Total-Factor Eco-Efficiency and Its Influencing Factors in the Yangtze River Delta Urban Agglomeration, China

Urban agglomerations are not only the core areas leading economic growth but also the fronts facing severe resource and environmental challenges. This paper aimed to increase our understanding of urban eco-efficiency and its influencing factors and thus provide the scientific basis for green development. We developed a model that incorporates super-efficiency, slacks-based-measure, and global-frontier technology to calculate the total-factor eco-efficiency (TFEE) and used a spatial panel Tobit model to take into account spatial spillover effects. An empirical study was conducted utilizing a prefecture-level dataset in the Yangtze River Delta Urban Agglomeration (YRDUA) from 2003 to 2016. The main findings reveal that significant spatial differences exist in TFEE in the YRDUA: high-TFEE cities were majorly located in the coastal areas, while low-TFEE cities were mostly situated inland. Overall, TFEE shows a trend of “decline first and then rise with fluctuation”; the disparity between inland and coastal regions has expanded. Further regression analysis suggests that industrial structure, environmental regulation, and innovation were positively related to TFEE, while foreign direct investment was not conducive to the growth in TFEE. The relationship between population intensity and urban eco-efficiency is an inverted U-shaped curve. Finally, several specific policy implications were raised based on the results.


Introduction
As environmental pollution and resource crises intensified, policymakers are increasingly concerned about the impact of economic activities on the ecological environment. Urban agglomerations are not only the core areas leading economic growth but also the fronts facing severe resource and environmental challenges. The spatial differentiation of economic development and of ecological environment exist simultaneously within urban agglomerations. To promote green and coordinated development, it is urgent to improve the performance appraisal system of regional development and to optimize the interregional cooperation mechanism. Eco-efficiency, which takes into account factors such as resources, environment, and economy, can reflect green development performance comprehensively, and therefore, providing an important reference for measuring the long-term development advantages of a region and formulating sustainable development policies. Undoubtedly, the calculation of urban mechanisms of the urban cluster on urban eco-efficiency. They find that the improvement of the urban cluster is conducive to enhancing urban eco-efficiency, and there is a "core-periphery" spatial structure in the process of urban cluster development. Li et al. [25] studied the relationship between government transparency and eco-efficiency utilizing the data of 262 cities in China from 2005 to 2012. Their results suggested that the overall eco-efficiency of Chinese cities was low, and a nonlinear relationship exists between government transparency and eco-efficiency performance.
However, despite all the fruitful results and substantial advances, there are still some limitations in the previous literature. Firstly, compared with numerous studies that focused on the provincial and sectoral level [21], much less attention has been paid to the eco-efficiency of cities, especially that of the emerging cities in developing countries [26]. Considering that urban areas make a tremendous contribution to resource consumption and pollution emissions in the developing world, it is of great significance to carry out in-depth research on their green development. Another essential drawback is that most studies ignored the spatial correlations between cities when discussing the influencing factors of urban eco-efficiency. However, apparent neglect of the spatial spillover effects, which tend to become significant with the intensification of interregional economic connections and the more frequent flow of factors of production, could result in biased estimations [1,27].
To address these limitations of extant studies, this paper extends research at the provincial and sectoral level to the urban level utilizing a prefecture-level panel dataset in the Yangtze River Delta Urban Agglomeration (YRDUA), China, during the period of 2003-2016, and constructs a model that incorporated the super-efficiency DEA model with slacks-based-measure and global-frontier technology (SSBM-GF) to estimate total-factor eco-efficiency (TFEE). Moreover, to take into account spatial spillover effects, a spatial panel Tobit model is constructed to analyze the influencing factors of urban eco-efficiency. The results are supportive of understanding the spatial difference and driving mechanism of urban eco-efficiency in the YRDUA, providing a scientific basis for governments to formulate policies to promote the development of green and sustainable urbanization.
The remainder of the paper is organized as follows. Section 2 introduces the methodology and data sources. Section 3 presents the empirical results. Section 4 provides some discussion and implications. Section 5 provides a conclusion.

Super-Efficiency Slacks-Based Measure Global Frontier Model
Following Fare et al. [28], the non-parametric Data Envelopment Analysis (DEA) piecewise linear production frontiers are adopted in this study to estimate total-factor eco-efficiency. To overcome the "discriminating power problem" [11,12], Tone [13,14] proposed the super-efficiency model based on the non-radial, non-angular Slacks-based Measure (SBM) model. Furthermore, Pastor and Lovell [15] introduced a global benchmark technology to address the "technical regress" problem [11,12]. Thus, to ensure that the calculations were accurate, we combined the super-efficiency SBM model with global frontier technology to construct a Super-efficiency Slacks-based Measure Global Frontier (SSBM-GF) model.
Given that x represents each of the M inputs of a decision-making unit (DMU, that is, the cities in this study) such that x = (x 1 · · · x M ) ∈ R + M , y represents each of the S desirable outputs of the DMU such that y = y 1 · · · y S ∈ R + S , b represents each of the V undesirable outputs of the DMU such is the vector for the inputs, desirable outputs, and undesirable outputs of DMU i (i = 1 ... N), and s x , s y , s b is the slacks vector, then a production possibility set (PPS) can be expressed by: where λ is an unknown weight vector.
Thus, the PPS of the global frontier is given by: where PPS Glb denotes the specific technologies of the global frontier (i.e., best practice frontier) [29]. The production technology is assumed to follow all the standard axioms of production theory, including the assumptions of bounded set, bounded convexity, etc. [30,31].
To calculate the super-efficiency of a specific DMU, PPS is constructed by eliminating the observations of that specific DMU. The non-radial distance function is used to measure the efficiency gap between the production frontier and a certain specific DMU. Hence, the TFEE, which measures the distance of observed DMU 0 from the global frontier, can be calculated by solving the linear programming (LP) problem given below: where s x , s ,b , and s y denote the slack in input, desirable output, and undesirable output, respectively.

Spatial Panel Tobit Model
Since TFEE, the dependent variable in our econometric model, always has a value that is no less than 0, it is not suitable to use the ordinary least squares (OLS) for coefficient estimation. Otherwise, it would lead to a problem in terms of the inconsistency of coefficient and biased estimates. On that account, the Tobit model [32], which has been widely used to investigate the influencing factors of environmental efficiency [33], was adopted in this paper.
Additionally, economic activities in one city generally exert a spillover effect on neighboring cities [34]. To take into account the spatial spillover effects of TFEE, a spatial panel Tobit model was constructed in this study. The spatial lag model and the spatial error model are widely used in spatial econometric modeling. The spatial lag Tobit model can be expressed by: where TFEE it is the total-factor eco-efficiency, α is the constant term; ρ is the spatial lag parameter, W is the spatial weights matrix, β is the coefficient of the explanatory variables, X it stands for the explanatory variables, and ε it is the error term.
The spatial error Tobit model can be expressed by: where θ is the spatial autocorrelation coefficient. As for how to select between the above two models, Anselin et al. [35] suggested the use of the Lagrange multiplier (LM-lag and LM-error) and its robust form (Robust LM-lag and Robust LM-error). The spatial weights matrix W was constructed based on geographical distances due to the consideration that, compared with the spatial adjacency matrix, the distance weights matrix is more frequently employed, as it can reflect the attenuation characteristics of spatial spillover. The matrix element w ij is defined as: where d ij is the distance between cities i and j.

Variables Selection
The calculation of TFEE integrated three dimensions: the economy, resources, and the environment: (1) Inputs. The inputs include capital, labor, energy, and water consumption. Capital input (K): the capital stock is estimated by adopting the Perpetual Inventory Method [36,37]. Labor force (L): the average number of employees each year. Energy (E): following extant literature, we used the amount of electricity consumption as a proxy for energy input due to the lack of data on final energy consumption at the city level [38]. Water (W): the amount of water consumption. (2) Undesirable outputs. The undesirable outputs consist of three types of pollutants, namely, Wastewater (WW), sulfur dioxide (SO2), and soot and dust (SD). (3) Desirable output. Gross domestic product (GDP) is defined as the desirable output of each city.
To investigate the determinants of total-factor eco-efficiency, we specify an econometric model. Based on previous research in this field, the following factors were included as the explanatory variables in this study: (1) Industrial structure (IS). Urban eco-efficiency can be affected by the industrial structure. Compared with the secondary industry, the tertiary sector is much less relying on resources and creates fewer pollutants; therefore, a higher ratio of the service sector may lead to lower pressure on the urban environment, which can, in turn, lead to a better eco-efficiency. Thus, we use the tertiary industry ratio as an indicator to represent the industrial structure. (2) Environmental regulations (ER). Environmental regulations are critical for resource-saving and pollution control, as well as the promotion of green development [39]. Compared with the operating costs of pollution control facilities, the income level is a better proxy for environmental regulation. The main reason is that using the former indicator may encounter significant endogenous problems in the econometric analysis since there is an apparent two-way causality; that is, not only operating costs of pollution control facilities can affect environmental quality, but also the level of pollution will determine the expenditure. By contrast, there is no such problem with the latter one. As the income level rises, the citizen's demand for a better environment will also increase, thus urging the government to improve environmental governance properly. Following Antweiler et al. [40], we used GDP per capita as a proxy for environmental regulation. (3) Innovation (INN). Innovation plays a crucial role in improving environmental performance [41].
Above all, innovation can promote the improvement of production technology, and therefore, can reduce the input of raw materials and energy consumption of unit products. In addition, innovation can also give impetus to the emerging industries and become a new driving force for the economy [42]. The ratio of scientific expenditure to total fiscal spending is used to define a city's innovation intensity. (4) Foreign direct investment (FDI). The relationship between FDI and green development is uncertain.
The advanced technologies that come with FDI could help to promote economic development, but resource consumption and pollution emission might also increase as a result of FDI and thus have a negative impact on eco-efficiency [43][44][45][46][47][48][49]. We use the ratio of real FDI to real GDP to measure FDI inflows. (5) Population density (PD). There is also an undetermined relationship exists between population density (the ratio of population to the built-up area) and urban eco-efficiency. Cropper and Griffiths [50] pointed out that higher population density may lead to higher pressure on the environment, which can, in turn, lead to a decrease in TFEE. However, Liu et al. [51] suggested that higher population density may urge society to pay more attention to the environment, and it may improve TFEE. Therefore, Population density and its squared term (PD2) are also included in our model to test whether there was an environmental Kuznets curve (EKC) for TFEE and population density.

Study Area and Data Sources
The Yangtze River Delta Urban Agglomeration is one of the six largest metropolitan areas in the world and the largest metropolitan area in China [52]. The YRDUA mainly consists of Shanghai City, Jiangsu Province, Zhejiang Province, and Anhui Province. This study included 26 prefecture-level cities of the YRDUA, among which nine cities are located in Jiangsu Province, eight cities are located in Zhejiang Province, and the other eight cities are located in Anhui Province (see Figure 1).

Study Area and Data Sources
The Yangtze River Delta Urban Agglomeration is one of the six largest metropolitan areas in the world and the largest metropolitan area in China [52]. The YRDUA mainly consists of Shanghai City, Jiangsu Province, Zhejiang Province, and Anhui Province. This study included 26 prefecture-level cities of the YRDUA, among which nine cities are located in Jiangsu Province, eight cities are located in Zhejiang Province, and the other eight cities are located in Anhui Province (see Figure 1). The study period covered the years 2003-2016, and the data were collected from the Chinese City Statistical Yearbook [53], the China Statistical Yearbook on Environment [54], and the Annual Statistical Report on Environment in China [55]. All monetary variables are adjusted to 2000 constant prices using the corresponding price indices. Table 1 lists the descriptive statistics for the relevant variables.

Analysis of TFEE
The TFEE of cities in the YRDUA during 2003-2016 were evaluated by solving Equation (3), and the results are analyzed in two dimensions: spatial distribution and temporal evolution.
3.1.1. Spatial Distribution of TFEE As shown in Figure 2, significant spatial differences existed in TFEE in the YRDUA. The cities with high eco-efficiency were majorly located in the coastal areas, while the cities with low eco-efficiency were mostly situated in the inland parts throughout the study period. Moreover, the number of cities with high eco-efficiency shrank during the period 2003-2012, and since then has increased significantly.

Analysis of TFEE
The TFEE of cities in the YRDUA during 2003-2016 were evaluated by solving Equation (3), and the results are analyzed in two dimensions: spatial distribution and temporal evolution.

Influencing Factors of Total-Factor Eco-Efficiency
Stata 14.0 software was used to estimate the spatial econometric models in this study, and variables measuring by non-percentage indicators were transformed to logarithms in the model to reduce the degree of dispersion. In order to explore the differences between areas and to test the robustness of the estimation results, we divided the 26 cities into high-income group, which mainly included the coastal cities, and the low-income group, which primarily included the inland cities, according to GDP per capita in the year of 2003. Then, we conducted an econometric analysis using the full sample and the two subsamples, respectively. Moreover, before estimating the coefficients of all samples, we firstly tested the spatial correlation based on the residuals of the corresponding nonspatial OLS estimations. The test results in Table 2 provide two essential information: (1) the values of Moran's I are greater than zero and significant at a 5% level in all samples, indicating that the TFEE of the cities has a positive spatial autocorrelation; (2) the strong statistical significance of the LM-lag tests suggest that the spatial lag models are suitable in all samples, while some of the LM-err statistics are insignificant. Therefore, the spatial lag Tobit model was selected rather than the spatial error Tobit model. Table 3 reports the estimated results of the spatial lag Tobit model assuming both fixed and random effects.

Influencing Factors of Total-Factor Eco-Efficiency
Stata 14.0 software was used to estimate the spatial econometric models in this study, and variables measuring by non-percentage indicators were transformed to logarithms in the model to reduce the degree of dispersion. In order to explore the differences between areas and to test the robustness of the estimation results, we divided the 26 cities into high-income group, which mainly included the coastal cities, and the low-income group, which primarily included the inland cities, according to GDP per capita in the year of 2003. Then, we conducted an econometric analysis using the full sample and the two subsamples, respectively. Moreover, before estimating the coefficients of all samples, we firstly tested the spatial correlation based on the residuals of the corresponding non-spatial OLS estimations. The test results in Table 2 provide two essential information: (1) the values of Moran's I are greater than zero and significant at a 5% level in all samples, indicating that the TFEE of the cities has a positive spatial autocorrelation; (2) the strong statistical significance of the LM-lag tests suggest that the spatial lag models are suitable in all samples, while some of the LM-err statistics are insignificant. Therefore, the spatial lag Tobit model was selected rather than the spatial error Tobit model. Table 3 reports the estimated results of the spatial lag Tobit model assuming both fixed and random effects.  Table 3. Estimated determinants of total-factor eco-efficiency. The coefficients for spatial autoregressive terms, W*TFEE, are positive and significant, indicating a clear spatial spillover effect of eco-efficiency across cities in the YRDUA. A comparison between the estimation results for the high-income group and low-income group suggests that the spatial spillover effect of eco-efficiency was more evident in the low-income cities than in the high-income cities. This difference could be explained by the following reasons to a certain extent: low-income cities tend to have low eco-efficiencies fundamentally due to their production inefficiency; however, quite a few high-income cities were considerably eco-inefficient because of their environmental inefficiency.

Variables
The coefficients for industrial structure, IS, are positive and significant at the 5% level, suggesting that an increase in the proportion of tertiary industry could promote the improvement of eco-efficiency in the YRDUA. This result is consistent with the theory and previous studies. Changes in the industrial structure will result in changes in intensities of energy consumption and pollution emission, and will consequently have great effects on the environment. It is asserted that the difference in industry structure may inevitably affect the level of eco-efficiency [46]. In addition, the effects of industrial structural change on urban eco-efficiency varied by income level. The effect of variation of industrial structure in low-income cities was greater than that in the high-income cities.
The coefficients for environmental regulation, ER, are significantly positive, implying that the impact of environmental regulation on eco-efficiency was quite positive in the YRDUA. This outcome is similar to Wang et al. [42]. For the cities of the YRDUA, strict environmental regulations can promote both economic prosperity and environmental quality and therefore lead to a win-win situation. Hence, the Porter hypothesis [27,28] is supported by this study. Moreover, the effects of environmental regulation on urban eco-efficiency also varied by income level. The effect of environmental regulation in the high-income cities was greater than that in low-income cities. Therefore, the government should take measures to upgrade environmental supervision and increase environmental investment in backward cities.
The coefficients for innovation intensity, INN, are positive and significant at the 1% level, suggesting that the enhancement of innovation intensities has a positive effect on eco-efficiency. Innovation is an essential driving force for sustained economic growth and the key to maintaining the core competitiveness of the industry [47]. Technological innovation can promote the development of green technologies such as energy-saving and emission cutting [36], and therefore, improve environmental quality by reducing resource consumption and pollution emission.
The coefficients for FDI are negative but not significant in the YRDUA, suggesting that FDI has not promoted the growth in eco-efficiency in this region. This result is similar to Wen [29], who suggested that the impacts of FDI on total productivity differed by region in China. The reason may be that foreign investment mainly focused on manufacturing, consequently bringing about tremendous pressure on the resources and environment in the YRDUA. Although "pollution transfer" may occur during the inflow of FDI, the advanced technology and management experience coming with the transfer process was also conducive to improving the environmental quality of the host country [52].
The coefficients for population intensity, PD, are positive and significant in all samples, while their squared terms, PD2, illustrating a significant (10% level) negative sign only in the high-income cities. This implies that the relationship between population intensity and eco-efficiency is an inverted U-shaped curve. That is, below some critical value, the increase in urban population intensity can promote eco-efficiency; otherwise, it may harm the sustainability of urban areas.

Discussion and Implications
This study shows that there were significant regional disparities of TFEE in the YRDUA. The cities with high eco-efficiency were majorly located in the coastal areas, while the cities with low eco-efficiency were mostly situated in the inland parts throughout the study period. This finding is similar to Xing et al. [56], which suggests that these spatial disparities are primarily attributable to geographical and economic differences among areas. Compared with the cities in the inland areas, the cities in the coastal areas are more developed. They have better infrastructure, more advanced technologies and more stringent environmental regulation, which contribute to improving resource utilization and reducing pollutant emissions and thus promoting urban eco-efficiency. In the post-financial crisis era, China gradually strengthened ecological protection and environmental governance and proposed an "ecological civilization construction" strategy. As the most developed region in China, the Yangtze River Delta has taken the lead in ecological construction by accelerating the elimination of low-end industries and encouraging innovation, which has significantly reduced pollution and improved resource utilization efficiency. Accordingly, a steady increase of TFEE in the YRDUA has been observed during the period 2012-2016.
Therefore, some policy implications for achieving green and sustainable development in urban areas can be put forward based on this research. Firstly, the pace of integrated development of the Yangtze River Delta Urban Agglomeration should be accelerated. There was an apparent disparity of ecological efficiency among cities in the Yangtze River Delta, and the gap between inland and coastal cities has been expanding, which is not conducive to the construction of regional ecological civilization. To achieve coordinated and shared development, it is urgent to strengthen guidance and support for the backward areas and give full play to positive spillover effects. While strengthening the responsibility for ecological protection, it is also indispensable to promoting the free flow of factors of production and encouraging the dispersion of advanced technologies and industries to backward areas, thereby gradually eliminate regional inequality.
Secondly, the improvement of ecological efficiency should be set as one of the core factors in the performance assessment system for local government. In a decade (2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012), the ecological efficiency of the Yangtze River Delta urban agglomeration has been stagnated. The fundamental reason is that the old development model places too much emphasis on GDP, thus neglecting resources, environment, and ecology. In recent years, especially after 2012, the reform of the performance assessment system for local government has had a profound impact on the resources and environment. The central government, whose incentives play a vital role in shaping regional plans [57], has firmly strengthened the supervision of the ecological environment, placing more focus on the preservation of natural resources and the improvement of environmental quality. Accordingly, the urban eco-efficiency of the Yangtze River Delta region has achieved steady growth since 2012. However, Chinese cities still have a lot of room to improve their eco-efficiency, which depends to a great extent on further reforming the performance evaluation mechanisms to improve green development.
Thirdly, a timely enhancement of environmental regulations is critical for developing economies to achieve high-quality development. As industrialization goes to a certain stage, developing economies should raise their environmental standards promptly, and put a limit on the energy-intensive and high-pollution industries, and therefore avoid becoming the "pollution heaven" for FDI. For the Yangtze River Delta Urban Agglomeration, more stringent environmental regulations are needed in the future to formulate and implement regional integrated environmental policies.
Furthermore, steady investing in innovation is indispensable to promoting eco-efficiency. Innovation is vital to the development of emerging industries and is the fundamental driving force for achieving sustainable development. Therefore, the government must continuously invest financial resources to support scientific research, technology development and business innovation in multiple dimensions, and create an efficient innovation system. Furthermore, the increase in the proportion of the tertiary industry can reduce resource consumption and pollution emissions, thereby improving ecological efficiency. Hence, the government should emphasize the transformation and upgrading of industries in the backward cities and encourage the development of service industries.
Finally, population concentration should be well-guided to make full use of scale effects and agglomeration effects. There was a weak inverted U-type relationship between population density and eco-efficiency in the YRDUA; thus, differentiated urbanization policies should be formulated for cities of different sizes. For small and medium-sized cities, population concentration should be further enhanced to make full use of scale effects and agglomeration effects to improve economic efficiency and environmental efficiency. However, for some big cities such as Shanghai, Hangzhou, and Nanjing, properly control of the population density is essential to prevent "big city diseases" from threatening sustainable development.

Conclusions
It is of great significance to carry out in-depth research on the dynamics and driving mechanisms of eco-efficiency of urban areas in developing countries since they make a tremendous contribution to resource consumption and pollution emissions in the developing world. This paper extends research at the provincial and sectoral level to the urban level utilizing a prefecture-level panel dataset in the Yangtze River Delta Urban Agglomeration, China, between 2003 and 2016, and proposes an SSBM-GF model that incorporated the super-efficiency DEA model with slack-based-measure as well as global-frontier technology to estimate total-factor eco-efficiency. Moreover, to take into account spatial spillover effects, a spatial lag Tobit model is constructed to analyze the factors influencing urban eco-efficiency.
Our measure revealed that there were great regional disparities of TFEE in the YRDUA; the cities with high eco-efficiency were majorly located in the coastal areas, while the cities with low eco-efficiency were mostly situated in the inland areas throughout the study period. TFEE in the YRDUA and its four regions demonstrated a similar trend of "decline first and then rise with fluctuation" during the period 2003-2016. The regression analysis shows that there was a noticeable positive spatial spillover effect of eco-efficiency across cities in the YRDUA. An increase in the proportion of tertiary industry could promote the improvement of eco-efficiency. Moreover, the impacts of environmental regulation and innovation on eco-efficiency both were significantly positive in the YRDUA. Notably, the inflow of FDI was not conducive to the growth in eco-efficiency in this region. The relationship between population intensity and eco-efficiency is an inverted U-shaped curve. The results are supportive of understanding the spatial difference and driving mechanism of urban eco-efficiency in the YRDUA, providing a scientific basis for governments to formulate policies to promote the development of green and sustainable urbanization.
This study inevitably has some limitations, which in turn point to directions for future research. Firstly, this paper used an SSBM-GF model based on the DEA method to estimate the total-factor eco-efficiency. Although DEA has some advantages, it is a non-parametric mathematical programming approach that does not consider statistical noise, which might lead to biased measures to a certain extent. Moreover, our analysis merely focused on the Yangtze River Delta region; therefore, the findings should not be taken as an accurate depiction of the overall picture of the urban development across China. Nevertheless, this approach can be extended to more parts of China and other countries without difficulty.