Peer effect, political competition, and eco-efficiency: Evidence from city-level data in the People's Republic of China

Chen, Xudong; Huang, Bihong; Yu, Yantuan Working Paper Peer effect, political competition, and eco-efficiency: Evidence from city-level data in the People's Republic of China ADBI Working Paper Series, No. 1125 Provided in Cooperation with: Asian Development Bank Institute (ADBI), Tokyo Suggested Citation: Chen, Xudong; Huang, Bihong; Yu, Yantuan (2020) : Peer effect, political competition, and eco-efficiency: Evidence from city-level data in the People's Republic of China, ADBI Working Paper Series, No. 1125, Asian Development Bank Institute (ADBI), Tokyo This Version is available at: https://hdl.handle.net/10419/238482 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/3.0/igo/ ADBI Working Paper Series PEER EFFECT, POLITICAL COMPETITION, AND ECO-EFFICIENCY: EVIDENCE FROM CITY-LEVEL DATA IN THE PEOPLE’S REPUBLIC OF CHINA Xudong Chen, Bihong Huang, and Yantuan Yu No. 1125 April 2020 Asian Development Bank Institute The Working Paper series is a continuation of the formerly named Discussion Paper series; the numbering of the papers continued without interruption or change. ADBI’s working papers reflect initial ideas on a topic and are posted online for discussion. Some working papers may develop into other forms of publication. The Asian Development Bank refers to “China” as the People’s Republic of China. Suggested citation: Chen, X., B. Huang, and Y. Yu. 2020. Peer Effect, Political Competition, and Eco-Efficiency: Evidence from City-Level Data in the People’s Republic of China. ADBI Working Paper 1125. Tokyo: Asian Development Bank Institute. Available: https://www.adb.org/publications/peereffect-political-competition-eco-efficiency-prc Please contact the authors for information about this paper. Email: [email protected], x[email protected], [email protected] Xudong Chen is a professor of economics and finance in the School of Business of Baldwin Wallace University in Berea, OH, United States. Bihong Huang is a research fellow at the Asian Development Bank Institute in Tokyo, Japan. Yantuan Yu is an assistant research fellow at the College of Economics of Jinan University in Guangzhou, People’s Republic of China. The views expressed in this paper are the views of the author and do not necessarily reflect the views or policies of ADBI, ADB, its Board of Directors, or the governments they represent. ADBI does not guarantee the accuracy of the data included in this paper and accepts no responsibility for any consequences of their use. Terminology used may not necessarily be consistent with ADB official terms. Working papers are subject to formal revision and correction before they are finalized and considered published. Asian Development Bank Institute Kasumigaseki Building, 8th Floor 3-2-5 Kasumigaseki, Chiyoda-ku Tokyo 100-6008, Japan Tel: +81-3-3593-5500 Fax: +81-3-3593-5571 URL: www.adbi.org E-mail: [email protected] © 2020 Asian Development Bank Institute ADBI Working Paper 1125 Chen, Huang, and Yu Abstract This study examines the impacts of political competition on eco-efficiency. We first develop a theoretical model in which local government officials compete against each other to maximize their own political score. We find that after an initial stage of decline, eco-efficiency eventually turns upwards, once environmental performance becomes a meaningful component of local government officials’ annual assessment. Eco-efficiency also exhibits a pattern of convergence. Lastly, level of political competition is found to be negatively correlated with eco-efficiency. For the empirical analysis, we use a data envelopment analysis (DEA) model to compute the eco-efficiency level for 191 Chinese cities from 2003 to 2015. Our empirical evidence presents a U-shape pattern in the trend of eco-efficiency, and helps us identify two peer effects that work in opposite directions for cities and regions: the incentivizing effect arising from higher performing neighbors, and the disincentivizing effect when a city outperforms its competitors. Both peer effects lead to convergence in eco-efficiency, and our spatial econometric modeling analysis suggests that the net peer effect is significantly positive. We also find evidence of political competition reducing eco-efficiency, as predicted in the theoretical model. Our findings are robust to alternative measures of eco-efficiency. Keywords: peer effect, political competition, eco-efficiency, spatial analysis, People’s Republic of China JEL Classification: C61, C67, Q56, R15 ADBI Working Paper 1125 Chen, Huang, and Yu Contents 1. INTRODUCTION ......................................................................................................... 1 2. THE THEORETICAL MODEL ..................................................................................... 4 3. METHODOLOGY AND DATA ..................................................................................... 9 3.1 The DEA Model ............................................................................................... 9 3.2 Econometric Specification ............................................................................. 11 3.3 Data ............................................................................................................... 13 4. RESULTS AND DISCUSSION .................................................................................. 14 4.1 Eco-efficiency: The National Trend ............................................................... 14 4.2 Eco-efficiency: Heterogeneities across Groups ............................................. 15 4.3 Baseline Results ............................................................................................ 17 4.4 Effects of Political Competition: Sub-sample Estimations ............................. 18 4.5 Robustness Check ........................................................................................ 21 5. CONCLUSIONS ........................................................................................................ 22 REFERENCES ..................................................................................................................... 24 APPENDIX A ........................................................................................................................ 29 APPENDIX B ........................................................................................................................ 31 APPENDIX C: ROBUSTNESS CHECKS: ALTERNATIVE MEASURE OF ECO-EFFICIENCY .......................................................................................................... 32 APPENDIX D: ESTIMATION RESULTS OF DIFFERENT CITIES: PERIOD 2003–2015 .... 37 APPENDIX E: ESTIMATION RESULTS OF DIFFERENT CITIES: PERIOD 2003–2010 .... 39 APPENDIX F: ESTIMATION RESULTS OF DIFFERENT CITIES: PERIOD 2011–2015 ..... 41 ADBI Working Paper 1125 Chen, Huang, and Yu 1 1. INTRODUCTION The relationship between economic growth and the environment remains controversial, and has received increasing attention due to the escalated level of environmental risk today.1 Over time, it becomes evident that economic growth tends to occur conjointly with environmental degradation. Numerous academicians and policy makers have contemplated the issue of how to engineer a transition away from unsustainable growth to eventual sustainable development. Eco-efficiency,2 the concept of producing more goods and services while using fewer resources and creating less waste and pollution, has been proposed as an indicator by which to monitor such a transition (Schmidheiny and Timberlake 1992; World Business Council for Sustainable Development 1992; OECD 2002; United Nations 2009). At the same time, strategic interactions across regions regarding environmental issues is unavoidable, due to the ability of local environmental policy to influence capital and population flows across sub-jurisdictions.3 Against this backdrop, this paper investigates the evolution of eco-efficiency over time, and assesses its association with peer effects and sub-jurisdiction political competition. We first build a simple theoretical model of political competition among regional government officials, each of whom allocates resources between economic production and environmental protection. Our comparative statics analysis suggests that ecoefficiency is negatively correlated with pollution intensity, and positively correlated with financial endowment and political emphasis on the environment. This means that as long as income increases, technology improves, and policy becomes environmentally friendlier, the level of eco-efficiency is expected to increase steadily over time. This is assuming, however, that environmental performance is a meaningful component of annual political assessment; otherwise, political competition would simply lead to a continual decline in eco-efficiency in a race-to-the-bottom fashion. This helps to explain the U-shape in the historical trend of eco-efficiency in countries that experience drastic changes in environmental policy and attitude. We also identify a pattern of convergence in the trends of eco-efficiency. Last, we find that a heightened level of political competition reduces eco-efficiency, while its impact on the rate of convergence varies depending on the different causes of convergence. The People’s Republic of China (PRC) provides a compelling setting to investigate this environmental issue. As the fastest-growing major economy in recent decades, the PRC has surged ahead of the rest of the world in terms of material consumption (e.g., construction minerals, metal ores, fossil fuels, biomass), creating intense environmental pressures, but at the same time it has remained among the most successful in improving its resource efficiency (UNDP 2013). While its growth has lifted millions out of poverty, it has also come with rising environmental challenges, such as deteriorating water quality, deforestation, and pollution, as well as frequent haze. In the PRC, the tournament competition among city leaders is an important feature of its political system (Xu 2011) which has motivated local authorities to compete with one another to facilitate rapid economic growth (Maskin, Qian, and Xu 2000; Li and Zhou 2005; Yu et al. 2016). However, as pollution is becoming one of the PRC’s greatest 1 This body of research, known as the environmental Kuznets curve (EKC) literature, has been enormously influential. The work by Grossman and Krueger (1995) is widely regarded as one of the earliest attempts at EKC hypotheses. An extensive overview of theoretical studies and empirical evidence regarding EKC can be found in Kaika and Zervas (2013). 2 The term “eco-efficiency” is a concept and philosophy geared toward sustainability, combining ecological and economic efficiency. 3 The pollution haven hypothesis was first developed by Pethig (1976) and McGuire (1982), and later improved by Copeland and Taylor (1994) and Levinson and Taylor (2008), among others. ADBI Working Paper 1125 Chen, Huang, and Yu 2 threats – especially to health – the country’s government has introduced a series of regulatory policies in recent decades. Although success has not occurred immediately,4 it seems that the desired turnaround has eventually occurred. For our empirical work, we use a data envelopment analysis (DEA) model to compute the eco-efficiency levels of 191 Chinese cities at and above the prefecture level between 2003 and 2015. The evolution in eco-efficiency in the PRC exhibited a U-shaped pattern during that period, with the rapid upward trend occurring from around 2013. We also examine heterogeneous patterns across different types of cities. Spatial econometric modeling analysis suggests that the net peer effect is significantly positive. Specifically, without consideration of political competition, a city would see its eco-efficiency decrease by 0.240% triggered by a 1% increase in its lower performing neighbors, but an increase by 0.775% with a 1% increase in its higher performing neighbors. In addition, the positive net peer effect holds significantly in heterogeneous groups. We then investigate the effects of strategic political competition on eco-efficiency. In particular, we focus on how a city’s eco-efficiency is affected by its neighboring cities’ eco-efficiency in two different scenarios: with and without the consideration of political competition. Our results show that in the presence of political competition, the peer effects (both disincentivizing and incentivizing effects) were amplified during the years 2003–2010, and attenuated during the years 2011–2015 as well as across the period as a whole. Our findings are robust to alternative measures of eco-efficiency. This study extends several strands of literature. First, productivity efficiency has been broadly used to measure environmental performance in the existing literature (Zhou et al. 2017) and DEA has been widely used to construct multi-factor measures of eco-efficiency (Mahlberg and Luptacik 2014). To accurately quantify the growthenvironment nexus and estimate eco-efficiency, we develop a comprehensive DEA model that simultaneously incorporates a non-convex metafrontier (O’Donnell et al. 2008; Tiedemann et al. 2011; Huang et al. 2013; Afsharian 2017; Afsharian and Podinovski 2018; Walheer 2018), super-efficiency (Andersen and Petersen 1993), and undesirable outputs (Tone 2003) into a modified slacks-based measure (MSBM) (Sharp et al. 2007). The strengths of our model comprise its comparable treatment of the efficiency of the same decision-making unit (DMU) in different years due to technological progress when considering the non-convex metafrontier, its identifiable treatment of DMUs located on efficient frontier to employ super-efficiency, its comprehensive treatment of DMUs’ slacks improvement in using MSBM, and its similarity to the real production process as regards undesirable outputs. Non-parametric DEA techniques or activity models make it possible to incorporate several environmental pressures as well as to measure eco-efficiency at specific environmental pressure levels. Ideally, measuring eco-efficiency for sub-jurisdictions within a country requires disaggregated data on inputs and outputs, which are usually unavailable in most 4 For example, in its National 10th Five-Year Plan (2001–2005), released in 2001, the central government for the first time added environmental protection and pollution reduction to its list of “national strategic goals,” and set a target to reduce pollutant discharges by 10% by the end of 2005. Under the new regulation framework, each province was assigned a specific target, and the provincial government officials were to be evaluated on, among other things, how well these targets were met. However, little improvement in environmental quality has been observed in the PRC based on data between 1998 and 2008, because the pollution mandates imposed by the central government have triggered strategic polluting responses from the provinces (Cai, Chen, and Gong 2016). ADBI Working Paper 1125 Chen, Huang, and Yu 3 developing countries. The unique city-level input and output data enable us to fill this gap in the Chinese context.5 Second, this study sheds new light on the ongoing research on environmental convergence. The concept of convergence, which has been adopted to assess whether poorer regions are able to catch up with relatively richer regions over time, has recently been extended to the study of environmental issues. 6 In theory, environmental convergence can be justified by income convergence and the environmental Kuznets curve (EKC). Given an inverted-U relation between income and emission (Grossman and Krueger 1995), environmental convergence is a plausible prediction if the low-income region in question exhibits a superior economic growth rate to its high-income counterpart. The existing research mainly focuses on testing the convergence of emission intensity (List 1999; Strazicich and List 2003; Ezcurra 2007; Romero-Avila 2008; Jobert, Karanfil, and Tykhonenko 2010) or energy efficiency (Mielnik and Goldemberg 2000; Markandya et al. 2006; Liddle 2009, 2010; Jaunky 2013; Meng et al. 2013), for which the approaches of unit root test and log-t test (Phillips and Sul 2007) have been widely adopted. However, most of these studies have only accounted for the environmental side of production processes (such as per capita emissions), and have only considered a single emission to represent the environmental impact of economic activity. Camarero et al. (2013) have attempted to measure eco-efficiency using the DEA model as well as to investigate its convergence across OECD countries. We extend this strand of literature by investigating the convergence of eco-efficiency in a sub-national jurisdiction context of the PRC, for which achieving a good balance between economic growth and environmental sustainability is of critical importance given the sheer size of economy and resource usage in this country. Moreover, we innovatively apply the two-regime spatial Durbin model to capture the peer effect on the convergence of eco-efficiency. A branch of the existing literature focuses on the “spillover effect” by examining differences in pollution levels or health outcomes between border and interior jurisdictions (Helland and Whitford 2003; Gray and Shadbegian 2004; Kahn 2004; Cai et al. 2016; Chen and Huang 2016). We incorporate the geographic spillover of eco-efficiency by including the attributes of neighboring regions. Indeed, a city’s eco-efficiency depends not just on its own attributes, but also on those of its peers. Failure to account for geographic spillovers in empirical analyses of eco-efficiency may lead to biased inferences. We find that the evolution of regional eco-efficiency depicts an interesting pattern of dynamics. Witnessing the excellent performance of its outstanding neighbors, a city is incentivized to catch up, leading to improvements in eco-efficiency. However, when a city outperforms its neighbors, its eco-efficiency improvement tends to slow down, perhaps due to complacency. This finding is important in that it predicts a pattern of convergence among different regions with different levels of eco-efficiency. Our analysis integrates political competition into the analysis of eco-efficiency. Strategic horizontal interaction among local governments has represented one of the main foci of public economics. Researchers generally find that for regions within a country, cross- 5 For example, one can collect the PRC’s provincial energy intensity data from the National Bureau of Statistics, but information on the country’s urban energy intensity is not consistently reported. The PRC’s urban energy consumption will be overestimated if we multiply provincial energy intensity statistics and prefecture gross domestic product (GDP). To overcome this shortcoming, Huang et al. (2018) have collected the PRC’s urban energy intensity and have estimated primary energy consumption using a bottom-up rather than a top-down approach. 6 The neoclassical growth theory developed by Solow (1956) predicts that a region’s growth rate in per capita income is inversely related to its initial per capita income, suggesting that poorer regions will catch up with relatively richer regions over time. A thorough review of the various definitions of convergence and approaches to testing it can be found in Islam (2003). ADBI Working Paper 1125 Chen, Huang, and Yu 4 border fiscal competition exists (Besley and Case 1992; Geys and Osterloh 2013; Janeba and Osterloh 2013; Agrawal 2015; Yu et al. 2016), with the degree of strategic interaction being more pronounced if governors are politically sensitive to fiscal policy changes in neighboring jurisdictions, and less pronounced if they are not (Elhorst and Freret 2009). Recently, the theory on regional strategic competition has been extended to investigate environmental issues, given the ability of local environmental policy to influence flows of capital and population across sub-jurisdictions. The allocation of authority over environmental decision making within a federal political system has also been discussed, deliberated, and agonized over for decades (Millimet 2013). Nevertheless, the debate cannot be settled without solid empirical evidence. Fredriksson and Millimet (2002) and Fredriksson et al. (2004) have provided evidence that the states of the United States set environmental regulations strategically based on the regulation of neighboring states. This paper extends the existing research by highlighting the impact of political competition on the convergence of eco-efficiency within a country. Our study provides both new theoretical extension and empirical evidence to the existing research on incentivizing bureaucrats with concrete evaluation criteria. The evaluation system and its impacts on the behavior of local government officials have constituted one of the main foci of public management literature (OECD 1994; Hood 2007). However, despite this topic’s importance, empirical evidence has remained mixed at best. Lockwood and Porcelli (2013) have shown that the comprehensive performance assessment system in England has enhanced the quality of public services, although no significant effect on efficiency has been discovered. By contrast, Rasul and Rogger (2015, 2018) claim that concrete targets tend to reduce public service delivery in Nigeria. Our theoretical model projects that without integrating environmental performance into governors’ evaluations, political competition inevitably leads to continual deterioration in environmental quality. The results of our empirical analysis echo these theoretical predictions by exhibiting the effectiveness of including environmental factors into the assessment system, altering the pattern of eco-efficiency over time. The remainder of this paper is organized as follows. Section 2 outlines our theoretical model, followed by a description of the methodology and the data in Section 3. In Section 4, we present and discuss the results. Section 5 concludes the paper. 2. THE THEORETICAL MODEL We build a simple model in which there are 𝑁𝑁 regions, each headed by a government official whose objective function is as follows: 𝑈𝑈𝑖𝑖=𝑈𝑈(𝑦𝑦𝑖𝑖,𝑃𝑃𝑖𝑖), (1) where 𝑦𝑦𝑖𝑖 is the GDP in region 𝑖𝑖, and 𝑃𝑃𝑖𝑖 is the local official’s probability of political promotion at the end of the year. It is reasonable to assume that 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕> 0, 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕> 0. (2) For the sake of simplicity, we do not consider factors such as corruption and bribery for political promotion, and instead assume a meritocratic process of competition among all 𝑁𝑁 local officials. The annual assessment is conducted by the higher government, which will then assign to each local official a score that considers both economic and environmental performance, the latter being measured in terms of regional environmental quality, 𝑞𝑞𝑖𝑖. We believe that it is reasonable to assume that, all else being ADBI Working Paper 1125 Chen, Huang, and Yu 11 3.2 Econometric Specification We use the spatial Durbin model with a two-regime framework as in Elhorst and Fréret (2009), with the added component of political competition given that we are investigating its impact on eco-efficiency. This kind of econometric specification includes the spatial lags of all the independent variables in addition to the spatial lag of the dependent variable with two regimes: 𝐸𝐸𝐸𝐸 𝑖𝑖𝑚𝑚 =𝛼𝛼+𝜌𝜌 1 𝑑𝑑 𝑖𝑖𝑚𝑚 �𝑤𝑤 𝑖𝑖𝑚𝑚 𝐸𝐸𝐸𝐸 𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 +𝜌𝜌 2 (1−𝑑𝑑 𝑖𝑖𝑚𝑚 )�𝑤𝑤 𝑖𝑖𝑚𝑚 𝐸𝐸𝐸𝐸 𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 +𝛽𝛽𝑋𝑋 𝑖𝑖𝑚𝑚 +�𝑤𝑤 𝑖𝑖𝑚𝑚 𝑋𝑋 𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 𝜃𝜃+𝜇𝜇 𝑖𝑖 +𝜆𝜆 𝑚𝑚 +𝑢𝑢 𝑖𝑖𝑚𝑚 (18) 𝑑𝑑𝑖𝑖𝑚𝑚=�1, 𝑖𝑖𝑖𝑖 𝐸𝐸𝐸𝐸𝑖𝑖𝑚𝑚>∑𝑤𝑤𝑖𝑖𝑖𝑖𝑚𝑚𝐸𝐸𝐸𝐸𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 ,𝑖𝑖≠𝑗𝑗 0, 𝑜𝑜𝑡𝑡ℎ𝑒𝑒𝑟𝑟𝑤𝑤𝑖𝑖𝑠𝑠𝑒𝑒 19) where 𝐸𝐸𝐸𝐸𝑖𝑖𝑚𝑚 denotes eco-efficiency of city 𝑖𝑖 at period 𝑡𝑡; 𝑁𝑁 is the number of spatial units; 𝑤𝑤𝑖𝑖𝑖𝑖 is an element of a spatial weight matrix 𝑊𝑊 that describes the spatial arrangement of the cities in the sample; and ∑𝑤𝑤𝑖𝑖𝑚𝑚𝑋𝑋𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 𝜃𝜃 is the spatial Durbin term. The binary variable 𝑑𝑑𝑖𝑖𝑚𝑚 takes a value of 1 if city 𝑖𝑖’s eco-efficiency level is higher than the weighted average of its neighbors, and 0 otherwise. Based on its set-up, it should be clear that 𝜌𝜌1 and 𝜌𝜌2 are the peer-effect coefficients. 𝜌𝜌1 measures how a city’s eco-efficiency might be affected by its neighboring regions with inferior environmental performance on average, while 𝜌𝜌2 is applicable in the scenario when the city is outperformed by its neighbors. The magnitude and the signs of 𝜌𝜌1 and 𝜌𝜌2 are important because they help us to determine changes in eco-efficiency levels (increase or decrease), as well as the pattern of the directional relationship between eco-efficiency trends (convergence or divergence). Finally, 𝑋𝑋𝑖𝑖𝑚𝑚 represents a vector of control variables that affect eco-efficiency, with associated parameters in vector 𝛽𝛽. In selecting the factors of eco-efficiency, we turn to the existing literature for wisdom. The renowned IPAT identity links anthropogenic impacts in environment (I) to three key driving forces – population size (P), affluence (A), and technology (T) – through a multiplicative function, i.e., I = PAT (Ehrlich and Holdren 1972; Fischer-Kowalski and Amann 2001; Ehrenfeld 2005). For our study, we adopt the framework of the STIRPAT model, a stochastic form of IPAT identity proposed by Dietz and Rosa (1997), and choose the following factors as control variables in 𝑋𝑋𝑖𝑖𝑚𝑚: a) Economic development (pgdp), wage level of employees (wage), and population agglomeration (popd). In this study, economic development is proxied by per capita GDP, population density is the demographic measurement, and average salary is used to represent the affluence of urban residents. b) Foreign direct investment (sfdi). FDI enhances the production efficiency of domestic enterprises via a technology spillover effect (Javorcik 2004), and is naturally an important agent in urban economic development. Based on data availability, we utilize the actual use of FDI as an indicator in this study. c) Industrial structure (sind). Following York et al. (2003), the percentage of GDP contributed by the secondary (industry) sector is used as the key indicator to reflect the economic structure of each city. ADBI Working Paper 1125 Chen, Huang, and Yu 12 d) Technology and innovation (inno). Jin et al. (2019) have found that technology and innovation have greater beneficial impacts on coastal cities than on inland cities in the PRC. In our study, we use city innovation to represent technological change, with a relevant indicator from Kou and Liu (2017). e) Environmental regulation (er). Based on data availability, environmental regulation is proxied by an indicator that aggregates the removal rate of SO2, the treatment rate of domestic sewage, and the comprehensive utilization ratio of solid waste. Commonly used in spatial econometrics, the spatial weights matrix 𝐖𝐖 models spatial interdependence among observations (LeSage and Pace 2009), or, specific to our study, it describes the exogenous pattern of how neighboring cities are spatially inter-connected to each other. In the spatial econometrics literature, weight matrices are conventionally constructed based on geographic proximity, such as contiguity and geographic distance; however, estimation results are not sensitive to the specifications of spatial weights matrices, according to LeSage and Pace (2014). In this study, we specify two different spatial weights matrices, the first being defined as follows: 7 𝑤𝑤𝑖𝑖𝑖𝑖=�1, if cities 𝑖𝑖 and 𝑗𝑗 belong to the same province 0, otherwise (20) Elements on the main diagonal of the matrix are set to zero by convention. There is one special note regarding the four municipalities: Beijing, Tianjin, Shanghai, and Chongqing. They are under the direct administration of the central government, so in reality they each have the same rank as provinces within the Chinese political hierarchical structure. However, for the purpose of our study they are set at the same level as other cities, i.e., these four municipalities would compete among themselves politically in the same sense that there are no other municipality cities from the same province. Next, we introduce the probability of political promotion, with the revised spatial weights matrix defined as follows: 𝑤𝑤�𝑖𝑖𝑖𝑖=�𝑐𝑐𝑖𝑖×𝑤𝑤𝑖𝑖𝑖𝑖, if cities 𝑖𝑖 and 𝑗𝑗 belong to the same province 0, otherwise 21) where 𝑐𝑐𝑖𝑖 represents the average political intensity during the study period, and is measured as below in Equation (22), following Liu et al. (2012): 𝑐𝑐𝑖𝑖=1 𝑇𝑇∑�∑𝛿𝛿𝑖𝑖𝑖𝑖 𝑄𝑄 𝑖𝑖=1 −𝛿𝛿𝑖𝑖𝑖𝑖 2𝑄𝑄 � 𝑇𝑇 𝑚𝑚=1 (22) where 𝑄𝑄 is the number of cities within the same province, T is the duration of the period, and 𝛿𝛿𝑖𝑖𝑚𝑚 denotes the number of Party secretary turnovers of city 𝑖𝑖 in period t. To reestimate Equation (18), it is also necessary that we revise the definition of the binary variable 𝑑𝑑𝑖𝑖𝑚𝑚 accordingly, as follows: 𝑑𝑑󰆻𝑖𝑖𝑚𝑚=�1, 𝑖𝑖𝑖𝑖𝐸𝐸𝐸𝐸𝑖𝑖𝑚𝑚>∑𝑤𝑤�𝑖𝑖𝑖𝑖𝑚𝑚𝐸𝐸𝐸𝐸𝑖𝑖𝑚𝑚 𝜕𝜕 𝑖𝑖=1 ,𝑖𝑖≠𝑗𝑗 0, 𝑜𝑜𝑡𝑡ℎ𝑒𝑒𝑟𝑟𝑤𝑤𝑖𝑖𝑠𝑠𝑒𝑒 (23) 7 To normalize the outside influence upon each city, the weights matrix is subject to standardization, so that elements in the same row sum up to one. ADBI Working Paper 1125 Chen, Huang, and Yu 13 3.3 Data In this study we examine 191 cities at prefecture level and above in the PRC during the period 2003–2015. Our sample does not include any of the following due to data unavailability: Hong Kong, China; Macau, China; Taipei,China; and Tibet Autonomous Region. The input and output data are introduced as follows. (1) Desirable output. We use data on real GDP, measured at constant 2010 prices, wherever applicable throughout this paper. (2) Undesirable output. We employ the entropy weight method to generate a composite environmental pollution index (EPI) of four pollutants: carbon dioxide emissions; industrial wastewater discharge; sulfur dioxide emissions; and industrial soot-dust emissions. We use Huang et al.’s (2018) approach to estimate carbon dioxide emissions, while the other three pollutants’ data are collected from the China City Statistical Yearbooks (CCSY), 2004–2016. (3) Labor force. We use the size of the employed population as a proxy in our study, based on data availability. (4) Capital stock. We adopt the perpetual inventory method by Ke and Xiang (2012) to estimate capital stock – 𝐾𝐾𝑖𝑖,𝑚𝑚=𝐼𝐼𝑖𝑖,𝑚𝑚+�1−𝜎𝜎𝑖𝑖,𝑚𝑚�𝐾𝐾𝑖𝑖,𝑚𝑚−1 – where the total capital stock of city 𝑖𝑖 at time 𝑡𝑡, 𝐾𝐾𝑖𝑖,𝑚𝑚, is the sum of fixed assets accumulated from the past year (𝜎𝜎𝑖𝑖,𝑚𝑚 is the depreciation rate) and new physical capital investment, 𝐼𝐼𝑖𝑖,𝑚𝑚. (5) Land usage. We use the construction land area as the proxy based on data accessibility. (6) Energy consumption. Information on primary energy consumption in 2003–2013 is directly extracted from Huang et al. (2018), and we also use their method to estimate for the data for the two years of 2014 and 2015. Primary energy consumption is converted to standard coal equivalent (SCE), the standard unit in Chinese energy statistics. Table 1: Descriptive Statistics Unit Obs. Mean Std. Dev. Min. Max. Panel A: Input and output variables for DEA Labor 104 persons 2,483 56.51 79.65 5.49 986.87 Capital 108 CNY 2,483 1,340.00 1,990.00 27.50 22,000.00 Land km2 2,483 14,908.09 12,618.12 1,565.00 90,659.00 Energy Tons of SCE 2,483 1,573.91 1,572.05 53.79 11,719.50 GDP 108 RMB 2,483 1,730.00 2,410.00 50.15 25,000.00 EPI – 2,483 4.03 3.70 0.07 56.19 Panel B: Variables used in econometric model ee – 2,483 0.47 0.20 0.20 1.16 pgdp CNY/person 2,483 3.76 4.31 0.30 42.25 wage – 2,483 10.20 0.55 5.93 11.64 popd 104 persons/km2 2,483 0.05 0.04 <0.01 0.27 sfdi – 2,483 0.13 0.20 0.00 1.44 sind – 2,483 0.49 0.11 0.15 0.91 inno – 2,483 0.08 0.39 0.00 8.49 er – 2,483 0.40 0.18 0.03 0.79 Note: CNY denotes Chinese Yuan. ADBI Working Paper 1125 Chen, Huang, and Yu 14 The descriptive statistics and correlation coefficients of the dependent and control variables are displayed in Tables 1 and 2, respectively. Definitions of each variable and data source are provided in Appendix B. Given that the correlation coefficients among the independent variables are small, the possibility of multicollinearity is dismissed. Table 2: Correlation Coefficients among Dependent and Control Variables Variable ee pgdp wage popd sfdi sind inno pgdp 0.28*** 1 wage –0.01 0.54*** 1 popd 0.38*** 0.31*** 0.16*** 1 sfdi 0.24*** 0.55*** 0.27*** 0.45*** 1 sind –0.02 0.20*** 0.17*** 0.11*** 0.12*** 1 inno 0.16*** 0.50*** 0.30*** 0.30*** 0.24*** –0.14*** 1 er –0.04** 0.36*** 0.65*** 0.19*** 0.22*** 0.18*** 0.18*** Note: *** p < 0.01, ** p < 0.05, * p < 0.1. 4. RESULTS AND DISCUSSION 4.1 Eco-efficiency: The National Trend We first check the national average of eco-efficiency (illustrated in Figure 3) and see how it has evolved over the years. The U-shaped pattern is immediately apparent. In particular, the eco-efficiency level declined in the earliest years of our study period; this adverse trend seemed to be halted in 2007, as the eco-efficiency level remained largely flat until around 2013, and then rose quite rapidly ever since. Figure 3: The Trend of Eco-efficiency in the PRC: 2003–2015 This phenomenon coincides with a critical period of change in environmental protection and policy making in the PRC. In 2010, the binding targets set for the National 11th Five- Year Plan (2005–2010) were largely met, with great strides made in terms of both energy conservation and emissions reduction. Environmental quality has improved significantly: .44 .46 .48 .5 .52 Eco-efficiency 2003 2005 2007 2009 2011 2013 2015 Year ADBI Working Paper 1125 Chen, Huang, and Yu 15 for example, the average sulfur dioxide concentrations in KEP cities have decreased by 26.3% compared to 2005. Encouraged by this success, the National 12th Five-Year Plan (2011–2015) set the ambitious goal of further reducing the total emissions of all major pollutants by 8% to 10%, and expanded the list of pollution indicators. Environmental protection was thus made a priority. According to the National 12th Five-Year Plan for Environmental Protection published by the State Council of China in 2011, for the first time ever local environmental quality was made an important consideration in the annual evaluations of government officials at all levels, and environmental protection was given the power of veto in this process. Such unprecedented change provides evidence of the central government’s will and determination. Moreover, given that it has been integrated into the political assessment of local officials, it has quickly started to take effect, as corroborated by the remarkable turnaround in the nationwide trend of eco-efficiency. 4.2 Eco-efficiency: Heterogeneities across Groups Understandably, eco-efficiency can vary quite drastically across different regions due to factors such as heterogeneity in resource endowment, geographic condition, and environmental policies. Next, we present four different criteria, each of which would help to categorize our sample of 191 Chinese cities into two mutually exclusive groups. In the PRC, cities in the eastern (or coastal) regions are generally at a more advanced level of economic growth, whereas their counterparts in the central and western (or inland) regions tend to perform worse economically. Using the first criterion, we divide all cities into two groups: eastern vs. central/western. In our sample, there are 71 cities in the eastern group. Different cities may exhibit different patterns of natural resource endowment and industrial structure, which in turn may have an impact on environmental quality. Based on the definition and list of resource-based (RB) cities in Planning for Sustainable Development of Resource-based Cities in China (2013-2020), a document issued by the State Council of China in 2013, we can divide all of the cities into two groups: RB vs. non-RB. Our sample has 78 RB cities. More than 100 cities in the PRC have been designated as KEP cities. They are usually cities of considerable political and/or economic importance, such as municipalities, provincial capital cities, open coastal cities, and major tourist cities, although the list also includes cities with historical environmental issues and a more pressing need for environmental protection. KEP cities are subject to more stringent environmental policies and standards compared to non-KEP ones. The list of 117 KEP cities in our study is compiled based on the China Environmental Yearbook, 2012–2016. According to the Law on the Prevention and Control of Atmospheric Pollution, areas with existing or potential acid rain or severe sulfur dioxide pollution may be designated as acid rain control zones or sulfur dioxide control zones, often collectively referred to as TCZ cities. Our sample contains 127 TCZ cities, and, similar to the KEP cities, they are subject to more stringent environmental policies and standards. Figure 4 illustrates the heterogeneities across groups. Figure 4a suggests that the gap between eastern and central/western cities seems to have widened over time. Similarly, in Figure 4c one can see that RB cities have always had significantly lower eco-efficiency levels than non-RB cities. Another concerning observation is that both RB cities and cities in the central/western regions witnessed a continual decline in eco-efficiency until around 2013, when the trend was reserved. This similarity might be explained by the fact that most RB cities are located in central/western regions with ADBI Working Paper 1125 Chen, Huang, and Yu 16 a historically unfavorable industrial structure, weaker economic foundation, less advanced technology, and more severe environmental issues. Figure 4: Eco-efficiency Trends: Heterogeneities across Groups According to Figure 4b, although KEP cities started with inferior records in eco-efficiency compared to their non-KEP counterparts, the gap gradually narrowed, until around 2012 when they overtook them. Considering that the group of KEP cities are a mixture of cities with historically poor environmental records and cities of political importance, this overall pattern of catching up and then overtaking makes sense. As can be seen in Figure 4d, the dynamic evolution of eco-efficiency in TCZ vs. non-TCZ looks very similar to the aforementioned KEP vs. non-KEP comparison. In addition to the above graphical evidence, we can further investigate the inter-group disparity in eco-efficiency with a violin plot and a mean test. In Table 3, we compare the means of eco-efficiency between each pair of groups. We can see that, on average, cities in the eastern region have superior eco-efficiency to those in the central/western regions; similarly, non-KEP, non-RB, and non-TCZ cities outperform their KEP, RB, and TCZ counterparts, respectively. Although the respective gaps vary between 0.02 and 0.10, the disparity in eco-efficiency between each pair of groups is statistically significant at the 1% level. The violin plot in Figure 5 largely confirms the patterns of the relationships between each pair, including that the median value of the non-TCZ cities’ eco-efficiency is higher than that of the TCZ cities, albeit only slightly. .1 .3 .5 .7 Eco-efficiency 2003 2015 (a) Total Eastern Central/Western .1 .3 .5 .7 2003 2015 (b) Total KEP Non-KEP .1 .3 .5 .7 2003 2015 (c) Total RB Non-RB .1 .3 .5 .7 2003 2015 (d) Total TCZ Non-TCZ ADBI Working Paper 1125 Chen, Huang, and Yu 17 Table 3: Mean Test of Eco-efficiency Statistics Eastern KEP RB TCZ Mean 0.5050 0.4585 0.4088 0.4578 Max. 1.0924 1.0919 1.1632 1.0924 Min. 0.1984 0.1984 0.2136 0.1984 Std.Dev. 0.2094 0.1898 0.1551 0.1882 Central/Western Non-KEP Non-RB Non-TCZ Mean 0.4440 0.4795 0.5065 0.4841 Max. 1.1632 1.1632 1.0924 1.1632 Min. 0.1985 0.2330 0.1984 0.2360 Std.Dev. 0.1770 0.1946 0.2043 0.1979 T-TEST 0.0610*** –0.0210*** –0.0977*** –0.0263*** Note: *** p < 0.01. Figure 5: Distributions of Eco-efficiency for Different Cities (Color figure online) 4.3 Baseline Results We present the estimation results of Equation (18) in Table 4, with a particular interest in examining how a city’s eco-efficiency is affected by the performance of its neighbors. The case without the consideration of political competition is shown in the first column, and the case with in the second. Our estimates of 𝜌𝜌1 and 𝜌𝜌2 depict an interesting pattern of dynamics. It seems that a city may be incentivized to emulate its outstanding neighbors after witnessing their excellence, resulting in improved eco-efficiency. On the other hand, outperforming its neighbors seems to serve as a disincentivizing factor for the leader, perhaps due to complacency, leading to diminished eco-efficiency. Based on our findings, we can call 𝜌𝜌1 the “incentivizing effect coefficient,” and 𝜌𝜌2 the “disincentivizing effect coefficient.” This finding is important because it reveals a pattern of convergence among different regions with different levels of eco-efficiency. .2 .4 .6 .8 1 1.2 Eco-efficiency Eastern Central/Western KEP Non-KEP RB Non-RB TCZ Non-TCZ ADBI Working Paper 1125 Chen, Huang, and Yu 18 It can also be noted that the positive, incentivizing effect appears to be stronger than the negative, disincentivizing effect. According to Table 4, without consideration of political competition, a city would see its eco-efficiency decrease by 0.240% triggered by a 1% increase in its lower performing neighbors, but an increase by 0.775% with a 1% increase in its higher performing neighbors. This results in a positive net peer effect, reinforcing the upward trend of eco-efficiency in addition to the pattern of convergence. All results are statistically significant at the 1% level. Moreover, among the seven control variables, we find that economic development (pgdp), average salary (wage), population agglomeration (popd), and technology spillover of innovation (inno) have positive effects on eco-efficiency, while industrial structure (sind) has negative effects. It also seems that popd exerts the greatest positive impact on eco-efficiency out of all the control variables. Table 4: Baseline Estimates of Political Competition on Eco-efficiency (2003–2015) Variable Without Political Competition With Political Competition 𝒅𝒅 𝒅𝒅 � 𝜌𝜌 1 –0.2404*** (–5.6749) –0.2376*** (–5.6174) 𝜌𝜌 2 0.7746*** (17.4174) 0.6756*** (15.1150) pgdp 0.0109*** (6.5396) 0.0112*** (6.6504) wage 0.0806*** (5.1131) 0.0742*** (4.6578) popd 0.7503** (2.2685) 0.7034** (2.1057) sfdi 0.0190 (0.5232) 0.0289 (0.7939) sind –0.0835 (–1.5109) –0.0973* (–1.7447) inno 0.0278*** (3.3584) 0.0286*** (3.4058) er 0.0091 (0.5334) 0.0125 (0.7264) Observations 2,483 2,483 R-squared 0.8548 0.8506 Log-likelihood 2,897.7201 2,881.6590 Notes: tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. The definition of each variable can be found in Appendix B. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. 4.4 Effects of Political Competition: Sub-sample Estimations We now turn our attention to the effects of political competition on eco-efficiency. Revisiting Table 4, we can see that when political competition is taken into consideration, both disincentivizing and incentivizing effects decline in magnitude in terms of absolute value, as can also be seen in Table 4. If a city currently outperforms its competitors, its eco-efficiency is induced to drop, but at a lower rate of 0.238%; however, if it currently has an inferior eco-efficiency level than its competitors on average, this (positive) peer pressure effect is reported as a 0.676% increase. Either way, it seems that the convergence is slowed down; moreover, the net peer effect remains positive, yet it is reduced in magnitude compared to the scenario without consideration of political ADBI Working Paper 1125 Chen, Huang, and Yu 19 competition. We further investigate these effects for different periods in our study. Based on the estimation results summarized in Table 5, we can see that political competition amplified the peer effects (both disincentivizing and incentivizing effects) during the 2003–2010 period, but attenuated them during the 2011–2015 period, as well as across the period as a whole. Table 5: Impacts of Political Competition on Eco-efficiency: Different Periods Variable 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌 1 –0.3474*** (–5.6191) –0.4145*** (–6.7515) –0.4334*** (–5.6862) –0.3818*** (–4.9151) 𝜌𝜌 2 0.6937*** (10.3821) 0.6940*** (10.2894) 0.8854*** (9.1345) 0.6759*** (8.5961) pgdp 0.0160*** (5.6573) 0.0155*** (5.4644) 0.0139*** (3.5823) 0.0145*** (3.7084) wage 0.0785*** (4.4271) 0.0729*** (4.0816) 0.0220 (0.6953) 0.0380 (1.1940) popd 0.5119 (1.4076) 0.4198 (1.1489) 1.4145 (1.5062) 1.4482 (1.5260) sfdi –0.0588 (–1.1446) –0.0555 (–1.0786) –0.0511 (–0.7576) –0.0267 (–0.3959) sind –0.1496* (–1.8875) –0.1325* (–1.6668) 0.0780 (0.7160) 0.0891 (0.8117) inno –0.0197 (–0.5428) 0.0379 (1.0389) –0.0009 (–0.0696) 0.0074 (0.5382) er 0.0035 (0.1446) –0.0002 (–0.0101) 0.0293 (1.1836) 0.0364 (1.4556) Observations 1,528 1,528 955 955 R-squared 0.8810 0.8791 0.9163 0.9139 Log-likelihood 1,902.8498 1,900.0051 1,450.7196 1,448.4074 Notes: tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. The definition of each variable can be found in Appendix B. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. We now take a closer look at peer effects across different regions and periods, with the estimation results summarized in Table 6. We can see that all estimated values of 𝜌𝜌1 with statistical significance are negative, and that all estimated values of 𝜌𝜌2 are positive. This is true both with and without the consideration of political competition, indicating an unambiguous pattern of convergence in eco-efficiency, in line with our theoretical projection. In the following step, we examine the effects of political competition on the pattern of convergence by calculating and analyzing 𝜌𝜌2−𝜌𝜌1 based on information from Table 6. Keep in mind that 𝜌𝜌1 and 𝜌𝜌2 both contribute to convergence, although they work in opposite directions. Furthermore, given that 𝜌𝜌1< 0 and 𝜌𝜌2> 0, we know that 𝜌𝜌2−𝜌𝜌1 is the same as |𝜌𝜌1|+|𝜌𝜌2|. We use this term to loosely measure the rate (or degree) of convergence, and use Figure 6 (a, b, c) to illustrate our findings. Note that the values estimated without political competition are marked by circles, while diamonds represent values with. First, it is easy to reaffirm the pattern of convergence, as all values appear on the right side of the zero value. Next, regarding the effect of political competition, we can see that ADBI Working Paper 1125 Chen, Huang, and Yu 20 it reduces the speed of convergence, based on the full sample during the entire period 2003–2015. When we break the sample down to different groups and different periods, the experiences vary. Particularly striking is the stark difference before and after 2010, a critical time of transition in the PRC when the 11th Five-Year Plan was completed and the 12th Five-Year Plan was about to commence. The most recent period witnessed much more significant (slowing) effects of political competition on the convergence of eco-efficiency, as illustrated. Although it would be intriguing to explore the spatial, temporal, and spatiotemporal heterogeneities in the convergence pattern, in our opinion an even more meaningful task is to identify the direction of the trend in eco-efficiency, as well as how it is affected by political competition. As previously mentioned, 𝜌𝜌1 and 𝜌𝜌2 work in opposite directions, and the former can perhaps be labeled as the negative peer effect out of the two. We calculate and analyze 𝜌𝜌1+𝜌𝜌2, and use this term to loosely measure the “net” peer effect for a city: it should be obvious that both its sign and its magnitude would disclose important information about the trend of eco-efficiency for any region. Based on Figure 6 (d, e, f), we find that the net peer effect is positive for the full sample and for most groups in general, which is encouraging because it indicates an upward trend of eco-efficiency. The only two exceptions on the chart are eastern and non-RB groups, both of which seem to receive negative net peer effects over time. This should not come as a complete surprise, however, as their opposite groups (central/western, and RB cities, respectively) have struggled historically and are responsible for inflicting a large 𝜌𝜌1 in terms of absolute value. This largely reaffirms the information in Figure 4, which displays the heterogeneous patterns of the trend in eco-efficiency across groups, and shows that only in very recent years have the trends started to turn upward in central/western regions and for RB cities. It is also interesting to note that the negative net peer effect has almost vanished for the eastern group in the most recent period, while the same cannot yet be said about the non-RB cities. This may be linked to the timing of different policies, as China Western Development8 was officially launched about two decades ago, long before the central government’s initiatives targeting resource-based cities (2013), so it may be natural to expect a later turning point. Lastly, it seems that political competition has an adverse impact on eco-efficiency for the full sample and for most groups, as suggested by Figure 4 (d, e, f). Again, this finding confirms our theoretical model’s projection. 8 In late 1999, after two decades of pursuing coastal development, the PRC’s leaders announced a change in the PRC’s regional development strategy and initiated the western drive. Starting from 2001 the PRC government has offered preferential policies to the western region in terms of capital input, investment environment, internal and external opening-up, development of science and education, and human resources. ADBI Working Paper 1125 Chen, Huang, and Yu 27 Markandya, A., Pedroso-Galinato, S., and Streimikiene, D. (2006). Energy intensity in transition economies: is there convergence towards the EU average?. Energy Economics, 28(1), 121–145. Maskin, E., Qian, Y., and Xu, C. (2000). Incentives, information, and organizational form. The Review of Economic Studies, 67(2), 359–378. McGuire, M. C. (1982). Regulation, factor rewards, and international trade. Journal of Public Economics, 17(3), 335–354. Meng, M., Payne, J. E., and Lee, J. (2013). Convergence in per capita energy use among OECD countries. Energy Economics, 36, 536–545. Mielnik, O., and Goldemberg, J. (2000). Converging to a common pattern of energy use in developing and industrialized countries. Energy Policy, 28(8), 503–508. Millimet, D. L. (2013). Environmental federalism: a survey of the empirical literature. Case Western Reserve Law Review, 64, 1669–1757. O’Donnell, C. J., Rao, D. P., and Battese, G. E. (2008). Metafrontier frameworks for the study of firm-level efficiencies and technology ratios. Empirical economics, 34(2), 231–255. OECD. (1994). Performance Management in Government: Performance Measurement and Results-oriented Measurement Technical report. OECD, Paris. OECD Secretariat. (2002). Indicators to measure decoupling of environmental pressure from economic growth. Sustainable Development, available at http://www.docstoc.com/docs/84838188/oecd_decoupling. Pethig, R. (1976). Pollution, welfare, and environmental policy in the theory of comparative advantage. Journal of Environmental Economics and Management, 2(3), 160–169. Phillips, P. C., and Sul, D. (2007). Transition modeling and econometric convergence tests. Econometrica, 75(6), 1771–1855. Rasul, I., and Rogger, D. (2015). The impact of ethnic diversity in bureaucracies: Evidence from the Nigerian civil service. American Economic Review, 105(5), 457–61. Rasul, I., and Rogger, D. (2018). Management of bureaucrats and public service delivery: Evidence from the Nigerian civil service. The Economic Journal, 128(608), 413–446. Romero-Ávila, D. (2008). Convergence in carbon dioxide emissions among industrialised countries revisited. Energy Economics, 30(5), 2265–2282. Schmidheiny, S., and Timberlake, L. (1992). Changing course: A global business perspective on development and the environment (Vol. 1). MIT press. Sharp, J. A., Meng, W., and Liu, W. (2007). A modified slacks-based measure model for data envelopment analysis with ‘natural’ negative outputs and inputs. Journal of the Operational Research Society, 58(12), 1672–1677. Solow, R. M. (1956). A contribution to the theory of economic growth. Quarterly Journal of Economics, 70(1), 65–94. Strazicich, M. C., and List, J. A. (2003). Are CO 2 emission levels converging among industrial countries?. Environmental and Resource Economics, 24(3), 263–271. ADBI Working Paper 1125 Chen, Huang, and Yu 28 Tiedemann, T., Francksen, T., and Latacz-Lohmann, U. (2011). Assessing the performance of German Bundesliga football players: a non-parametric metafrontier approach. Central European Journal of Operations Research, 19(4), 571–587. Tone, K. (2001). A slacks-based measure of efficiency in data envelopment analysis. European Journal of Operational Research, 130(3), 498–509. ———. (2003). Dealing with undesirable outputs in DEA: A slacks-based measure (SBM) approach. GRIPS Research Report Series, 2003. United Nations. (2009). Eco-efficiency indicators: measuring resource-use efficiency and the impact of economic activities on the environment. Greening of Economic Growth Series, ST/ESCAP/2561. UNDP. (2013). Resource Efficiency: Economics and Outlook for China, available at http://wedocs.unep.org/handle/20.500.11822/8119. Walheer, B. (2018). Aggregation of metafrontier technology gap ratios: The case of European sectors in 1995–2015. European Journal of Operational Research, 269(3), 1013–1026. Xu, C. (2011). The fundamental institutions of China’s reforms and development. Journal of Economic Literature, 49, 1076–1151. York, R., Rosa, E. A., and Dietz, T. (2003). STIRPAT, IPAT and ImPACT: analytic tools for unpacking the driving forces of environmental impacts. Ecological Economics, 46(3), 351–365. Yu, J., Zhou, L. A., and Zhu, G. (2016). Strategic interaction in political competition: Evidence from spatial effects across Chinese cities. Regional Science and Urban Economics, 57, 23–37. Zhou, P., Delmas, M. A., and Kohli, A. (2017). Constructing meaningful environmental indices: A nonparametric frontier approach. Journal of Environmental Economics and Management, 85, 21–34. ADBI Working Paper 1125 Chen, Huang, and Yu 29 APPENDIX A First, we derive the first-order condition (FOC) for region 𝑖𝑖, with the following optimization condition: (𝛼𝛼+𝑁𝑁)𝑞𝑞𝑖𝑖−𝛽𝛽(𝑏𝑏𝑖𝑖+ 1)𝑟𝑟𝑖𝑖𝑖𝑖𝑦𝑦𝑖𝑖= 0. Combined with (8), we have the following expanded conditions, one for each region: (𝛼𝛼+𝑁𝑁)(𝑞𝑞1 �−𝑟𝑟11𝑒𝑒1−𝑟𝑟12𝑒𝑒2−𝑟𝑟13𝑒𝑒3−⋯−𝑟𝑟1𝜕𝜕𝑒𝑒𝜕𝜕)−𝛽𝛽(𝑏𝑏1+ 1)𝑟𝑟11𝑦𝑦1= 0, (𝛼𝛼+𝑁𝑁)(𝑞𝑞2 �−𝑟𝑟21𝑒𝑒1−𝑟𝑟22𝑒𝑒2−𝑟𝑟23𝑒𝑒3−⋯−𝑟𝑟2𝜕𝜕𝑒𝑒𝜕𝜕)−𝛽𝛽(𝑏𝑏2+ 1)𝑟𝑟22𝑦𝑦2= 0, (𝛼𝛼+𝑁𝑁)(𝑞𝑞3 �−𝑟𝑟31𝑒𝑒1−𝑟𝑟32𝑒𝑒2−𝑟𝑟33𝑒𝑒3−⋯−𝑟𝑟3𝜕𝜕𝑒𝑒𝜕𝜕)−𝛽𝛽(𝑏𝑏3+ 1)𝑟𝑟33𝑦𝑦3= 0, ……. (𝛼𝛼+𝑁𝑁)(𝑞𝑞𝜕𝜕 �−𝑟𝑟𝜕𝜕1𝑒𝑒1−𝑟𝑟𝜕𝜕2𝑒𝑒2−𝑟𝑟𝜕𝜕3𝑒𝑒3−⋯−𝑟𝑟𝜕𝜕𝜕𝜕𝑒𝑒𝜕𝜕)−𝛽𝛽(𝑏𝑏𝜕𝜕+ 1)𝑟𝑟𝜕𝜕𝜕𝜕𝑦𝑦𝜕𝜕= 0. We sum up all the above FOCs and utilize the property 𝑟𝑟𝑖𝑖𝑖𝑖+∑𝑟𝑟𝑖𝑖𝑖𝑖= 1 𝑖𝑖 as in (9), and can have the following: (𝛼𝛼+𝑁𝑁)�(𝑞𝑞𝚤𝚤 �−𝑒𝑒𝑖𝑖) 𝑖𝑖−𝛽𝛽�(𝑏𝑏𝑖𝑖+ 1)𝑟𝑟𝑖𝑖𝑖𝑖𝑦𝑦𝑖𝑖 𝑖𝑖= 0. Assuming structural symmetry among regions, and combining equation (7), we can derive the results as (10): 𝑦𝑦𝑖𝑖=(𝛼𝛼+𝑁𝑁)(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �) (𝑏𝑏𝑖𝑖+ 1)(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖). For results (12), we have: 𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑚𝑚𝑖𝑖=𝑏𝑏𝑖𝑖(𝛼𝛼+𝑁𝑁)𝑞𝑞𝚤𝚤 � (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖2(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)> 0, 𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑏𝑏 =−(𝛼𝛼+𝑁𝑁)(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �) (𝑏𝑏𝑖𝑖+ 1)2𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)< 0, 𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝛽𝛽 =𝑏𝑏𝑖𝑖(𝛼𝛼+𝑁𝑁)(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �)𝑟𝑟𝑖𝑖𝑖𝑖 (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)2> 0, 𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑞𝑞𝚤𝚤 �=−𝑏𝑏𝑖𝑖(𝛼𝛼+𝑁𝑁) (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)< 0, 𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑁𝑁 =−𝑏𝑏𝑖𝑖(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �)𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖 (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)2< 0. ADBI Working Paper 1125 Chen, Huang, and Yu 30 As for results (13) and (14), we have 𝜕𝜕2𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑚𝑚𝑖𝑖2=−2𝑏𝑏𝑖𝑖(𝛼𝛼+𝑁𝑁)𝑞𝑞𝚤𝚤 � (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖3(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)< 0, 𝜕𝜕2𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝛽𝛽2=−2𝑏𝑏𝑖𝑖(𝛼𝛼+𝑁𝑁)(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �)𝑟𝑟𝑖𝑖𝑖𝑖2 (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)3< 0, 𝜕𝜕 𝜕𝜕𝑁𝑁�𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝑚𝑚𝑖𝑖�=𝑏𝑏𝑖𝑖𝑞𝑞𝚤𝚤 �𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖 (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖2(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)2> 0, 𝜕𝜕 𝜕𝜕𝑁𝑁�𝜕𝜕𝑒𝑒𝑒𝑒𝑖𝑖 𝜕𝜕𝛽𝛽�=−𝑏𝑏𝑖𝑖(𝑚𝑚𝑖𝑖+𝑞𝑞𝚤𝚤 �)𝑟𝑟𝑖𝑖𝑖𝑖(𝛼𝛼+𝑁𝑁−𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖) (𝑏𝑏𝑖𝑖+ 1)𝑚𝑚𝑖𝑖(𝛼𝛼+𝑁𝑁+𝛽𝛽𝑟𝑟𝑖𝑖𝑖𝑖)3< 0. ADBI Working Paper 1125 Chen, Huang, and Yu 31 APPENDIX B Table B1: Definitions of Variables and Data Sources Variable Definition Source(s) Panel A: DEA model Input Labor Size of employed population CCSY (2004–2016) Input Capital Total capital stock Estimated by the perpetual inventory method, CCSY (2004–2016) and China Statistical Yearbook (CSY, 2004–2016) Input Land Construction land area CCSY (2004–2016) Input Energy Primary energy consumption GDP energy intensity (manually collected from various official documents) multiplied by GDP, China Energy Statistical Yearbook (CESY 2004–2016); Huang et al. (2018) Desirable output GDP Gross domestic product CCSY (2004–2016) Undesirable output EPI Composite environmental pollution index CCSY (2004–2016); CEY (2004– 2016); Huang et al. (2018) Panel B: Econometric model Dependent variable ee Eco-efficiency Measured by Model (17) Control variables pgdp Per capita GDP CCSY (2004–2016) wage Logarithm value of average salary CCSY (2004–2016) popd Population density CCSY (2004–2016) sfdi The ratio of FDI to GDP CCSY (2004–2016) sind The percentage of GDP contributed by the secondary industry sector CCSY (2004–2016) inno City innovation Kou and Liu (2017) er Environmental regulation CCSY (2004–2016) ADBI Working Paper 1125 Chen, Huang, and Yu 32 APPENDIX C: ROBUSTNESS CHECKS: ALTERNATIVE MEASURE OF ECO-EFFICIENCY Table C1: Estimation Results for Different Periods: Full Sample Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.3428*** –0.3103*** –0.6012*** –0.6780*** –0.4225*** –0.3098*** (–7.4022) (–6.8630) (–8.3283) (–9.3292) (–5.4546) (–4.1369) 𝜌𝜌2 0.6297*** 0.5387*** 0.5670*** 0.5218*** 0.8362*** 0.6613*** (12.7283) (11.3725) (7.2437) (6.5945) (10.4018) (8.6647) pgdp 0.0118*** 0.0134*** 0.0212*** 0.0214*** 0.0129*** 0.0133*** (7.2420) (8.2464) (7.5770) (7.6611) (3.8216) (4.0397) wage 0.0904*** 0.0834*** 0.0803*** 0.0765*** 0.0343 0.0495* (5.9655) (5.4909) (4.6492) (4.4015) (1.2568) (1.8467) popd 1.1843*** 1.0793*** 1.0535*** 0.9880*** 1.0586 1.3534* (3.7212) (3.3896) (2.9746) (2.7789) (1.3064) (1.6945) sfdi 0.0108 0.0103 –0.1480*** –0.1259** –0.0225 0.0014 (0.3079) (0.2972) (–2.9585) (–2.5173) (–0.3861) (0.0239) sind –0.0661 –0.1047** –0.1045 –0.1407* –0.0538 –0.0292 (–1.2428) (–1.9699) (–1.3522) (–1.8177) (–0.5738) (–0.3156) inno 0.0284*** 0.0276*** –0.0595* –0.0102 0.0162 0.0249** (3.5508) (3.4245) (–1.6752) (–0.2855) (1.3719) (2.1372) er 0.0096 0.0112 –0.0096 –0.0154 0.0262 0.0278 (0.5877) (0.6866) (–0.4084) (–0.6578) (1.2251) (1.3188) Observations 2,483 2,483 1,528 1,528 955 955 R-squared 0.8302 0.8286 0.8563 0.8542 0.9223 0.9235 Log-likelihood 3,049.5805 3,045.1621 1,988.4864 1,989.5981 1,598.9133 1,608.2151 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. Table C2: Estimation Results for Different Periods: Eastern Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.6369*** –0.5547*** –1.4269*** –1.3838*** –0.3927** –0.2485* (–6.1724) (–5.6238) (–6.8649) (–6.7614) (–2.5405) (–1.7210) 𝜌𝜌2 0.3183*** 0.2837*** –0.3936* –0.3991* 0.4815*** 0.3528*** (3.5191) (3.1673) (–1.9060) (–1.8608) (3.8306) (2.8361) pgdp 0.0112*** 0.0122*** 0.0255*** 0.0254*** 0.0039 0.0079 (5.3242) (5.7902) (7.8725) (7.5945) (0.6602) (1.4405) wage 0.1368*** 0.1236*** 0.1607** 0.1546** 0.2028*** 0.1942 (2.9499) (2.6574) (2.2706) (2.1266) (2.6883) (2.7799) popd 0.9831*** 0.9758*** 0.8820** 0.8200** 1.5674 1.4552 (2.8940) (2.8661) (2.4983) (2.2667) (0.9922) (1.0014) sfdi –0.0255 –0.0273 –0.2240*** –0.1875*** 0.0935 0.1218 (–0.5969) (–0.6351) (–3.9289) (–3.2237) (0.8308) (1.1737) sind –0.2243** –0.2466** 0.0215 0.0150 –0.9367*** –0.9868*** (–2.2626) (–2.4772) (0.1628) (0.1111) (–2.6633) (–3.0335) inno 0.0254*** 0.0216** –0.0497 0.0023 0.0066 0.0184 (2.9376) (2.4679) (–1.2821) (0.0583) (0.3421) (1.0366) er 0.0050 –0.0030 –0.0341 –0.0393 0.0554 0.0096 (0.1859) (–0.1128) (–1.0051) (–1.1327) (1.0201) (0.1897) Observations 923 923 568 568 355 355 R-squared 0.8641 0.8635 0.8686 0.8620 0.9087 0.9218 Log-likelihood 1,162.1211 1,158.8959 788.2658 774.6058 486.5910 513.2557 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. ADBI Working Paper 1125 Chen, Huang, and Yu 33 Table C3: Estimation Results for Different Periods: Central/Western Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.3209*** –0.3438*** –0.5592*** –0.5624*** –0.2909*** –0.3108*** (–5.4100) (–5.6543) (–6.6471) (–6.6083) (–3.5076) (–3.6347) 𝜌𝜌2 0.6260*** 0.5925*** 0.6913*** 0.6580*** 0.9817*** 0.9307*** (9.0413) (8.5919) (7.2847) (6.9685) (10.5944) (10.0887) pgdp 0.0199*** 0.0201*** 0.0184*** 0.0194*** 0.0299*** 0.0302*** (5.9539) (6.0449) (2.7660) (2.9439) (8.6140) (8.7209) wage 0.0874*** 0.0842*** 0.0818*** 0.0777*** –0.0156 –0.0160 (5.4343) (5.2334) (4.5099) (4.2995) (–0.8840) (–0.9084) popd 3.3999*** 3.4084*** 1.4807 1.3506 1.6535** 1.6609** (3.4944) (3.5095) (0.7755) (0.7110) (2.4813) (2.4933) sfdi –0.0249 –0.0243 –0.0464 –0.0205 –0.0753 –0.0617 (–0.3622) (–0.3532) (–0.4293) (–0.1903) (–1.3593) (–1.1136) sind 0.0132 –0.0176 –0.1883* –0.2216** 0.1054* 0.1241** (0.2039) (–0.2713) (–1.9123) (–2.2640) (1.8476) (2.1774) inno 0.1485*** 0.1510*** 0.5039*** 0.5256*** 0.1096*** 0.1098*** (4.2218) (4.2948) (2.8141) (2.9491) (4.2909) (4.3025) er 0.0216 0.0285 0.0116 –0.0018 –0.0022 0.0023 (1.0411) (1.3789) (0.3595) (–0.0545) (–0.1552) (0.1625) Observations 1,560 1,560 960 960 600 600 R-squared 0.7899 0.7902 0.8476 0.8489 0.9488 0.9488 Log-likelihood 1,915.9532 1,920.8198 1,232.6034 1,237.0253 1,344.4720 1,345.9532 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. Table C4: Estimation Results for Different Periods: KEP Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.3192*** –0.3130*** –0.5518*** –0.5803*** –0.5461*** –0.4065*** (–6.5814) (–6.3501) (–7.8627) (–8.1817) (–5.8105) (–4.3952) 𝜌𝜌2 0.6251*** 0.5669*** 0.7214*** 0.7347*** 0.5667*** 0.4154*** (12.8856) (11.9720) (10.1114) (10.1925) (6.2993) (4.9261) pgdp 0.0106*** 0.0121*** 0.0109*** 0.0099*** 0.0151*** 0.0143*** (5.2271) (5.9993) (3.1962) (2.8546) (3.3360) (3.2355) wage 0.0976*** 0.0939*** 0.0716** 0.0678** 0.0543 0.0561 (3.4667) (3.3065) (2.2635) (2.1305) (1.1217) (1.1783) popd 1.3035** 1.2316** 1.8431* 0.8638 1.0238 1.0261 (2.1555) (2.0206) (1.6780) (0.7833) (1.0059) (1.0308) sfdi 0.0137 0.0035 –0.1575*** –0.1272*** –0.0999 –0.0972 (0.3736) (0.0953) (–3.2215) (–2.5877) (–1.4253) (–1.4094) sind 0.0499 0.0221 –0.0119 –0.0191 0.2080 0.2656 (0.6460) (0.2834) (–0.1265) (–0.2016) (1.1652) (1.5121) inno 0.0337*** 0.0323*** –0.0227 0.0444 0.0117 0.0226* (4.1417) (3.9206) (–0.6759) (1.3284) (0.8934) (1.7631) er 0.0330 0.0313 –0.0221 –0.0352 0.0252 0.0247 (1.5004) (1.4095) (–0.7799) (–1.2346) (0.7633) (0.7615) Observations 1,521 1,521 936 936 585 585 R-squared 0.8676 0.8656 0.9047 0.9037 0.9281 0.9304 Log-likelihood 1,976.6216 1,970.1381 1,389.8528 1,387.1780 962.3307 969.0667 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. ADBI Working Paper 1125 Chen, Huang, and Yu 34 Table C5: Estimation Results for Different Periods: Non-KEP Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.3322*** –0.3368*** –0.5409*** –0.5075*** –0.8947*** –0.4449*** (–4.2525) (–4.1955) (–4.7261) (–4.3994) (–5.5635) (–3.2470) 𝜌𝜌2 0.4911*** 0.4485*** 0.5071*** 0.4356*** 0.7812*** 0.5723*** (7.4421) (6.5867) (5.6729) (4.6359) (12.0694) (5.5533) pgdp 0.0275*** 0.0273*** 0.0361*** 0.0373*** 0.0022 0.0047 (6.5041) (6.4361) (5.9290) (6.1443) (0.1889) (0.4079) wage 0.0696*** 0.0671*** 0.0877*** 0.0839*** –0.0337 –0.0386 (3.5845) (3.4456) (3.8060) (3.6318) (–1.1830) (–1.3362) popd 1.1239*** 1.1672*** 1.3378*** 1.3588*** –1.0647 –1.2596 (2.7742) (2.8726) (2.9895) (3.0364) (–0.8311) (–0.9742) sfdi –0.0729 –0.0695 0.0258 0.0673 –0.0804 –0.0602 (–0.8084) (–0.7688) (0.1850) (0.4831) (–0.6779) (–0.5022) sind –0.3093*** –0.3475*** –0.2941** –0.3538** –0.1519 –0.1489 (–3.6517) (–4.0982) (–2.0857) (–2.5200) (–1.5808) (–1.5377) inno –1.5013*** –1.3712*** –3.5495 –3.6178 –0.9904*** –1.0722*** (–6.6829) (–6.0520) (–1.5458) (–1.5766) (–3.5420) (–3.7853) er –0.0249 –0.0169 –0.0069 –0.0142 0.0243 0.0244 (–0.9437) (–0.6377) (–0.1693) (–0.3467) (0.9928) (0.9862) Observations 962 962 592 592 370 370 R-squared 0.7902 0.7890 0.8070 0.8070 0.9297 0.9284 Log-likelihood 1,158.6718 1,156.1296 696.6822 695.7065 740.0271 727.2455 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. Table C6: Estimation Results for Different Periods: RB Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.1390** –0.1510*** –0.3480*** –0.3582*** –0.0450 –0.0348 (–2.4530) (–2.6349) (–4.0354) (–4.1264) (–0.6172) (–0.4839) 𝜌𝜌2 0.4824*** 0.4822*** 0.5923*** 0.5773*** 0.8610*** 0.8488*** (10.7708) (10.9091) (10.8844) (10.4419) (27.4293) (25.2104) pgdp 0.0190*** 0.0196*** 0.0198*** 0.0207*** 0.0311*** 0.0316*** (5.5503) (5.7378) (2.7613) (2.9255) (9.8594) (10.0333) wage 0.0657*** 0.0658*** 0.0600*** 0.0591*** –0.0222 –0.0192 (3.5370) (3.5330) (2.9658) (2.9257) (–1.2119) (–1.0502) popd 4.8435* 5.4196** 6.5617* 7.5701** –7.0118** –7.1183** (1.9406) (2.1654) (1.8118) (2.0941) (–2.2746) (–2.3088) sfdi –0.0791 –0.0488 –0.0125 0.0482 0.0114 0.0280 (–0.7729) (–0.4748) (–0.0990) (0.3824) (0.1600) (0.3915) sind 0.0269 –0.0095 0.0406 –0.0181 0.0718 0.0779 (0.3261) (–0.1140) (0.3506) (–0.1568) (1.1304) (1.2259) inno 0.2824 0.2093 1.6516 1.3571 –0.3932*** –0.4005*** (1.2817) (0.9482) (1.3110) (1.0824) (–3.0411) (–3.0923) er 0.0314 0.0315 –0.0098 –0.0108 –0.0079 –0.0139 (1.0822) (1.0824) (–0.2175) (–0.2394) (–0.4663) (–0.8271) Observations 1014 1014 624 624 390 390 R-squared 0.7824 0.7815 0.8509 0.8514 0.9521 0.9521 Log-likelihood 1321.7205 1318.8161 847.8766 848.7807 964.8801 962.4305 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. ADBI Working Paper 1125 Chen, Huang, and Yu 35 Table C7: Estimation Results for Different Periods: Non-RB Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.5935*** –0.5674*** –0.7465*** –0.7676*** –0.7987*** –0.6532*** (–9.0244) (–8.6124) (–7.7610) (–7.8119) (–6.1041) (–5.2786) 𝜌𝜌2 0.2769*** 0.2245*** 0.1365* 0.0914 0.1653* 0.0890 (5.5221) (4.3644) (1.7268) (1.1046) (1.8295) (0.9934) pgdp 0.0124*** 0.0133*** 0.0254*** 0.0247*** 0.0045 0.0072 (5.7809) (6.2368) (7.2804) (6.9900) (0.7723) (1.2940) wage 0.1430*** 0.1358*** 0.1236*** 0.1259*** 0.0587 0.0535 (4.5402) (4.3150) (2.9570) (2.9829) (1.0820) (1.0316) popd 0.9962*** 0.9430*** 1.0326*** 0.9559** 0.5993 0.8721 (2.9174) (2.7649) (2.7216) (2.4946) (0.5518) (0.8431) sfdi 0.0226 0.0112 –0.0941 –0.0617 –0.0771 –0.0885 (0.5278) (0.2625) (–1.5306) (–0.9941) (–0.8005) (–0.9638) sind –0.2382*** –0.2751*** –0.1715 –0.1919 –0.3426 –0.2891 (–2.9299) (–3.3938) (–1.4218) (–1.5715) (–1.6372) (–1.4469) inno 0.0331*** 0.0308*** –0.0537 –0.0015 0.0300* 0.0354** (3.8488) (3.5714) (–1.3688) (–0.0374) (1.9115) (2.3578) er 0.0114 0.0090 0.0040 –0.0061 0.0285 0.0364 (0.5205) (0.4122) (0.1301) (–0.1941) (0.8010) (1.0738) Observations 1,469 1,469 904 904 565 565 R-squared 0.8367 0.8371 0.8507 0.8477 0.8961 0.9052 Log-likelihood 1,800.9122 1,806.4777 1,174.2957 1,170.7943 839.98095 862.7707 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. Table C8: Estimation Results for Different Periods: TCZ Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌1 –0.3134*** –0.3015*** –0.6606*** –0.6737*** –0.5534*** –0.4204*** (–6.4841) (–6.2484) (–9.3656) (–9.5476) (–5.5487) (–4.4072) 𝜌𝜌2 0.7155*** 0.6841*** 0.9683*** 1.0049*** 0.5549*** 0.4231*** (14.4347) (13.8075) (14.4208) (14.2387) (5.4453) (4.2604) pgdp 0.0143*** 0.0158*** 0.0270*** 0.0270*** 0.0032 0.0049 (7.8204) (8.7026) (10.0712) (9.9325) (0.7250) (1.1495) wage 0.1576*** 0.1554*** 0.0661 0.0705 0.0982** 0.0981** (4.8902) (4.8269) (1.5202) (1.6123) (2.1592) (2.2346) popd 1.1990*** 1.1897*** 0.9680*** 0.8806*** 0.4944 0.4847 (3.9130) (3.8877) (3.1896) (2.8857) (0.4663) (0.4753) sfdi 0.0255 0.0148 –0.2037*** –0.1754 –0.0581 –0.0467 (0.6623) (0.3851) (–4.1884) (–3.5958) (–0.7219) (–0.6009) sind –0.0184 –0.0302 0.0637 0.0733 –0.1379 –0.0186 (–0.2591) (–0.4251) (0.7196) (0.8223) (–0.6701) (–0.0935) inno 0.0339*** 0.0310*** –0.0700** –0.0056 0.0241* 0.0329** (4.4231) (4.0200) (–2.2520) (–0.1817) (1.7651) (2.4897) er 0.0132 0.0169 –0.0411 –0.0444* 0.0160 0.0229 (0.6601) (0.8454) (–1.5953) (–1.7092) (0.5173) (0.7665) Observations 1,651 1,651 1,016 1,016 635 635 R-squared 0.8564 0.8568 0.8975 0.8963 0.9157 0.9214 Log-likelihood 2,120.2997 2,123.2639 1,482.0985 1,476.1617 1,025.2875 1,043.9151 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1. ADBI Working Paper 1125 Chen, Huang, and Yu 36 Table C9: Estimation Results for Different Periods: Non-TCZ Cities Variable 2003–2015 2003–2010 2011–2015 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝒅𝒅 𝒅𝒅 � 𝜌𝜌 1 –0.1212* –0.1344** –0.1836** –0.1777** –0.4715*** –0.4696*** (–1.8194) (–1.9768) (–2.0413) (–1.9661) (–3.6012) (–3.5805) 𝜌𝜌 2 0.3757*** 0.3665*** 0.3367*** 0.3241*** 0.7047*** 0.6889*** (7.6619) (7.4942) (5.0823) (4.8882) (17.1239) (15.9910) pgdp 0.0230*** 0.0253*** 0.0254** 0.0295** 0.0369*** 0.0389*** (4.3663) (4.8779) (2.0224) (2.4222) (4.6346) (4.8580) wage 0.0490** 0.0426** 0.0617*** 0.0544** –0.0220 –0.0229 (2.5300) (2.1949) (2.7454) (2.4283) (–0.7739) (–0.7922) popd 0.1056 0.1964 –1.2364 –1.4920 –0.1334 –0.1068 (0.0474) (0.0880) (–0.2332) (–0.2833) (–0.1118) (–0.0884) sfdi –0.0429 –0.0568 0.1164 0.1342 –0.0224 –0.0503 (–0.3466) (–0.4587) (0.6466) (0.7493) (–0.1431) (–0.3185) sind –0.3094*** –0.3652*** –0.4959*** –0.5503*** –0.0817 –0.0667 (–3.3863) (–3.9940) (–3.2276) (–3.6176) (–0.9712) (–0.7840) inno 0.3312*** 0.2986*** 1.0508** 1.0658** 0.0060 –0.0192 (2.8784) (2.5921) (1.9927) (2.0320) (0.0497) (–0.1573) er –0.0055 –0.0100 0.0284 0.0192 0.0276 0.0253 (–0.1768) (–0.3212) (0.5836) (0.3966) (0.9724) (0.8837) Observations 832 832 512 512 320 320 R-squared 0.8083 0.8077 0.8250 0.8269 0.9524 0.9512 Log-likelihood 978.7677 978.8077 599.5602 601.8643 642.0050 637.9659 Notes: The tstatistics are given in parentheses. All regressions include spatial- and time period-fixed effects. Spatial lags of the control variables are not presented in order to save space. *** p < 0.01, ** p < 0.05, * p < 0.1.