Eco-Efficiency and Its Evolutionary Change under Regulatory Constraints: A Case Study of Chinese Transportation Industry

Abstract

The transportation industry is characterized as a capital-intensive industry that plays a crucial role in economic and social development, and the rapid expansion of this industry has led to serious environmental problems, which makes the eco-efficiency analysis of the transportation industry an important issue. Previous research paid little attention to the regulatory scenarios and suffered from the incomparability problem, hence this paper aims to reasonably estimate the eco-efficiency and identify its evolutionary characteristics. We measure the eco-efficiency and the corresponding global Malmquist–Luenberger productivity index using a modified model of the data envelopment analysis framework, in which different regulatory constraints are incorporated. Based on the empirical study on the transportation industry of thirty provinces in China, we find that the eco-efficiency of Chinese transportation industry experienced a slight increase during 2015–2016, a sharp decline during 2016–2017, and a continuous rise since year 2017. The Middle Yangtze River area was the best performer among the eight regions in terms of eco-efficiency, while the Southwest area was placed last. The global Malmquist–Luenberger productivity index showed an earlier increase and later decrease trend, which was quite consistent with the reality of the variation of inputs and outputs and the emergence of COVID-19. Moreover, the best practice gap change was found to be the main driven force of productivity. The empirical results verify the practicability of our measurement models and the conclusions can be adopted in guiding the formulation of corresponding policies and regulations.

1. Introduction

The transportation industry has long served as a strategic and basic industry, composed of a long industrial chain and a huge amount of fixed assets. According to the statistics from the China Statistical Yearbook, the total investment in the transportation fixed assets in China increased from about 2.05 trillion RMB in 2011 to 3.48 trillion RMB in 2020, with more than 50% cumulative and 4.4% annual average growth rate. With the vigorous development of the transportation industry, there was an enormous increase in energy consumption which led to environmental problems, such as climate change and environmental pollution. The International Energy Agency also stated that the transportation industry is the world’s largest carbon emitting industry and the main cause of global climate change. To deal with these global environmental problems, many international conventions and documents devoted to emission reduction have been introduced, e.g., the United Nations Framework Convention on Climate Change, the Kyoto Protocol, the Copenhagen Accord, etc. In the context of global climate change, China put forward the goal of carbon peaking and carbon neutrality. China pledged to gradually reduce carbon dioxide emissions before it reaches the peak by 2030 and to offset them altogether by planting trees, saving energy, and reducing emissions before 2060. Therefore, for the sustainable development of the transportation industry and Chinese economy, it is necessary to consider the improvements in transportation capacity, energy conservation, and emission reduction, which means that it is urgent to improve the eco-efficiency of this industry.

Currently, environmental efficiency or eco-efficiency evaluation has become one of the important topics in academia, and it includes many different methods, such as multi-attribute decision-making (MADM, Gohari et al. [1]; Sheu [2]; Zhang et al. [3]), analytic hierarchy process (AHP, Garg et al. [4]), stochastic frontier analysis (SFA, Wanke et al. [5]; Liu et al. [6]), and data envelopment analysis (DEA, Hampf and Rødseth [7]; Xia et al. [8]). Among these evaluation methods, DEA became the most popular because it is applicable to evaluate the efficiency of decision-making units (DMUs) with multiple inputs and multiple outputs. Moreover, it not only dispenses the preset of parameters and specific models of the production function but also effectively avoids the bias caused by function errors. However, DEA-based efficiency values suffer from the problem of incomparability across periods (Song et al. [9]). This is because of the incorporation of different technology frontiers in different periods. Hence, even the same input–output point may have different efficiencies based on frontiers in different periods. To solve this problem, scholars have proposed several productivity indices to reveal the evolutionary change of efficiencies, e.g., the Malmquist productivity index (MPI). Furthermore, to deal with scenarios with undesirable outputs, the Luenberger productivity index and the Malmquist–Luenberger productivity index were also proposed based on the directional distance function (DDF, Chung et al. [10]). Nevertheless, existing performance evaluation studies in the field of transportation paid little attention to this incomparability problem. Therefore, it is necessary to supplement relevant studies to improve the rationality and accuracy of the various performance evaluations in this industry.

Actually, the eco-efficiency measurement results are highly affected by the application of different environmental policies and regulations. For example, in the transportation industry, on one hand, enterprises adopting low-carbon technologies could become more competitive and cost-saving to get more efficient as it expands. On the other hand, enterprises under environmental control are more willing to carry out technological research and development activities and thereby promote environmental and production efficiency. Hence, under the guidance of different policies, the regional efficient outputs and environmental control outcomes could be quite different. However, most of the existing research ignores the impact of these environmental regulations, which will mislead the evaluation of eco-efficiency and the corresponding policy formulation. Hence, this paper aims to propose a DEA-based eco-efficiency measurement model that incorporates the regulatory constraints and conducts an empirical analysis of Chinese transportation industry. This paper advances previous investigations in two ways. First, our new model better constructs the input–output structure under real environmental regulations within the Chinese transportation industry. Second, the combination of both eco-efficiency and the productivity index enhances the comparability of eco-efficiency evaluation results based on DEA.

The rest of this paper unfolds as follows: Section 2 reviews the corresponding previous studies. Section 3 presents our proposed eco-efficiency measurement model and the formation of the global Malmquist–Luenberger productivity index. Section 4 describes the setting and results of the empirical study, and Section 5 concludes this paper.

2. Literature Review

2.1. Performance Evaluation of the Transportation Industry

The transportation industry has long been characterized as a capital-intensive industry that depends heavily on natural conditions, and the development of this industry is conducive to driving the regional economy, narrowing regional gaps, and achieving coordinated economic and social development. With the advancement of urbanization and industrialization, the transportation industry has been experiencing rapid development, but it is also accompanied by serious environmental pollution problems (Xie et al. [11]). Hence, many studies on the performance evaluation of this industry have been carried out to find feasible paths for its sustainable and efficient development (Lior [12]).

To conduct the performance evaluation in various scenarios, different research methods have been adopted, e.g., MADM, AHP, SFA, DEA, etc. The first group of studies employed the most widely used DEA method in assessing the transformation performance. Xia et al. [8] proposed a meta-frontier DEA-based decomposition method to measure the spatial carbon intensity inequality, decomposing the carbon intensity inequality of China’s transportation industry from 2004 to 2018. Beltrán-Esteve et al. [13] used DEA and DDF models in analyzing the environmental performance changes of the transportation industry in 38 countries during 1995–2019 and identified the Luenberger productivity index and its determinants of environmental performance changes. Cui and Li [14] used a virtual frontier DEA model to assess the carbon efficiency of transportation in 15 countries. Choi et al. [15] employed the DEA-based DDF in measuring the environmental efficiency of the Korean freight transportation industry, and they also tested the regulatory cost. Yang et al. [16] proposed a life-cycle DEA approach to measure the atmospheric environmental efficiency of the Chinese transport sector. Egilmez and Park [17] combined both economic input–output life cycle assessment and DEA to assess sustainability performance for the carbon, energy, and water footprints associated with transportation in the U.S. manufacturing sector. Gupta et al. [18] used AHP and DEA methods to study the sustainable transportation mode of the Indian mining industry and established a comprehensive multi-objective optimization model. Du et al. [19] modified the three-stage DEA measurement model and assessed the transport carbon efficiency of 52 Belt and Road Initiative countries from 2005 to 2017.

The second group of studies used the parametric regression model, SFA, to conduct the transportation performance evaluation research. Liu et al. [6] employed the SFA method in measuring the total factor energy efficiency and identifying its driving factors for China’s road transport industry. Considering the carbon inequality, Wanke et al. [5] proposed a novel robust Bayesian SFA approach to measure the sustainable efficiency of China’s transportation system. Bai et al. [20] proposed a parameter element frontier analysis method, explored the carbon emission performance and emission reduction potential of the transportation sector in various provinces in China, and proposed corresponding emission reduction strategies. Martín et al. [21] evaluated the Spanish airport efficiency using both the Markov chain Monte Carlo simulation and the SFA model and verified a significant level of inefficiency in airport operations. Cullinane et al. [22] applied both DEA and SFA approaches to compare the efficiency of the world’s largest container ports and revealed a high degree of correlation.

The third group of studies is conducted based on several prevalent decision-making models, such as AHP and MADM. Zhang et al. [3] proposed an interval 2-tuple language MADM method in ranking the zero-carbon measures for the sustainable freight. Garg et al. [4] evaluated the sustainable development of the container freight railway in India using a fuzzy AHP method. In the logistics industry with complex and uncertain supply and demand conditions, Sheu [2] proposed a hybrid method combining both fuzzy AHP and MADM to identify global logistics strategies.

2.2. Eco-Efficiency Evaluation Based on DEA

DEA is a nonparametric approach established on mathematical programming models for measuring the relative efficiency of DMUs. Charnes et al. [23] proposed the seminal DEA model and defined relative efficiency as the ratio of weighted outputs to weighted inputs. Banker et al. [24] further changed the constant returns to scale assumption used in Charnes et al. [23] into the variable return to scale assumption and gave rise to the BCC model. To fit the evaluation needs of various new scenarios and new fields, a series of advanced DEA models were proposed based on these two classic DEA modes, e.g., stochastic DEA (Sengupta [25]), network DEA (Färe and Grosskopf [26]), slacks-based model (Tone [27]), etc. Since the DEA method does not need to assume the specific production function form and parameter distribution in advance, it overcomes the influence of subjectivity on the measurement results. Hence, it gradually becomes a strong and effective analytical tool in the field of economics and management, especially in performance evaluation in various fields.

The increasingly severe problems of resource shortage and environmental pollution have attracted great attention around the world, and more and more researchers have realized the importance of introducing undesirable outputs in the process of evaluating organizational performance. In the framework of DEA, there are five main methods of dealing with undesirable outputs that appeared in the environmental efficiency or eco-efficiency evaluation. (1) The first method is termed the input method, which treats undesired outputs as strong disposable inputs and incorporates them into the production function. Considering the impact of environmental factors on productivity, Hailu and Veeman [28] incorporated pollutant outputs in the efficiency and productivity analysis and conducted the environmental efficiency analysis in the field of agriculture. (2) The second method is the data transformation method, which combines the life cycle assessment (LCA) with DEA and treats undesirable outputs as desirable outputs only through the kinds of data transformation approaches. To evaluate the efficiency of winter wheat cultivation systems in Poland, Pishgar-Komleh et al. [29] employed six transformation methods to deal with undesirable outputs. (3) The weak disposability assumption is the third type of undesirable outputs modeling method, which supposes that desirable and undesirable outputs are closely interrelated. Hence, the reduction in undesirable outputs comes at the cost of a simultaneous reduction in desirable outputs. Färe and Grosskopf [30] verified the rationality of modeling undesirable outputs in DEA based on the weak disposability assumption. Wu et al. [31] conducted eco-efficiency measurements based on a stochastic DEA model using the weak disposability assumption to incorporate undesirable outputs. (4) The fourth method is called weak G-disposability and materials balance principles. Chung et al. [10] defined the G-disposability based on the commonly used DDF, and many scholars became accustomed to combining DDF with DEA models to conduct environmental efficiency analysis. Wang et al. [32] combined the principle of material balance and DEA method to evaluate the environmental efficiency and emission reduction efficiency of China’s thermal power industry. (5) The fifth method is natural and management disposability. Murty [33] considered that previous treatment methods of undesirable outputs are based on a single production system, which is insufficient to illustrate the balanced relationship between desirable and undesirable outputs, so he proposed the by-production technique and provided a theoretical basis for modeling pollution generation technology in multiple production systems. Sueyoshi et al. [34] employed the non-radial distance adjustment measures to propose the concepts of natural disposability and management disposability and explored the adaptation potential of companies in pollution reduction.

Due to the continuous improvement of social environmental awareness, eco-efficiency evaluation in various fields has become an important issue in academia. Nodin et al. [35] evaluated the eco-efficiency of rice self-sufficiency in Malaysia from 2009 to 2018 using the non-radial slacks-based model combined with the Malmquist–Luenberger index, considering undesirable outputs. Wang et al. [36] proposed a hybrid super-efficiency DEA model and introduced the Malmquist index to analyze the eco-efficiency of 22 industrial sectors in China. Zuo et al. [37] adopted a two-stage DEA model in analyzing the technological innovation efficiency and eco-efficiency of the mining technology in 30 provinces in China from 2008 to 2018. Zhang et al. [38] proposed a dynamic series-parallel recycling DEA model based on DDFs and assessed the industrial efficiency of 30 provinces in China from 2011 to 2019. Liang et al. [39] used a two-stage network DEA model in measuring the water resource efficiency of 11 provinces in western China during 2008–2017. Kutty et al. [40] evaluated the sustainable development performance of 35 leading smart cities in Europe during 2015–2020 using a new DEA model based on the double frontiers.

In summary, the rapid development of the transportation industry and the consequent environmental pollution problems have made the eco-efficiency assessment an important topic. However, existing investigations pay little attention to the aforementioned problems, and this may mislead the corresponding industry development and policymaking. First, DEA-based eco-efficiency evaluation results are incomparable across periods. Second, few studies considered the constraints of environmental regulations in their models, which could lead to deviations between the evaluation results and the real situation. Therefore, this paper aims to propose eco-efficiency measurement models and conduct evolutionary investigations on them. Meanwhile, environmental regulations will be introduced in our models to make them more consistent with reality. On this basis, we will also carry out empirical research on Chinese transportation industry.

3. Methodology

To solve the abovementioned problems, in this section, we first introduce the eco-efficiency measurement model, in which different environmental regulations can be reflected through the setting of the direction vector. We further construct the Malmquist–Luenberger productivity index to explore the evolutionary change in eco-efficiency.

3.1. Eco-Efficiency under Regulatory Constraints

Suppose there are $n$ DMUs being evaluated, and each of them uses input $x^t$ to produce desirable outputs $y^t$ and undesirable outputs $b^t$ during period $t\left(1,2,\dots ,T\right)$, so the environmental production possibility set (PPS) during period $t$ can be expressed as $P^t=\left\{\left(x^t,y^t,b^t\right)\left|x^t \mathrm{can} \mathrm{produce} \left(y^t,b^t\right)\right.\right\}$. To incorporate undesirable outputs in DEA models reasonably, different techniques have been proposed by scholars, e.g., the weak disposability assumption, the cost disposability assumption, and the by-production modeling method (Seiford and Zhu, [41]; Dakpo et al. [42]). The weak disposability assumption establishes a linkage between desirable outputs and undesirable outputs and limits the increase in desirable outputs to grow independently of undesirable outputs. Hence, based on the weak disposability assumption and the null-jointness assumption, the environmental PPS can be transformed as:

$$P^t=\left\{\left(x^t,y^t,b^t\right)\left|\begin{array}{l}{\displaystyle {\sum }_{j=1}^n{\lambda }_j{x}_{ij}^t\le x_i^t},\forall i\\ {\displaystyle {\sum }_{j=1}^n{\theta }_j{\lambda }_j{y}_{rj}^t\le y_r^t},\forall r\\ {\displaystyle {\sum }_{j=1}^n{\theta }_j{\lambda }_j{b}_{lj}^t=b_l^t},\forall l\\ {\displaystyle {\sum }_{j=1}^n{\lambda }_j=1}\\ {\lambda }_j\ge 0,\forall j\\ 0\le {\theta }_j\le 1,\forall j\end{array}\right.\right\}$$

(1)

where, ${\lambda }_j$ represents the intensity vector, and ${\theta }_j$ is the abatement factor that relates desirable and undesirable outputs.

It is difficult for traditional radial DEA models to achieve an increase in desirable outputs and a decrease in undesirable outputs, simultaneously. Hence, we further introduce the DDF to measure the eco-efficiency of DMU_k as:

$$\begin{array}{ll}\hfill \overrightarrow{D}& \left(x_k^t,y_k^t,b_k^t;g\right)=\max \beta \hfill \\ \hfill s.t.& {\displaystyle {\sum }_{j=1}^n{\lambda }_j{x}_{ij}^t\le x_{ik}^t},\forall i\hfill \\ & {\displaystyle {\sum }_{j=1}^n{\theta }_j{\lambda }_j{y}_{rj}^t\ge y_{rk}^t+}\beta g_r^Y,\forall r\hfill \\ & {\displaystyle {\sum }_{j=1}^n{\theta }_j{\lambda }_j{b}_{lj}^t=b_{lk}^t-\beta g_l^B},\forall l\hfill \\ & {\displaystyle {\sum }_{j=1}^n{\lambda }_j=1}\hfill \\ & {\lambda }_j\ge 0,\forall j;0\le {\theta }_j\le 1,\forall j; \beta free\hfill \end{array}$$

(2)

where, $g=(g_{}^Y,g_{}^B)$ is the exogenous direction vector that determines the improvement direction of desirable and undesirable outputs towards the technology frontier. Hence, the choice of the direction vector reflects the specific environmental regulation orientation. In reality, the industrial development regulations in different regions of China are determined by local policymakers, so there may be differences in environmental regulations among regions. Here, we present two major environmental regulatory scenarios, the win-win policy orientation and the environmental priority orientation, which is illustrated in Figure 1. Scenario one, the win-win policy orientation, describes a situation in which the government is simultaneously guiding an increase in good outputs and a decrease in bad outputs. This scenario is fairly common because most governments are committed to improving the environment without dragging down the economy. Correspondingly, the environmental priority orientation, scenario two, identifies environmental improvement as the primary goal. This scenario applies to several parts of China where the environment has been severely damaged. Hence, for a specific evaluated DMU₀, with $\left(y_0,b_0\right)$ being its output vector, $g_1=\left(y_0,-b_0\right)$ applies to scenario one and $g_2=\left(0,-b_0\right)$ applies to scenario two.

It should be noted that in addition to these two kinds of direction vectors mentioned above, there have been studies using other settings of direction vectors, e.g., $g=\left(1,-1\right)$ and $g=\left(y_0,0\right)$. We do not adopt these direction vectors, because it is necessary to set the direction of $g_{}^B$ as the actual value of undesirable outputs precisely. In this manner, the DDF reflects the extent to which the actual undesirable output can be reduced, i.e., the environmental inefficiency (please see similar explanations in Song et al. [43]).

Due to the incorporation of the abatement factor, model (2) becomes a nonlinear programming. Thus, we let ${\mu }_j={\theta }_j{\lambda }_j$ and ${\nu }_j=\left(1-{\theta }_j\right){\lambda }_j$, so ${\lambda }_j={\mu }_j+{\nu }_j$ holds. Hence, we can transform the DDF as:

$$\begin{array}{ll}\hfill \overrightarrow{D}& \left(x_k^t,y_k^t,b_k^t;g\right)=\max \beta \hfill \\ \hfill s.t.& {\displaystyle {\sum }_{j=1}^n\left({\mu }_j+{\nu }_j\right)x_{ij}^t\le x_{ik}^t},\forall i\hfill \\ & {\displaystyle {\sum }_{j=1}^n{\mu }_j{y}_{rj}^t\ge y_{rk}^t+}\beta g_r^Y,\forall r\hfill \\ & {\displaystyle {\sum }_{j=1}^n{\mu }_j{b}_{lj}^t=b_{lk}^t-\beta g_l^B},\forall l\hfill \\ & {\displaystyle {\sum }_{j=1}^n\left({\mu }_j+{\nu }_j\right)=1};{\mu }_j+{\nu }_j\ge 0,\forall j\hfill \\ & {\mu }_j\ge 0,\forall j;{\nu }_j\ge 0,\forall j; \beta free\hfill \end{array}$$

(3)

The optimal solution of model (3), ${\overrightarrow{D}}^{*}\left(x_k^t,y_k^t,b_k^t;g\right)$, represents the distance of the evaluated DMU_k from the technology frontier alongside the specific improvement direction, the environmental inefficiency. Using the direction vector of $g_1=\left(y_0,-b_0\right)$, we can define the eco-efficiency in the win-win policy orientation scenario as ${\vartheta }_k^{t,ww}=1-{\overrightarrow{D}}^{*}\left(x_k^t,y_k^t,b_k^t;y_0,-b_0\right)$. Similarly, if we use the direction vector of $g_2=\left(0,-b_0\right)$, the eco-efficiency in the environmental priority orientation scenario can be defined as ${\vartheta }_k^{t,ep}=1-{\overrightarrow{D}}^{*}\left(x_k^t,y_k^t,b_k^t;0,-b_0\right)$.

3.2. Global Malmquist–Luenberger Productivity Index

Even though we have obtained the eco-efficiency in two scenarios based on DEA models, the efficiency scores do not have comparability among different periods (Song et al. [9]). To conduct intertemporal comparison among periods or the evolutionary change research on eco-efficiency, we need to resort to the prevalent productivity indices. Färe et al. [44,45] proposed the global Malmquist–Luenberger productivity index (GMLPI) based on the convex union of DMUs of all periods, and this index can identify the efficiency change over time in cases with undesirable outputs. Moreover, due to the use of the global technique, the GMLPI avoids the infeasibility problem that might occur when calculating the Malmquist–Luenberger productivity index (Oh and Heshmati [46]).

We first construct the global PPS through the convex union of observations of all periods as $P^G=conv\left\{P^1,P^2,\dots ,P^T\right\}$. With the global PPS as the reference set, the original DDF can be rewritten as:

$$\begin{array}{ll}\hfill {\overrightarrow{D}}^G& \left(x_k^t,y_k^t,b_k^t;g\right)=\max \beta \hfill \\ \hfill s.t.& {\displaystyle {\sum }_{t=1}^T{\displaystyle {\sum }_{j=1}^n\left({\mu }_j+{\nu }_j\right)x_{ij}^t\le x_{ik}^t}},\forall i\hfill \\ & {\displaystyle {\sum }_{t=1}^T{\displaystyle {\sum }_{j=1}^n{\mu }_j{y}_{rj}^t\ge y_{rk}^t+}\beta g_r^Y},\forall r\hfill \\ & {\displaystyle {\sum }_{t=1}^T{\displaystyle {\sum }_{j=1}^n{\mu }_j{b}_{lj}^t=b_{lk}^t-\beta g_l^B}},\forall l\hfill \\ & {\displaystyle {\sum }_{t=1}^T{\displaystyle {\sum }_{j=1}^n\left({\mu }_j+{\nu }_j\right)=1}};{\mu }_j+{\nu }_j\ge 0,\forall j\hfill \\ & {\mu }_j\ge 0,\forall j;{\nu }_j\ge 0,\forall j; \beta free\hfill \end{array}$$

(4)

where, ${\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)$ reflects the extent to which DMU_k can improve its desirable and undesirable outputs from the range of all DMUs and all periods. Referencing the formulation of the productivity index proposed by Färe et al. [44], we can further define the GMLPI index as:

$$GMLPI\left(x_k^t,y_k^t,b_k^t,x_k^{t+1},y_k^{t+1},b_k^{t+1}\right)=\frac{1+{\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)}{1+{\overrightarrow{D}}^G\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)}$$

(5)

Consequently, the result of $GMLPI\left(x_k^t,y_k^t,b_k^t,x_k^{t+1},y_k^{t+1},b_k^{t+1}\right)>1$ indicates that DMU_k obtained productivity improvement from period $t$ to period $t+1$, while results of less than unity and equal to unity represent a decline in the productivity and a maintenance of the status quo. To explore the internal causes of the productivity or efficiency evolution, we further decompose the GMLPI index as follows:

$$\begin{array}{l}GMLPI\left(x_k^t,y_k^t,b_k^t,x_k^{t+1},y_k^{t+1},b_k^{t+1}\right)=\frac{1+{\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)}{1+{\overrightarrow{D}}^G\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)}\\ =\frac{1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)}{1+{\overrightarrow{D}}^{t+1}\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)}\times \frac{\left(1+{\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)\right)/\left(1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)\right)}{\left(1+{\overrightarrow{D}}^G\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)\right)/\left(1+{\overrightarrow{D}}^{t+1}\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)\right)}\\ =\frac{T{E}^{t+1}}{T{E}^t}\times \frac{BP{G}^{t+1}}{BP{G}^t}=E{C}^{t,t+1}\times BP{C}^{t,t+1}\end{array}$$

(6)

where, the technical efficiency (TE) during period $t$ is defined as $T{E}^t=\frac{1}{1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)}$. The ratio of TE of two adjacent periods represents the efficiency change (EC), and it is denoted as $E{C}^{t,t+1}=\frac{1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)}{1+{\overrightarrow{D}}^{t+1}\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)}$. Moreover, changes in productivity may also result from changes in technology, so we define the best practice gap (BPG) between the contemporaneous and the global technology frontiers as $BP{G}^t=\frac{1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)}{1+{\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)}$. Therefore, the best practice gap change (BPC), $BP{C}^{t,t+1}=\frac{\left(1+{\overrightarrow{D}}^G\left(x_k^t,y_k^t,b_k^t;g\right)\right)/\left(1+{\overrightarrow{D}}^t\left(x_k^t,y_k^t,b_k^t;g\right)\right)}{\left(1+{\overrightarrow{D}}^G\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)\right)/\left(1+{\overrightarrow{D}}^{t+1}\left(x_k^{t+1},y_k^{t+1},b_k^{t+1};g\right)\right)}$, reflects the technical change from period $t$ to period $t+1$.

Concretely, the formulation of the GMLPI index is the same under different regulatory scenarios, while there could be significant differences in the efficiency and GMLPI measurement results due to the use of different direction vectors. In the empirical analysis part, we will compare the results caused by different environmental regulation scenarios.

4. Empirical Study

4.1. Data Description

The selection of input and output variables plays an important role in evaluating eco-efficiency reasonably, so this paper follows the variables commonly used in previous eco-efficiency research in the transportation industry (e.g., Zhang et al. [47]; Cui and Li [48]; Mavi et al. [49]; and Zhang et al. [50]). Our employed input variables are divided into three main factors, including assets, labor, and energy. The fixed asset investment is the proxy of assets, which reflects the total cost of new construction, expansion, reconstruction, relocation, and repair within the transportation sector during a certain time. Employees, whether part-time or full-time, are considered as the labor force as long as they obtain remuneration from this industry. Energy is another important source of power for the transportation industry, so the amount of energy consumption measured using the standard coal is employed as the proxy of energy. In terms of the outputs, we take both passenger and freight into consideration. To reflect both transport capacity and distance, we use passenger turnover volume and freight turnover volume as two desirable outputs, respectively. As for the undesirable output, we select CO₂ emission, the most important pollution, by referring to numerous papers. Detailed information about these input and output variables and their data sources are listed in

Table 1. Detailed information about input and output variables.

Variables	Types	Units	Data Sources
Fixed asset investment	Input	Million RMB	China Transportation Statistical Yearbook 2016–2021
Employees	Input	Thousand person	China Transportation Statistical Yearbook 2016–2021
Energy consumption	Input	Thousand tonne	China Energy Statistical Yearbook 2016–2021
Passenger turnover volume	Desirable output	Million person kilometer	China Transportation Statistical Yearbook 2016–2021
Freight turnover volume	Desirable output	Million tonne kilometer	China Transportation Statistical Yearbook 2016–2021
CO2 emission	Undesirable output	Thousand tonne	Calculated referencing the methods provided by Intergovernmental Panel on Climate Change Guidelines

. It is worth mentioning that data on energy consumption and CO₂ emission are not available directly, so they are calculated based on the prevalent methods provided by the Intergovernmental Panel on Climate Change Guidelines using data on coal, oil, and natural gas. Moreover, due to data availability, thirty provinces of mainland China, excluding Tibet, Hong Kong, Macao, and Taiwan, are considered as our evaluated DMUs in this paper. The intertemporal eco-efficiency assessment brings in a problem of inflation in the monetary variable. Hence, we collect the consumer price index (CPI) data from the OECD database to deflate the data of fixed asset investment.

Table 2. The CPI data employed in the deflation process.

Years	2015	2016	2017	2018	2019	2020
CPI	100.00	102.00	103.63	105.78	108.84	111.48

shows the CPI data concretely, using 2015 as the base year.

After gathering and deflating the original data, we further present the descriptive statistics of all variables in

Table 3. Descriptive statistics of input and output variables.

Year	Statistics	Fixed Asset Investment (Million RMB)	Employees (Thousand Person)	Energy Consumption (Thousand Tonne)	Passenger Turnover Volume (Million Person Kilometer)	Freight Turnover Volume (Million Tonne Kilometer)	CO2 Emission (Thousand Tonne)
2015	Max	115,833.11	613.16	28,160.34	179,651.76	1,949,588.47	60,217.39
	Min	1499.84	32.31	1462.46	11,292.78	44,557.70	3043.52
	Mean	54,522.35	226.86	10,407.94	75,793.54	521,803.20	22,451.39
	Median	57,710.67	202.55	9471.79	63,894.44	359,601.30	20,008.64
	Standard deviation	31,504.31	130.66	5740.58	47,264.96	464,050.98	12,287.61
2016	Max	147,058.82	635.53	31,536.66	188,744.51	2,180,164.99	67,454.11
	Min	12,320.96	31.76	1682.63	10,971.96	47,579.99	3496.86
	Mean	63,004.11	227.20	10,925.54	76,135.17	555,858.42	23,484.26
	Median	57,303.29	201.31	9633.08	64,412.46	373,160.37	20,279.49
	Standard deviation	38,654.86	132.44	6379.10	48,423.73	526,776.80	13,598.60
2017	Max	180,704.43	641.73	31,973.98	201,247.23	2,791,979.08	68,388.00
	Min	5311.43	32.28	1729.14	9919.00	51,946.25	3637.30
	Mean	71,217.41	226.34	11,285.63	77,516.68	632,243.72	24,200.59
	Median	65,212.94	199.44	9901.98	66,988.95	420,118.40	21,510.37
	Standard deviation	47,516.35	132.42	6720.49	50,591.52	651,675.57	14,324.26
2018	Max	177,941.24	631.96	32,583.71	208,558.68	2,833,832.92	69,696.31
	Min	8236.09	34.04	1515.78	8830.62	55,135.92	3131.92
	Mean	71,530.10	216.83	11,437.93	78,196.70	662,951.29	24,385.66
	Median	62,501.30	188.01	10,037.10	71,633.86	443,800.07	21,010.05
	Standard deviation	48,063.55	127.35	6858.28	51,513.08	687,040.45	14,666.19
2019	Max	215,077.86	630.23	32,924.97	212,572.02	3,032,490.16	70,535.38
	Min	7508.23	33.82	1684.82	8703.60	39,842.89	3441.57
	Mean	70,423.05	216.98	11,753.12	78,662.11	645,077.81	24,994.64
	Median	61,295.74	187.90	9800.51	72,544.81	373,721.31	20,497.62
	Standard deviation	53,287.94	124.66	7057.53	52,189.18	702,288.86	15,108.38
2020	Max	237,061.92	616.67	29,002.17	119,090.73	3,279,499.58	62,128.00
	Min	8770.15	34.28	1600.01	5251.05	41,573.93	3263.76
	Mean	70,483.57	214.63	10,508.83	43,044.09	654,087.99	22,342.16
	Median	57,493.80	195.51	8323.82	40,526.86	385,406.46	17,540.49
	Standard deviation	54,761.39	122.25	6335.88	29,196.03	728,058.87	13,563.25

to show their distribution features. It is obvious that data of most variables were fairly evenly distributed and remained stable, while the passenger turnover volume experienced a sharp decline in year 2020. This phenomenon was mainly due to the outbreak of COVID-19. Additionallly, the average and median scores of the freight turnover volume showed large deviations for each year, indicating that there were significant differences in the freight traffic among different regions. We also depict the line chart in Figure 2 to exhibit the trends for the average value of input and output variables. We can see that employees declined slightly over these six years. Energy consumption and CO₂ emissions continued to fluctuate within a correspondingly small amplitude. Due to the rapid expansion of infrastructures and e-commerce in recent years, fixed asset investment and freight turnover volume have risen sharply. Finally, the most obvious feature in this figure is the sharp decline in the passenger turnover volume. It decreased by more than 45% in year 2020, illustrating the devastating impact of the pandemic on the passenger transport sector. All these characteristics indicate that divergent input–output patterns exist in Chinese transportation industry among different regions, which can cause large differences in eco-efficiency.

4.2. Empirical Results

4.2.1. Analysis of Eco-Efficiency Results

Using the DDF model (3), we first conduct the eco-efficiency measurement in the win-win policy orientation scenario. Based on the specific direction vector $g_1=\left(y_0,-b_0\right)$ and the definition of eco-efficiency ${\vartheta }_k^{t,ww}=1-{\overrightarrow{D}}^{*}\left(x_k^t,y_k^t,b_k^t;y_0,-b_0\right)$, we present the eco-efficiency of 30 provinces during 6 years in

Table 4. Eco-efficiency of different provinces in the win-win policy orientation scenario.

Provinces	2015	2016	2017	2018	2019	2020	Average
Beijing	1.0000	1.0000	0.2462	0.2686	0.4127	0.5027	0.5717
Tianjin	0.7527	1.0000	1.0000	1.0000	1.0000	1.0000	0.9588
Hebei	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Shanxi	0.5965	0.6970	0.5149	0.4705	0.5909	0.4984	0.5614
Inner Mongolia	0.3491	0.5491	0.5365	0.5849	0.5549	0.4348	0.5015
Liaoning	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Jilin	0.5596	0.6278	0.7036	0.8247	0.7914	0.8739	0.7302
Heilongjiang	1.0000	0.8695	0.5166	0.6790	0.6370	0.4901	0.6987
Shanghai	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Jiangsu	0.9304	1.0000	1.0000	0.8440	0.9731	1.0000	0.9579
Zhejiang	0.8567	0.8109	0.7385	0.7574	0.8732	0.8773	0.8190
Anhui	1.0000	1.0000	0.9449	1.0000	1.0000	1.0000	0.9908
Fujian	0.5466	0.6062	0.5514	0.5757	0.6987	0.6799	0.6097
Jiangxi	1.0000	1.0000	1.0000	0.9517	1.0000	1.0000	0.9919
Shandong	0.7593	0.7274	0.6831	0.7242	0.7450	0.7027	0.7236
Henan	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Hubei	0.8008	0.7653	0.7536	0.7301	0.7554	0.5916	0.7328
Hunan	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Guangdong	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Guangxi	0.7706	0.7766	0.7569	0.8059	0.8230	0.9886	0.8203
Hainan	1.0000	0.8368	0.7747	0.6410	0.8615	1.0000	0.8523
Chongqing	0.6076	0.5741	0.4999	0.5487	0.5468	0.4742	0.5419
Sichuan	0.8010	0.6792	0.5711	0.5919	0.5900	0.6982	0.6552
Guizhou	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Yunnan	0.4759	0.4642	0.4342	0.3890	0.3777	0.3718	0.4188
Shaanxi	0.7869	0.9262	0.8745	0.8638	0.8981	0.9799	0.8882
Gansu	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Qinghai	1.0000	1.0000	1.0000	0.9144	0.8361	0.9375	0.9480
Ningxia	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Xinjiang	0.6706	0.5653	0.4260	0.4194	0.4433	0.3555	0.4800
Average	0.8421	0.8492	0.7842	0.7862	0.8136	0.8152	0.8151

. According to the results, we find that nine provinces have maintained the eco-efficiency of unity for six years, including Hebei, Liaoning, Shanghai, Henan, Hunan, Guangdong, Guizhou, Gansu, and Ningxia. This shows that these provinces have achieved environmental protection while maintaining high transportation capacity, so they are in the top tier among all the evaluated provinces. Additionally, the average eco-efficiency of all provinces across all examined years was 0.8151, which was a relatively high score and represents an average improvement space of 19.49% for the transportation capacity enhancement and environmental pollution reduction. Apart from the nine provinces within the first tier, nine provinces exceeded the average eco-efficiency, i.e., Tianjin, Jiangsu, Zhejiang, Anhui, Jiangxi, Guangxi, Hainan, Shaanxi, and Qinghai. These provinces are in the second tier according to the eco-efficiency and can continue to improve their environmental protection standards. Eco-efficiency in the remaining 12 provinces was below average, so these provinces had great potential for both transport capacity and environmental improvement. The provinces in the third tier include Beijing, Shanxi, Inner Mongolia, Jilin, Heilongjiang, Fujian, Shandong, Hubei, Chongqing, Sichuan, Yunnan, and Xinjiang. It is worth noting that Beijing’s eco-efficiency was the most unstable one, which experienced a rapid decline from 2016 to 2017 and remained at a low level until 2020. The underlying reasons for this result can be found in Beijing’s input and output data. Concretely, the fixed asset investment of Beijing was 1499.84, 12,567.38, 25,449.07, 23,124.44, 14,662.99, and 11,776.99 million RMB during 2015–2020, respectively. Meanwhile, the passenger turnover volume was 27,943.16, 26,849.40, 25,315.90, 25,443.28, 26,368.31, and 11,436.44 million person kilometer during 2015–2020, respectively. Data on the freight turnover volume had similar changing trends as the passenger turnover volume. This illustrates that Beijing greatly increased its fixed asset investment in 2017 and 2018, while there was no significant improvement in its transport outputs, leading to the reduction in eco-efficiency.

Furthermore, we employ the direction vector of $g_2=\left(0,-b_0\right)$ and DDF model in the second scenario for calculation, and eco-efficiency results ${\vartheta }_k^{t,ep}=1-{\overrightarrow{D}}^{*}\left(x_k^t,y_k^t,b_k^t;0,-b_0\right)$ are shown in

Table 5. Eco-efficiency of different provinces in the environmental priority policy orientation scenario.

Provinces	2015	2016	2017	2018	2019	2020	Average
Beijing	1.0000	1.0000	0.1373	0.1321	0.2081	0.3223	0.4666
Tianjin	0.4049	1.0000	1.0000	1.0000	1.0000	1.0000	0.9008
Hebei	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Shanxi	0.2734	0.4741	0.2711	0.3076	0.4193	0.3319	0.3462
Inner Mongolia	0.2115	0.3784	0.3666	0.4133	0.3692	0.2346	0.3290
Liaoning	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Jilin	0.3181	0.3707	0.3948	0.6446	0.6124	0.7372	0.5130
Heilongjiang	0.4317	0.4261	0.2963	0.4514	0.3925	0.2292	0.3712
Shanghai	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Jiangsu	0.5772	0.5718	0.5150	0.5478	0.5312	1.0000	0.6238
Zhejiang	0.4998	0.5650	0.4819	0.5115	0.6247	0.5605	0.5406
Anhui	1.0000	1.0000	0.7159	0.7197	1.0000	1.0000	0.9059
Fujian	0.3760	0.4349	0.3807	0.4042	0.4429	0.3221	0.3935
Jiangxi	1.0000	1.0000	1.0000	0.8119	1.0000	1.0000	0.9687
Shandong	0.4321	0.4646	0.3835	0.4167	0.4530	0.3413	0.4152
Henan	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Hubei	0.5384	0.5171	0.4611	0.4627	0.4162	0.2902	0.4476
Hunan	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Guangdong	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Guangxi	0.5455	0.5925	0.5167	0.5444	0.5826	0.9683	0.6250
Hainan	1.0000	0.7033	0.5975	0.3844	0.7060	1.0000	0.7319
Chongqing	0.3994	0.3957	0.3332	0.3781	0.3746	0.3042	0.3642
Sichuan	0.6680	0.4867	0.3996	0.4204	0.4139	0.3774	0.4610
Guizhou	0.8601	0.8443	1.0000	1.0000	1.0000	1.0000	0.9507
Yunnan	0.2970	0.3023	0.2773	0.2415	0.2316	0.1997	0.2582
Shaanxi	0.6486	0.8461	0.7769	0.7602	0.8122	0.9373	0.7969
Gansu	1.0000	1.0000	0.9592	1.0000	1.0000	1.0000	0.9932
Qinghai	1.0000	1.0000	1.0000	0.8169	0.6533	0.8587	0.8882
Ningxia	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Xinjiang	0.4024	0.3398	0.2706	0.2654	0.2818	0.2162	0.2960
Average	0.6961	0.7238	0.6512	0.6545	0.6842	0.7077	0.6862

. The most obvious difference is that the average values across all years as well as the overall average value of the eco-efficiency showed a sharp decline. The overall average eco-efficiency of all provinces across all examined years was 0.6862, indicating that the environmental pollution of Chinese transportation industry can be reduced by 31.38% on average. That was relatively dramatic but reasonable, because under this scenario the only optimization direction of our model is environmental improvement, so a greater extent of environmental inefficiency can be identified. As for the provinces located in the first tier of this model, there were only seven provinces left since Guizhou and Gansu are no longer environmentally efficient across all years. Additionally, the number of provinces situated in the second tier declined to eight, including Tianjin, Anhui, Jiangxi, Hainan, Guizhou, Shaanxi, Gansu, and Qinghai. Hence, the remaining fifteen provinces are in the third tier, dragging down the overall average eco-efficiency. According to the eco-efficiency results under this scenario, the overall trend and distribution characteristics of the efficiency value are similar to that under the first scenario. For instance, the eco-efficiency of Beijing also showed a rapid decline in 2017. This reflects that models in the two scenarios have a certain degree of consistency in the identification of eco-efficiency, but substantial differences also exist in to the detailed situation.

We further draw a line chart in Figure 3 for the yearly average eco-efficiency to compare the difference in eco-efficiency in the two scenarios. As seen from the graph, the eco-efficiency in the win-win scenario was always higher than that in the environmental priority scenario, and the difference between these two scenarios gradually decreased. Moreover, the eco-efficiency of Chinese transportation industry experienced a slight increase during 2015–2016 and a sharp decline during 2016–2017, and it continued to rise since year 2017. It should be noted that it is reasonable to use the DEA-based eco-efficiency measurement results for comparison among a group of DMUs in the same period. However, due to the problem of intertemporal incomparability for DEA-based efficiency scores, it is unreasonable to conduct an evolutionary analysis on the results of different periods. Therefore, it is necessary to conduct intertemporal comparisons among periods, i.e., the evolutionary change research on the eco-efficiency, using the GMLPI index in the following parts.

There are obvious differences in the economic and geographical characteristics of these 30 provinces in China, but the neighboring provinces share some similarities. Therefore, we divide these provinces into eight economic geographic regions, including the North coast area (Beijing, Tianjin, Hebei, and Shandong), the Middle Yellow River area (Shanxi, Shaanxi, Henan, and Inner Mongolia), the Northeast area (Liaoning, Jilin, and Heilongjiang), East coast area (Shanghai, Jiangsu, and Zhejiang), South coast area (Fujian, Guangdong, and Hainan), Middle Yangtze River area (Anhui, Jiangxi, Hubei, and Hunan), Southwest area (Guangxi, Chongqing, Sichuan, Guizhou, and Yunnan), and Northwest area (Gansu, Qinghai, Ningxia, and Xinjiang). According to the above regional division, we further calculate the average eco-efficiency for each region, as shown in

Table 6. The average eco-efficiency of Chinese eight economic geographic regions.

Regions	Average Eco-Efficiency in the First Scenario	Ranks	Average Eco-Efficiency in the Second Scenario	Ranks
North coast area	0.8135	5	0.6957	5
Middle Yellow River area	0.7378	7	0.6180	7
Northeast area	0.8096	6	0.6281	6
East coast area	0.9256	2	0.7215	3
South coast area	0.8207	4	0.7084	4
Middle Yangtze River area	0.9289	1	0.8305	1
Southwest area	0.6872	8	0.5318	8
Northwest area	0.8570	3	0.7944	2

. Regardless of the scenario, the Middle Yangtze River area performed the best among the eight regions, while the Southwest area ranked last. Moreover, regional rankings remained relatively constant in both scenarios, except for slight fluctuations in the East coast area and the Northwest area.

4.2.2. Analysis of GMLPI Results

Since the measurement results of eco-efficiency are essentially not comparable, we further calculated the results of the GMLPI index and its decompositions to evaluate the evolution of eco-efficiency more accurately. Based on model (4) and the construction of the GMLPI index, we present the detailed results of the two specific scenarios in

Table 7. Results of the GMLPI index and its decompositions of different provinces in the win-win scenario.

Provinces	2015–2016			2016–2017			2017–2018			2018–2019			2019–2020
	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC
Beijing	0.6058	1.0000	0.6058	0.9385	0.5702	1.6460	0.9953	1.0129	0.9826	1.0228	1.0908	0.9377	0.9562	1.0602	0.9020
Tianjin	1.1415	1.2473	0.9152	1.1584	1.0000	1.1584	0.9832	1.0000	0.9832	0.9583	1.0000	0.9583	0.9271	1.0000	0.9271
Hebei	0.9399	1.0000	0.9399	1.0640	1.0000	1.0640	1.0000	1.0000	1.0000	0.9908	1.0000	0.9908	1.0093	1.0000	1.0093
Shanxi	1.0237	1.0772	0.9503	0.9706	0.8774	1.1062	0.9773	0.9710	1.0066	1.0269	1.0854	0.9461	1.0040	0.9384	1.0699
Inner Mongolia	1.0846	1.1378	0.9532	1.0105	0.9914	1.0192	1.0092	1.0341	0.9759	0.9817	0.9792	1.0025	0.9723	0.9233	1.0531
Liaoning	1.0681	1.0000	1.0681	1.0360	1.0000	1.0360	1.1188	1.0000	1.1188	1.0000	1.0000	1.0000	0.7309	1.0000	0.7309
Jilin	1.0051	1.0497	0.9575	1.0617	1.0585	1.0031	1.1330	1.1030	1.0272	0.9438	0.9725	0.9705	0.8217	1.0733	0.7656
Heilongjiang	0.9316	0.8845	1.0532	0.9872	0.7621	1.2953	1.0797	1.1230	0.9614	0.9277	0.9692	0.9572	0.8481	0.9027	0.9395
Shanghai	0.9590	1.0000	0.9590	1.0722	1.0000	1.0722	1.0780	1.0000	1.0780	1.1284	1.0000	1.1284	1.0000	1.0000	1.0000
Jiangsu	1.0012	1.0696	0.9361	1.0221	1.0000	1.0221	1.0270	0.8650	1.1872	1.0031	1.1258	0.8911	0.7859	1.0269	0.7653
Zhejiang	0.9898	0.9615	1.0294	1.0097	0.9426	1.0712	1.0359	1.0152	1.0204	1.0320	1.1028	0.9358	0.9034	1.0037	0.9001
Anhui	0.9571	1.0000	0.9571	0.9225	0.9478	0.9733	1.1326	1.0551	1.0734	0.9046	1.0000	0.9046	0.9220	1.0000	0.9220
Fujian	0.9920	1.0428	0.9513	0.9993	0.9622	1.0386	0.9966	1.0170	0.9800	1.0267	1.0945	0.9380	1.0362	0.9858	1.0511
Jiangxi	1.0019	1.0000	1.0019	1.0349	1.0000	1.0349	0.9353	0.9539	0.9805	0.9964	1.0483	0.9505	0.7874	1.0000	0.7874
Shandong	1.0132	0.9749	1.0393	1.0123	0.9664	1.0475	1.0243	1.0322	0.9924	1.0209	1.0166	1.0043	0.8482	0.9674	0.8767
Henan	1.0219	1.0000	1.0219	1.0089	1.0000	1.0089	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	0.7145	1.0000	0.7145
Hubei	0.9630	0.9713	0.9915	1.0147	0.9906	1.0243	0.9952	0.9815	1.0139	0.9568	1.0203	0.9377	0.8019	0.8837	0.9075
Hunan	0.9923	1.0000	0.9923	1.0078	1.0000	1.0078	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	0.6757	1.0000	0.6757
Guangdong	0.9476	1.0000	0.9476	1.0553	1.0000	1.0553	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	0.9250	1.0000	0.9250
Guangxi	1.0021	1.0049	0.9972	1.0239	0.9841	1.0404	1.0332	1.0411	0.9924	1.0009	1.0145	0.9866	0.8932	1.1637	0.7676
Hainan	0.6871	0.8597	0.7992	1.0314	0.9493	1.0866	0.9448	0.9016	1.0478	1.1914	1.1937	0.9980	1.2536	1.1385	1.1011
Chongqing	0.9629	0.9765	0.9861	0.9823	0.9505	1.0334	1.0253	1.0337	0.9919	0.9884	0.9987	0.9897	0.9113	0.9524	0.9568
Sichuan	0.8700	0.9078	0.9583	0.9641	0.9243	1.0431	1.0058	1.0148	0.9911	0.9861	0.9986	0.9875	0.9075	1.0831	0.8379
Guizhou	0.9692	1.0000	0.9692	1.0812	1.0000	1.0812	1.0336	1.0000	1.0336	1.0000	1.0000	1.0000	0.7649	1.0000	0.7649
Yunnan	0.9847	0.9924	0.9922	0.9989	0.9808	1.0185	0.9666	0.9719	0.9945	0.9887	0.9930	0.9956	0.9324	0.9964	0.9358
Shaanxi	1.0754	1.1298	0.9518	1.0003	0.9540	1.0485	0.9781	0.9906	0.9874	1.0122	1.0311	0.9817	0.8575	1.0802	0.7938
Gansu	0.9833	1.0000	0.9833	0.9874	1.0000	0.9874	1.0481	1.0000	1.0481	1.0000	1.0000	1.0000	0.7843	1.0000	0.7843
Qinghai	0.9608	1.0000	0.9608	0.9353	1.0000	0.9353	0.9816	0.9212	1.0657	0.9168	0.9327	0.9829	0.8685	1.0954	0.7929
Ningxia	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Xinjiang	0.9178	0.9266	0.9905	0.9600	0.9115	1.0532	0.9898	0.9959	0.9939	1.0075	1.0154	0.9922	0.9074	0.9466	0.9586
Average	0.9684	1.0071	0.9620	1.0117	0.9575	1.0671	1.0176	1.0012	1.0176	1.0004	1.0228	0.9789	0.8917	1.0074	0.8872

and

Table 8. Results of the GMLPI index and its decompositions of different provinces in the environmental priority scenario.

Provinces	2015–2016			2016–2017			2017–2018			2018–2019			2019–2020
	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC	GMLPI	EC	BPC
Beijing	0.5444	1.0000	0.5444	0.9859	0.5368	1.8364	0.9952	0.9972	0.9979	1.0034	1.0424	0.9625	0.9753	1.0681	0.9131
Tianjin	1.1684	1.5951	0.7325	1.3681	1.0000	1.3681	0.9575	1.0000	0.9575	0.9189	1.0000	0.9189	0.8704	1.0000	0.8704
Hebei	0.8780	1.0000	0.8780	1.1389	1.0000	1.1389	1.0000	1.0000	1.0000	0.9601	1.0000	0.9601	1.0416	1.0000	1.0416
Shanxi	0.9881	1.1315	0.8733	1.0040	0.8826	1.1376	1.0188	1.0216	0.9973	1.0208	1.0707	0.9534	1.0032	0.9476	1.0586
Inner Mongolia	1.0591	1.1029	0.9602	1.0080	0.9928	1.0153	1.0072	1.0294	0.9784	0.9750	0.9730	1.0021	0.9769	0.9238	1.0575
Liaoning	1.0772	1.0000	1.0772	1.0701	1.0000	1.0701	1.2715	1.0000	1.2715	1.0000	1.0000	1.0000	0.6222	1.0000	0.6222
Jilin	1.0100	1.0323	0.9784	1.0087	1.0150	0.9938	1.1236	1.1843	0.9487	0.9909	0.9767	1.0145	0.8470	1.0989	0.7708
Heilongjiang	0.9649	0.9964	0.9684	1.0035	0.9238	1.0863	1.0346	1.1002	0.9404	0.9817	0.9633	1.0191	0.9043	0.9078	0.9961
Shanghai	0.9539	1.0000	0.9539	1.0634	1.0000	1.0634	1.0853	1.0000	1.0853	1.0481	1.0000	1.0481	1.2248	1.0000	1.2248
Jiangsu	1.0009	0.9962	1.0047	0.9937	0.9617	1.0332	0.9878	1.0226	0.9660	0.9885	0.9887	0.9998	0.8807	1.4688	0.5996
Zhejiang	0.9936	1.0454	0.9504	0.9932	0.9453	1.0507	1.0114	1.0199	0.9917	1.0410	1.0823	0.9618	0.9036	0.9554	0.9457
Anhui	0.8270	1.0000	0.8270	0.9408	0.7787	1.2081	0.9874	1.0030	0.9844	1.0217	1.2803	0.7980	0.8592	1.0000	0.8592
Fujian	0.9933	1.0376	0.9572	0.9994	0.9665	1.0341	0.9970	1.0147	0.9825	0.9894	1.0249	0.9653	0.9884	0.9280	1.0652
Jiangxi	1.0023	1.0000	1.0023	1.0802	1.0000	1.0802	0.8210	0.8417	0.9754	0.9874	1.1881	0.8311	0.8044	1.0000	0.8044
Shandong	0.9975	1.0211	0.9769	0.9866	0.9498	1.0387	1.0143	1.0210	0.9935	1.0106	1.0235	0.9874	0.9257	0.9327	0.9925
Henan	1.0669	1.0000	1.0669	1.0347	1.0000	1.0347	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	0.6185	1.0000	0.6185
Hubei	0.9444	0.9857	0.9581	1.0072	0.9635	1.0453	0.9919	1.0010	0.9909	0.9605	0.9707	0.9896	0.8973	0.9263	0.9687
Hunan	0.9371	1.0000	0.9371	0.9912	1.0000	0.9912	1.0766	1.0000	1.0766	0.7809	1.0000	0.7809	0.7676	1.0000	0.7676
Guangdong	1.1126	1.0000	1.1126	1.2240	1.0000	1.2240	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	0.8328	1.0000	0.8328
Guangxi	0.9918	1.0334	0.9598	0.9879	0.9489	1.0411	1.0084	1.0191	0.9895	1.0133	1.0269	0.9868	0.9468	1.3740	0.6891
Hainan	0.5827	0.7712	0.7557	1.0026	0.9246	1.0844	1.0066	0.8681	1.1595	1.0957	1.2486	0.8775	1.5520	1.2940	1.1994
Chongqing	0.9734	0.9977	0.9756	0.9857	0.9625	1.0241	1.0206	1.0277	0.9931	0.9904	0.9978	0.9925	0.9324	0.9585	0.9727
Sichuan	0.8574	0.8802	0.9741	0.9697	0.9456	1.0254	1.0052	1.0131	0.9922	0.9878	0.9959	0.9918	0.9270	0.9775	0.9483
Guizhou	0.9524	0.9864	0.9656	1.1709	1.1557	1.0131	1.0876	1.0000	1.0876	1.0000	1.0000	1.0000	0.6342	1.0000	0.6342
Yunnan	0.9885	1.0031	0.9855	0.9993	0.9855	1.0140	0.9758	0.9796	0.9961	0.9921	0.9944	0.9976	0.9560	0.9823	0.9733
Shaanxi	1.0999	1.1711	0.9391	1.0017	0.9434	1.0618	0.9693	0.9865	0.9825	1.0170	1.0438	0.9744	0.8326	1.1177	0.7449
Gansu	0.9526	1.0000	0.9526	0.9755	0.9608	1.0153	1.1352	1.0408	1.0907	1.0000	1.0000	1.0000	0.6986	1.0000	0.6986
Qinghai	0.9182	1.0000	0.9182	0.8765	1.0000	0.8765	0.9740	0.8453	1.1523	0.8633	0.8785	0.9827	0.8680	1.1799	0.7356
Ningxia	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000	1.0000
Xinjiang	0.9644	0.9623	1.0021	0.9792	0.9600	1.0201	0.9925	0.9970	0.9956	1.0054	1.0096	0.9959	0.9367	0.9632	0.9725
Average	0.9600	1.0250	0.9396	1.0284	0.9568	1.0875	1.0185	1.0011	1.0192	0.9881	1.0260	0.9664	0.9076	1.0335	0.8859

, respectively. According to our observation, no province had GMLPI results consistently higher than unity over six years in both scenarios. Nevertheless, the GMLPI index in Yunnan and Qinghai remained below unity across six years regardless of the scenario. This indicates that productivity in these two provinces kept declining. Interestingly, results of the GMLPI index and its two decompositions of Ningxia were always at a constant value of unity during the whole examined periods in the two scenarios. Moreover, the eco-efficiency results of every year in every scenario were unity as well, which illustrates that the transportation industry in Ningxia was the national benchmark in terms of environmental protection. Furthermore, we also calculated the results of average GMLPI index and its decompositions, as listed in

Table 9. Results of average GMLPI index and its decompositions.

Scenarios	Indices	2015–2016	2016–2017	2017–2018	2018–2019	2019–2020
The win-win orientation scenario	GMLPI	0.9684	1.0117	1.0176	1.0004	0.8917
	EC	1.0071	0.9575	1.0012	1.0228	1.0074
	BPC	0.9620	1.0671	1.0176	0.9789	0.8872
The win-win orientation scenario	GMLPI	0.9600	1.0284	1.0185	0.9881	0.9076
	EC	1.0250	0.9568	1.0011	1.0260	1.0335
	BPC	0.9396	1.0875	1.0192	0.9664	0.8859

. In both scenarios, the GMLPI index showed an earlier increase and later decrease trend, which was consistent with reality. On one hand, a significant and steady increase existed in freight turnover volume and passenger turnover volume independently, corresponding to the earlier increase in the GMLPI. On the other hand, in subsequent years, massive infrastructure did not lead to rapid growth in different outputs along with the COVID-19 pandemic, causing a significant decrease in the GMLPI. From the perspective of the decomposition indices, EC showed a slight trend of earlier decrease and later increase, while BPC basically kept consistent with the trend of GMLPI, indicating that BPC was the main driven force affecting the trend of GMLPI.

4.3. A Summary of the Main Findings

Based on the empirical study of the transportation industry in 30 provinces of China, we obtained the following findings: (1) Data of most input and output variables remained stable, while the passenger turnover volume experienced a sharp decline in year 2020, which was mainly affected by the outbreak of the COVID-19. (2) Amongst the 30 provinces, the eco-efficiency in Beijing was the most unstable, which experienced a rapid decline from 2016 to 2017 and remained at a low level until 2020, resulting from the rapid expansion of fixed assets investment and the modest changes in transportation capacity. (3) Our eco-efficiency measurement models in two scenarios have a certain degree of consistency in the results, and a greater extent of environmental inefficiency can be identified in the environmental priority orientation scenario, illustrating the practicability of this model in evaluating the resource transformation efficiency and the potential of emission reduction. (4) The eco-efficiency of Chinese transportation industry experienced a slight increase during 2015–2016 and a sharp decline during 2016–2017, and it continued to rise since year 2017. (5) The Middle Yangtze River area was the best performer among the eight regions in terms of the eco-efficiency, while the Southwest area was always the last. (6) No province maintained a higher than unity GMLPI results over six years, but the GMLPI index and its two decompositions of Ningxia kept at unity during the whole examined period, illustrating its outstanding performance in the environmental protection. (7) The GMLPI index solves the intertemporal incomparability problem of the eco-efficiency evaluation model based on DEA, and the results showed an earlier increase and later decrease trend, and BPC acted as the main driven force of GMLPI.

5. Conclusions and Discussions

As a basic industry affecting the economic development, the transportation industry should improve its eco-efficiency in rapid construction and expansion. To reasonably estimate the eco-efficiency and identify its evolutionary characteristics, this paper constructs and transforms the DDF in the framework of DEA, in which we also incorporate the influence of different policy scenarios through direction vectors. An empirical study is conducted on the transportation industry of 30 provinces in China, and a group of main findings about the eco-efficiency and its evolutionary GMLPI index were obtained.

Consequently, different provinces can formulate their transportation capacity improvement and environmental protection regulations according to their and the benchmarking provinces’ eco-efficiency. First, provinces with eco-efficiency scores less than unity, especially those with eco-efficiency below the average level for a long time, should formulate strict policies or regulations for energy conservation and emission reduction in the transportation industry, to facilitate green transformation through administrative intervention. Second, since BPC is the most influential factor affecting the GMLPI, departments of environmental protection shall support and encourage research institutions in the field of transportation to carry out research activities on improving the environmental protection technology, from which benchmarking provinces will benefit a lot and can set examples and guide the development of the national transportation industry toward a cleaner and more efficient direction. Third, from the regional perspective, regions with poor eco-efficiency can optimize their energy structure and improve their eco-efficiency through environmental protection subsidies and green infrastructure construction.

Even though several clear phenomena and results are presented in this paper, we admit that the model proposed in this paper is only applicable to two limited direction vectors because of the definition of environmental efficiency. Hence, future research can further refine the model of policy scenarios according to the actual research questions and policy background. Moreover, the regulatory costs, governance effects, and economic impacts brought by different environmental regulations can be investigated in possible future studies.