Dependent Variable CEE
	Model 1	Model 2	Model 3	Model 4	Model 1	Model 2	Model 3	Model 4
S1a	−1.009 ** (0.174)				0.296 *** (0.097)
S1b		−0.977 *** (0.173)				0.299 *** (0.097)
S2			−1.587 *** (0.177)				0.672 *** (0.100)
S3				−1.929 *** (0.243)				0.454 *** (0.146)
Controlled variable	control	control	control	control	control	control	control	control
Observed	300	300	300	300	300	300	300	300

Note: ***, ** are significant at the 1% and 5% levels, respectively, and the values in parentheses represent standard errors.

, all configurations significantly reduce carbon intensity at the 1% level, and removing extreme values does not substantially alter the significance or coefficient direction of the variables. This suggests that the conclusions are robust.

5. Discussion

5.1. Results and Analysis

Improving CEE is crucial for coordinating economic development and carbon emission reduction, addressing global climate change, and achieving sustainable development. Given that CEE is influenced by numerous factors, our exploration of the impacts of various factor combinations on CEE in this study provides a novel and intriguing perspective.

Based on the TOE framework, DQCA is employed in this study to investigate the impact of six antecedent conditions on the CEE of 30 provinces (municipalities) in China. These conditions include technological progress, energy efficiency, industrial structure, government support, economic development, and environmental regulation. The results indicate that no single factor is a necessary condition for achieving high CEE levels; instead, multiple factors must work together to enhance CEE, which aligns with the conclusions of Yan et al. (2023) [27]. Additionally, this study identifies four CEE enhancement pathways. The industrial structure is a core or peripheral condition in all the pathways, which highlights its role as a key driver of CEE enhancement. Further analysis reveals that provinces have multiple pathways to improve CEE, and, interestingly, some provinces exhibit different improvement pathways at different stages of transformation. For example, Tianjin was “environment-dominated” from 2013 to 2015. However, as the government’s emphasis on green environmental protection increased and technological advancements progressed, it shifted to a “multi-equilibrium” pathway from 2016 to 2022. This further underscores that provinces need to consider their resource endowments and development stages when formulating CEE enhancement strategies and choose the most suitable pathway.

Furthermore, an additional contribution of this paper lies in the introduction of the time dimension into the research on pathways for CEE improvement. Although the results show no significant time effects across configurations, certain years exhibit concentrated changes due to unobserved factors. The significant differences in WICONS among the configurations mean that the distribution of cases explained by each configuration varies. This confirms the rationality of different provinces choosing CEE improvement pathways on the basis of their own characteristics. More importantly, this study responds to the call for a “comprehensive” methodology in complex systems theory. It integrates configuration pathways into a composite variable from a “weak holism” perspective and thereby validates the configuration pathways at a deep level. This study reveals that improvement pathways have different driving effects, with the “multi-equilibrium” pathway having the strongest driving effect on CEE. Then the “environment-dominated” pathway follows and finally the “technology–organization dual-driven” pathway. This result strongly supports the notion that the synergistic effect of multiple antecedent conditions can produce a “1 + 1 > 2” effect and provides new insights and methodological approaches for the study of complex systems.

5.2. Policy Implications

On the basis of the empirical findings of this study, the following policy implications are proposed to enhance CEE. These implications are designed not only to address current challenges but also to inform future emissions management strategies and policy planning under evolving economic and environmental conditions.

First, the core position of industrial structure optimization should be strengthened. Given the universal role of the industrial structure in all improvement pathways, industrial structure optimization should be placed at the core of efforts to improve CEE. Specifically, on the one hand, the government should accelerate the transformation and upgrading of traditional industries and promote the development of high-energy-consuming industries in cleaner and lower-carbon directions. On the other hand, strategic emerging industries and modern service sectors should be vigorously cultivated to optimize the industrial structure. This structural adjustment will lay a solid foundation for the long-term enhancement of CEE.

Second, a multifactor collaborative promotion mechanism should be established. This study finds that a single factor is insufficient to enhance CEE, which suggests the need to establish a multifactor collaborative promotion mechanism. It is recommended to increase investment in scientific and technological innovation, improve energy efficiency, and refine the environmental governance system. Additionally, the selection and application of environmental regulatory tools should be optimized, government guidance and support should be strengthened, and market-oriented incentive mechanisms should be established. More importantly, interactions among various factors should be promoted to achieve policy synergy.

Third, differentiated CEE enhancement policies should be implemented on the basis of local conditions. Regions should deploy resources in a scientifically rational way to enhance CEE, considering policy guidance and their own endowments and development stages. For economically developed regions, the “multi-equilibrium” pathway can be prioritized. This pathway can achieve continuous improvement in CEE through the coordinated advancement of technological innovation, industrial upgrading, and environmental governance. For economically underdeveloped regions, Qinghai and Hainan can serve as references regarding the development of clean energy and tourism industries to gradually increase CEE levels. Additionally, a dynamic adjustment mechanism should be established. This mechanism must regularly evaluate both policy implementation effects and the impact of emerging trends (e.g., global trade tensions, international climate agreements) on the effectiveness of the current pathway, thereby driving proactive shifts in the policy portfolio.

5.3. Limitations and Future Prospects

This study also has several limitations that warrant further exploration. First, this study establishes a foundational framework at the macro-provincial level. However, this means that potential variations at more granular levels are beyond the current scope of our analysis. Consequently, city-level and sub-sectoral analyses are critical next steps to capture local heterogeneity and formulate more precise policies. Second, CEE is influenced by multiple factors, and the six key antecedent variables selected on the basis of the TOE framework may not fully reflect all the influencing factors. In reality, various complex factors may impact CEE. Third, despite parameter sensitivity tests being conducted to ensure robustness, potential biases arising from unobserved heterogeneity and measurement errors in fuzzy-set calibration may still exist. Finally, this study uses secondary public data, and future research could include field surveys to obtain primary data for a more detailed microlevel study.

6. Conclusions

This paper uses the refined three-stage DEA method to measure the CEE of 30 provinces (municipalities) in China from 2013 to 2022 and adopts the DQCA method based on the TOE framework, to explore the pathways for improving CEE. Finally, the driving effects of each configuration are quantified by multiple linear regression. Our investigation yields the following conclusions:

First, no single factor is identified as a necessary condition for achieving high levels of CEE, and the CEE level is influenced by multiple factors; thus, it is necessary to conduct configurational analysis. In the sufficiency analysis of condition combinations, three main types of improvement pathways are identified: “technology–organization dual-driven”, “environment-dominated”, and “multi-equilibrium”. All three configurations include the industrial structure as a core or peripheral condition, which indicates that industrial structure upgrading is a key factor in enhancing CEE.

Second, the between-group results show that the improvement pathways possess temporal universality, but the BECONS collectively declined in 2020, possibly because of the external shock of the COVID-19 pandemic. The within results reveal significant regional differences in the coverage of each configuration, with some underdeveloped regions showing limited effects of configuration pathways on CEE enhancement. This provides empirical evidence for formulating differentiated CEE improvement policies.

Third, all three improvement pathways have significant effects on CEE at the 1% significance level, with the “multi-equilibrium” path exhibiting the strongest driving effect, followed by the “environment-dominated” path and finally the “technology–organization dual-driven” path. These effects remain robust to the replacement of the dependent variable and the removal of extreme values.