The mechanism and effects of national smart city pilots in China on environmental pollution: empirical evidence based on a DID model

The policy of the national smart city (NSC) pilots, a new type of urbanization for future development, has been implemented in China in batches. This paper investigates the mechanism and effects of the NSC pilots on the environment. Using the prefecture-level panel data during 2004–2018 period, our multi-period difference-in-differences (DID) estimation shows that the NSC pilots causally mitigate SO2 (wastewater) pollution by 13.76% (14.36%), which is supported by a series of robustness tests. The mediating effect model indicates that green total factor productivity (GTFP) plays a partial mediating role in mitigating both SO2 and wastewater emissions, while technological innovation plays a partial mediating role in mitigating wastewater emissions. After introducing the two mediating channels into the multi-period DID model, the reduction effect for SO2 and wastewater emissions drops to 11.04% (1-e− 0.117) and 12.1% (1-e− 0.129), respectively. Comparatively, the NSC pilots contribute to the improvement of GTFP and GTFP takes the major mediating role in reducing SO2 and wastewater pollution. The heterogeneous effects of the NSC pilots show that the mitigation effect is more pronounced in cities with strong fiscal support, cities with abundant human capital, and cities with high R&D expenditure. Based on these novel findings, this study provides some policy implications for achieving better mitigation effects.


Introduction
Since the accession to the WTO, China has experienced unprecedented economic growth with an average annual growth rate of about 10% (Wang et al. 2021), accompanied by many environmental issues, including air pollution, water deterioration, massive carbon dioxide emissions (Chang 2015;He et al. 2020;Gong et al. 2021). During the process of industrialization and urbanization, China has witnessed severe "urban diseases" and suffered a lot from unreasonable spatial layout, traffic congestion, inefficient governance and environmental degradation (Liang et al. 2019;Wang et al. 2019;Yu 2021). Thus, it is essential Responsible Editor: Baojing Gu Zhijiu Yang live large@163.com 1 focusing on restricting manufacture (Lin and Du 2015). In addition, as illustrated in official guidelines for smart city construction, green and low-carbon development are two of the four requirements. In terms of efficiency improvement and policy priorities, the NSC pilots might have possible mitigation effects on environmental pollution.
The NSC pilots are closely associated with new-type urbanization in China, both of which highlight the intensive, intelligent, green, and low-carbon development in the process of urbanization. To some extent, the NSC pilots are not only pilots for smart city construction, but also pilots for new-type urbanization, which will be undoubtedly expanded to the rest of China. Hence, investigating the effects of the NSC pilots can be of great practical significance for developing countries, as they are on the path towards the second-stage urbanization with billions of people . The previous rapid urbanization, with an annual urbanization rate rising to 1.09% during 1980-2021, was characterized by population and industrial agglomeration (NBS 2021). This urbanization was investment-driven and fueled by high levels of energy combustion . Due to daily activities of urban residents and industrial production, cities have been long regarded as the main source of pollution (Lamsal et al. 2013), causing lots of environmental issues, like global warming (Han et al. 2018;Kocoglu et al. 2021), harmful pollution emissions Xu et al. 2020), deteriorated land and depleted resources (Kasman and Duman 2015). Low efficiency in energy and urban operation even exacerbated the conflict between economic growth and environmental protection (Yu 2021). As a result, the traditional urbanization exerted tremendous pressure on public health and ecological resources, making the development increasingly unsustainable (Gasana et al. 2012;Wu et al. 2020). On the contrary, the new-type urbanization is people-oriented and takes efficiency improvement, green and sustainable development as its core objectives. In this sense, the NSC pilots will contribute to promoting the coordination of social, economic, cultural, and environmental development during the process of urbanization, thereby influencing long-term sustainable economic development.
There are several problems that need to be examined and quantified. Firstly, can smart city construction in China mitigate environmental pollution? If so, what are the channels, and how significant are the impacts? Secondly, is there any difference in effects among these pilot cities? because China has a vast territory, smart city construction in different regions varies a lot. Exploring the heterogeneous effects of the NSC pilots on environmental pollution is of great significance for illustrating in which conditions the smart city construction performs well.
Since the NSC pilots emphasize the intensive, intelligent, green, and low-carbon development during the process of urbanization, they will improve the efficiency of urban operation and affect total factor productivity correspondingly. However, traditional TFP has limitations in evaluating how efficiently and intensively the inputs are utilized in production because it ignores the byproducts behind economic growth: energy combustion and environmental pollution (Easterly and Levine 2001;Comin 2010). Instead, green total factor productivity (GTFP), which takes into account environmental pollution as undesirable output, might be a better indicator measuring economic efficiency comprehensively. The NSC pilots integrate advanced technologies and management into urbanization and improve resource allocation efficiency, industrial upgrading, and optimization of spatial layouts, which are the keys to enhancing GTFP (Cheng and Jin 2020). For example, big data helps enterprises make better planning for production, cloud computing reduces corporate operation costs, and advanced management as well as internet infrastructure improve operating efficiency, resulting in mitigating environmental pollution eventually. As a result, this study argues that the NSC pilots might contribute to mitigating pollution emissions through enhancing GTFP. In addition, the NSC pilots also provide the R&D criteria for assessing the construction performance. In this sense, another influencing channel to mitigate pollution emissions might be through technological innovation.
To accurately quantify the effect of smart city construction, we should first know about the drivers of environmental pollution. There are two strands of literature related to our research. The first one is to explore concrete factors that might cause or alleviate environmental pollution. Existing studies investigate several causal factors from different perspectives that have environmental impacts, like industrial structure updating (Sun and Zhou 2016;Liu and Lin 2019), industrial agglomeration (Lu and Feng 2014;Zhang et al. 2019b), urbanization (Wang et al. 2016;Cui and Shi 2012;Li et al. 2012), technical process and innovation (Bosetti et al. 2011;Li and Zhao 2011;Xu et al. 2019;Shahbaz et al. 2020), environmental regulation (Levinson and Taylor 2008;Zhao et al. 2020;Song et al. 2020), international trade (Deng and Song 2008;Levinson 2009), etc. The literature provides meaningful perspectives for evaluating environmental pollution, which can assist in measuring the environmental effects.
The second strand of literature measures the policy effects of new-type urbanization and the NSC pilots. Some researchers investigate the mechanism of new-type urbanization and analyze the direct and indirect effects on energy consumption as well as carbon emissions (Liu et al. 2017;Wang et al. 2019). The result indicates that energy-saving technologies may have a rebound effect while environmental technologies can reduce carbon emissions. Yu (2021) finds that new-type urbanization has obviously positive ecological effects of "pollution reduction and efficiency improvement." As for the NSC pilots, Yang (2018) leverages the DID method to evaluate the impacts of the NSC pilots on economic efficiency, and finds that the NSC pilots lead to improving efficiency by 1%, compared to the non-pilot cities. Similarly, Liu and Lin (2019) show that the smart city construction policy significantly improves urban industrial ecological efficiency, increasing by about 5%-8%. Inspired by their work, this paper deems that the NSC pilots help improve the GTFP of pilot cities, thereby mitigating environmental pollution. Shi et al. (2018) evaluate the consequences of smart city construction in China under Schumpeter's innovation theory framework, illustrating the smart city contributes to the reduction of environmental pollution. But due to the limitations of the data range, their work measures the short-term effect and ignores the different batches of the NSC pilots. Based on the previous work, this paper regards the NSC pilots as the new stage of urban development and evaluates the effects on pollution from the perspective of GTFP and technological innovation.
We may contribute to the existing literature in the following points. First, we employ the multi-period DID model to quantify the long-term effects of the NSC pilots on environment. According to the official plan, each pilot city will have a 3-5 years construction period. Our sample covering the period could better reflect the construction effect. Our baseline results suggest that the NSC pilots do negatively affect environmental pollution and lead to 13.76% (1-e −0.148 ) and 14.36% (1-e −0.155 ) mitigation of SO 2 emissions and wastewater emissions respectively. In addition, the parallel trend test might reveal the time-varying effects, implying the long-term impacts on environmental pollution.
Second, we put forward and verify the mechanism that illustrates the possible channels of the NSC pilots on environmental pollution. Based on the planning and requirements of the NSC pilots, this paper explains some environmental consequences of the NSC pilots from a policy perspective, making a meaningful complement to the environmental effect of smart city construction. We highlight the role of GTFP in mitigating pollution emissions. Employing the mediating effect model, we find that the NSC pilots contribute to improving GTFP and technological innovation by 0.0906 and 13.56% respectively. The Sobel test statistics indicate that GTFP plays a partial mediating role in mitigating both SO 2 and wastewater emissions, while innovation has a partial mediating effect on water pollution. Comparatively, the mediating effect of GTFP is more salient, which will help deepen the understanding of the NSC pilots and enrich relevant literature. Third, we explore the heterogeneous effects of the NSC pilots on environmental pollution. The heterogeneous effects indicate that the mitigation effect is more pronounced in cities with agglomerated human capital, cities with strong fiscal support and cities with more R&D expenditure. Therefore, we suggest that governments should encourage the movement of population to cities, improve their education attainment and subsidize enterprises, which could lead to economies of scale and improve GTFP greatly. Because of the appreciable reduction of pollution emissions, governments should expand the scope of pilot cities and promote the development of new-type urbanization in China, integrating urbanization with advanced technologies and management.
The remainder of the paper is organized as follows: "Institutional background and hypotheses" generally describes the NSC pilots in China and proposes the impacting mechanism and some causal hypotheses. "Methodology and data" introduces the variables and our empirical strategy. "Empirical results" presents the empirical results of the NSC pilots on environmental pollution. "Heterogeneous effects of the NSC pilots" shows the heterogeneous effects of the NSC pilots. The last section concludes and provides some policy implications.

Policy background
With the rapid expansion of urbanization and industrialization, China experienced inefficient governance and suffered a lot from these so-called urban diseases (Chen et al. 2013;Xu et al. 2019). To abate "urban diseases" and promote the sustainable development of cities, the Ministry of Housing and Urban-Rural Development (MHURD) in China officially issued the NSC pilots' plan at the end of 2012. The construction period is 3-5 years in principle. During the construction period, local governments should submit the construction report concerning the construction progress and the performance of predetermined objectives to the MHURD before the end of December every year. For further monitoring the construction of smart city, four first-level assessment indicators are designed and published to the public, focusing on the aspects of network infrastructure, intelligent management and services, industrial upgrading, and new industries (MOHURD 2012). Thus, local governments have pressure and incentives to complete the evaluation, leading to the difference in policies and actions with non-pilot cites. The NSC pilots follow the principle of "experiment first, expand later," which is not only conducive to implementing construction plan according to local conditions, but also provides valuable experience for promoting new-type urbanization.
The first batch of NSC pilots covered 90 cities, including 37 prefecture-level cities, 50 counties (districts), and three towns. In these NSC pilots, China Development Bank collaborates with the local governments to invest in intelligent infrastructure, including cloud computing, 5G network, high-efficiency online platform, efficient organization etc., to realize the intelligence of the city management (Yang 2018). Then, in 2013, the government released the second batch of the NSC pilots list, which included 103 pilot cities, including 83 at the prefecture-level and 20 at the county and town level, further expanding the scale of the NSC pilots (MOHURD 2013). In 2015, the third batch of the NSC pilots list was released. And by this time, the number of the NSC pilots has reached 290, effectively promoting the development of China's new urbanization.
Generally, as demonstrated in official guidelines, smart city construction is closely related to the new-type urbanization in China, both of which have the requirements of intensive, intelligent, green, and low-carbon development. 1 Therefore, pilot cities have motives and will take measures to satisfy the green and low-carbon requirement, thereby reflecting on their environmental pollution levels. The main objectives of the NSC pilots are illustrated in the following aspects.
To integrate information resources and improve urban governance For a long time, the operating efficiency of cities and governance capability of governments are not admirable, imposing restrictions on business matters and individual mobility. And in the guidelines, smart city construction should focus on providing convenience and benefits to the people, promoting innovation in urban management and public services, and avoiding repeated construction. For achieving this purpose, governments are encouraged to improve network infrastructure and establish urban public information platform. These measures could enhance the intelligence of public infrastructure and make the operation precise. To support innovative activities and upgrade industry structure. The NSC pilots stress the importance of innovative activities and evaluate the construction performance based on the research and development (R&D) as well as innovative outputs, especially for emerging industries. Innovation and industry upgrading is an important dimension and the first-level 1 According to the Guidelines on Promoting the Healthy Development of Smart Cities, the guiding ideology of smart city construction is intensive, intelligent, green and low-carbon, consistent with the newtype urbanization in China (see smartcity.pdf). In addition, smart city construction should refer to National Plan for New Urbanization (2014)(2015)(2016)(2017)(2018)(2019)(2020). In this sense, the NSC pilots could be regarded as pilots for future new-type urbanization.
indicator for national smart city pilots. As a response, pilot cities tend to subsidize and encourage enterprises to invest more in R&D. To utilize urban planning and improve cities' functions. The assessment indicators emphasize the intelligence of cities from various perspectives, including transportation, energy, logistics, finance, and environmental protection. For intelligent construction, cities are required to formulate a complete and reasonable urban planning and a digital management. These measures make urban space and industrial layout reasonable and effectively promote the efficiency of cities. To make the environment more livable. The NSC pilots are people-oriented and emphasize sustainable developments, which have a priority concern over living environment. For achieving this purpose, the NSC pilots require an intelligent environmental monitoring system for water, air, noise, and soil. For obtaining better records, pilot cities are incentivized to prevent and control pollution emissions and energy consumption, thus improving the living environment in cities.
The NSC pilots' policy in China can be viewed as a quasi-natural experiment, where pilot cities are regarded as the treatment and non-pilot cities as the control group. In evaluating the policy effects, DID model is universally used in literature (Beck et al. 2010;Hu and Shi 2021). Since the NSC pilots are carried out in batches, we adopt the multi-period DID model to estimate the policy effect. Although cities have the right to apply for becoming the NSC pilots themselves, the MOHURD might select them based on their social and economic conditions. Therefore, to cope with possible selection biases, we conduct a series of robustness tests. The year-specific result will reveal the time-varying effects of the NSC pilots on the environment. Besides, to verify the impacting mechanism, we design the mediating effect model to examine whether the channels are significant empirically.

Related hypotheses
This study aims to quantify the impacts of the NSC pilots on environmental pollution in China. Based on the requirements and planning, the NSC pilots may have both direct and indirect impacts on environmental pollution.
First, China is undergoing a rapid process of urbanization. The NSC pilots can improve the level of urban services and attract more residents to the cities, thus affecting environmental pollution. Besides, to satisfy the assessments of the NSC pilots construction, local governments tend to focus more on environmental issues and have more motivations to introduce less-polluting industries into the local than non-pilot cities. And better performance of smart city construction might contribute to officials' political promotion. In addition, the development of emerging industries, usually environmentally friendly, is also an important evaluation indicator for the NSC pilots and governments are encouraged to facilitate them. These highly possible measures contribute to the reduction of pollution immediately. Thus, the NSC pilots may mitigate environmental pollution through the selection and differential treatments for enterprises directly. This paper proposes hypothesis 1 as follows: H1: the NSC pilots can reduce environmental pollution directly.
Second, the NSC pilots emphasize the coordinated development of population, society, industry and environment (Yu 2021). With the assistance of 5G, big data, and other information technologies, smart city construction can promote information communication and coordination within cities, reduce transaction costs and friction costs, and improve resource allocation efficiency. With the aid of Internet infrastructures and advanced management, governments will transform their serving concepts into more efficient ways (Xu et al. 2012). Meanwhile, the spatial layouts of industries are more optimal, avoiding unnecessary energy consumption and pollution emissions. Enterprises can continuously monitor the market demand and plan the best route with intelligent information technology (Shi et al. 2018). For example, the rise of ride-hailing platforms help improve energy efficiency and protect the environment by reducing the time idling. Xu et al. (2019) measure the carbon emission reduction by LaLaMove in 2016 with appropriate 100 million tons. These consequences, including improvements in resource allocation, promotion of industrial structure, efficient governance and optimal spatial layouts, contribute to improving GTFP (Cheng and Jin 2020). Hence, this paper argues that smart city construction could reduce environmental pollution emissions by improving GTFP.

H2: the NSC pilots can reduce pollution by improving GTFP.
Since GTFP is a comprehensive indicator evaluating the quality of economic growth, we also focus on the concrete channel from the perspective of innovation.According to the assessment indicators, pilot cities have extra money and concrete guidance for investing in research and development (R&D), which can improve the innovation capacity of cities in resource-saving and environment-friendly technologies (MOHURD 2012). In addition, for accomplishing the smart city construction plan, government may impose appropriate environmental regulation, which can motivate enterprises to improve their technological level and ability to control pollution (Shi et al. 2018). Thus, it reduces environmental pollution through technological innovation correspondingly.
H3: the NSC pilots can reduce pollution by promoting technological innovation. Based on the analysis above, the impacting mechanism is shown in Fig. 1. The two channels, including GTFP and technological innovation, might take the mediating role in reducing environmental pollution.

Empirical strategy
Based on the analysis, China had three batches of the national smart city pilots in 2012, 2013, and 2015 respectively. We use a multi-period difference-in-differences (DID) specification to measure the impacts of smart city construction on the environment. And the multi-period DID model is shown as follows: (1) where subscript i denotes ith city, and t is the time. P ollute it denotes the pollution of city i at time t, X it represents a set of control variables and θ denotes the corresponding coefficients. P ilot it implies the DID factor, equaling one in the years after city i chosen as the pilot city, otherwise zero. Due to the implementation of the NSC pilots concentrating in the second half of the year, the lagging form of DID variable is used in the regression. For example, Dongying city, located at Shandong province, China, was chosen as a NSC pilot in 2012. And the DID factor takes the value one when the year after 2012, otherwise zero. Besides, we include city-specific dummy variables to control unobserved time-invariant characteristics that might influence pollution emission, and μ i captures the city fixed effect. Since the central government is gradually paying attention to environmental issues, we include the time trend variable to capture time trend characteristics. For robustly estimating, standard errors are clustered at the city level.

Data source
To explore the effect of the NSC pilots on environmental pollution, we construct our sample data based on the China City Statistical Yearbook (2005-2019), namely data ranging from 2004 to 2018. This is primarily because the period is the latest and contains the whole NSC pilots, which could be employed to measure the longterm effects of the NSC pilots. The patents data within a city, denoting the innovation output, are collected from the Intellectual Property Office of the People's Republic of China. The inflation-adjusted indicator and the exchange rate, converting foreign investment denominated in US dollars into China Yuan, are obtained from the National Bureau of Statistics of China. For missing data of some indicators, the average growth rate is used to make up. Because of the adjustments of administrative jurisdiction and data unavailability, we eliminate the cities do not cover the whole period. All variables denoted by nominal value are converted into the constant price of year 2004. As for the NSC pilots, this study only considers the prefecture-level cities. These non-prefecture-level and provincial capital cities may influence the effects on the environment of the policy. Besides, samples for Tibet and Hainan provinces are dropped due to their unique locations and low-quality data. After being processed, there are 213 prefecture-level cities remained.

Dependent variables
Two leading indicators of pollution are adopted: industrial sulfur dioxide emissions and industrial wastewater emissions. Industrial production is the main contributor to environmental pollution (Zhang and Zhang 2019a). According to National Bureau of Statistics of China, industrial SO 2 emissions account for 29.9% of the total emissions of major air pollutants in 2017, imposing serious threats to public health and the environment. Owing to its harmfulness, governments in China have taken strict measures to control and prevent industrial SO 2 emissions and scholars usually investigates the drivers of SO 2 emissions as well as how to mitigate them Yuan et al. 2020). As for wastewater, academics usually take industrial wastewater emissions to analyze water pollution.

Green total factor productivity (GTFP)
Although China has achieved unprecedent economic growth, it also has brought severe environmental problems (Chang 2015;Tang et al. 2016). It is important and necessary to consider undesirable or detrimental outputs, such as pollutants or hazardous wastes, as byproducts for economic growth (Álvarez et al. 2016). Hence, this study employs a Malmquist-Luenberger (ML) productivity measure considering undesirable output to assess economic efficiency comprehensively, namely green total factor productivity (GTFP). By contrast, traditional total factor productivity (TFP) ignores the environmental pollution behind economic growth and therefore, cannot make an accurate and comprehensive evaluation of economic performance, especially for development quality (Cheng and Jin 2020). In our analysis, the NSC pilots focus on the intensive, intelligent, green and low-carbon development and therefore contribute to the improvement of long-term sustainability of the economy, improving cities' GTFP.
Each decision-making unit (DMU) refers to the city respectively, and input of each city is expressed as X Assume the production possibility set P (x) satisfies the closed and convex features. Following constant returns to scale production, the possibility set is denoted as follows: Let g=(g d , g u ) be a direction vector. The efficiency corresponds to the solution of the following program: This function seeks the maximal increase of desirable outputs while simultaneously reducing undesirable outputs (Oh 2010). The ML index of a city i is calculated based on two consecutive benchmark technologies, as follows: where s = t, t + 1. If ML > 1, efficiency increases and the DMU is capable of producing more desirable output with less undesirable production. And ML < 1 indicates the opposite. Since ML t = ML t+1 , scholars often use the geometric mean of the ML index between two consecutive periods, illustrated by: We employ the data envelopment analysis toolbox for MATLAB, developed byÁlvarez et al. (2016), to calculate GTFP. As the requirement of ML, the calculation process needs input variables, desirable output variables, and undesirable output variables. Considering that labor and capital are essential for production from macro-economic perspective, these relevant indicators are explained as follows: (1) Labor input: As labor is a vital element for production, this study uses the number of employed persons of each city to denote labor input.
(2) Capital input: Since the capital given in the national statistics is a flow variable, following Chen and Tang (2018), we use the equation K t+1 = (1 − δ)K t + I t+1 to transform the flow variable into capital stock.
(3) Electricity input: Along with the industrialization process of China, electricity plays increasingly important roles in promoting economic growth. This study employs the total electricity consumption to denote electricity input. (4) Desirable output: In this study, GDP of each city, adjusted to the base year 2004, is adopted as the desirable output. (5) Undesirable outputs: As previous literature indicates, environmental pollution is important byproduct of economic growth. Therefore, in view of the availability of data, this paper uses industrial SO 2 emissions and industrial wastewater to represent undesirable outputs.

Technological innovation
Patents are the important output of innovative activities. The strength of technological innovation can be reflected in the number of patents. According to the property of patents, they can be divided into invention patents, utility patents and design patents. Among these types of patents, invention patents could contribute to the mitigation of environmental pollution, while the other two do not influence pollution emissions basically. The more the invention patents are, the stronger the innovation (Shi et al. 2018;Liu and Lin 2019). Hence, we use the invention patents of each city as a proxy variable of technological innovation, also adopted in the logarithmic form. On the other hand, it takes some time when patents are licensed to apply. Thus, the lagging form of patents is used in the following regressions.

Control variables
To ensure the effect on environmental pollution is not driven by other factors, we include a set of prefecturelevel variables to control their potential impacts on environmental pollution. In line with the extant literature, these variables include: (1) Economic development (Pgdp), defined as the GDP per capita in each city. Additionally, the quadratic form is adopted.
(2) Economic growth (Growth), characterized by the year-on-year growth in GDP. Economic growth, especially rapid growth, is highly correlated with environmental pollution (Wang et al. 2021). (3) Total population (Pop), defined as the natural logarithm of total population. The agglomeration of population affects production and life style, further influencing pollution emissions (Cui and Shi 2012). (4) Industrial structure (Ind), defined as the proportion of the second industry to all industries. Owing to the difference in energy combustion, industrial structure has a significant influence on environmental pollution (Liu and Lin 2019). (5) Urbanization (Urb), characterized by the proportion of the non-farm payrolls to the total employment. Generally, scholars usually employ the proportion of non-agricultural population to the total population (Panayotou 1997;Shi et al. 2018). However, the integrity and quality of the prefecture-level non-agricultural population data are not applicable because of missing many values. Hence, following Sun and Zhou (2015), we take an alternative indicator, the ratio of non-farm payrolls, to denote urbanization rate. (6) Foreign direct investment (Open), denoted by the amount of foreign investment. According to "pollution heaven" and "pollution halo" hypotheses, foreign investment might have both positive and negative effects on environmental pollution (Fredriksson et al. 2003;Balsalobre-Lorente et al. 2019). (7) Density (Den), defined as the ratio of population to area. For mitigating the effect of outliers, we winsorize all continuous variables at the 1st and

Mediating effect model
In addition, we design the mediating effect model to verify the mechanism of the NSC pilots on pollution, which can be interpreted as a two-stage model (Hill et al. 2010). At the first stage, we should examine whether the NSC pilots have impacts on GTFP and technological innovation. That is: And at the second stage, we introduce GTFP and technological innovation variables into the multi-period DID model and examine whether it is still statistically significant. The model is shown as follows: Suppose the coefficient before the DID factor is significant in the first stage. In that case, the result indicates that the NSC pilots contribute to improving GTFP and technological innovation. At the second stage, the coefficient before the DID factor is still significant, and the coefficients before the GTFP and technological innovation are also significant, indicating that these two channels play a partial mediating role in the environmental pollution reduction preliminarily. If the coefficient of DID factor becomes insignificant, while the two channels are significant, it indicates that GTFP and technological innovation play a complete mediating role (see Baron and  Z 1 =β 11 ·β 2 / β 2 11 · Var β 11 +β 2 2 · Var β 2 (5) Z 2 =β 21 ·β 3 / β 2 21 · Var β 21 +β 2 3 · Var β 3 Besides, the form of Aroian test and Goodman test is similar to the Sobel test, while the two have an additional cross term in the denominator. We employ the three tests to further verify the mediating role of GTFP and technological innovation.

Baseline results
Using the multi-period DID model, we examine the effect of the NSC pilots on environmental pollution, as shown in Table 2. We are concerned about the estimated coefficient before the variable Pilot. Column (1)-(2) show the NSC pilots contribute to mitigating SO 2 emissions whether includes the covariates or not. In detail, after controlling the covariates, the construction of smart city is associated with the mitigation in SO 2 emissions by 13.76% (1e −0.148 ) comparatively, which is significant at the 5% significance level. Similarly, Column (3)-(4) indicate the NSC pilots negatively affect industrial wastewater emissions. On average, the NSC pilots lead to a 14.36% (1-e −0.155 ) reduction in wastewater emissions. The baseline model illustrates that the NSC pilots significantly mitigate environmental pollution regarding SO 2 and wastewater emissions. Besides, the coefficients before the GDP per capita and its quadratic form show the inverted U-shape  (Kaika and Zervas 2013). Due to the lack of capital and technology in the early stages of development, China boosted its economy at the expense of energy combustion and environmental pollution. And now China is moving towards a stage of high-quality development, which is expected to limit environmental degradation eventually, presenting a pattern of "grow first, clean up later" like other developed countries. Undoubtedly, the NSC pilots will contribute to the green and high-quality economic development.

Parallel trend test
One premise before employing the multi-period DID model is that the control and treatment group should have same trends before the implementation of the NSC pilots. Following Beck et al. (2010), this paper employs the parallel trend test to examine whether there are differences before  Standard errors reported in the parentheses are clustered at the city level. ***, **, and * indicate the result is significant at the 1%, 5%, and 10% significance level respectively. Column (1)- (2) show the result eliminating cities in the ETS pilot provinces. Column (3)-(4) show the result eliminating cities highly influenced by the Action Plan. Column (5)- (6) show the result eliminating cities influenced by the ETS or the Action Plan and after the pilot to verify the robustness. We set different dummies relative to the start of the NSC pilots. The parallel trend test is designed as follows: where D it is the year-specific variable, taking a value of zero, except as follows: D −j equals one for cities in the j th year before entering the NSC pilots' list, while D +j equals one for cities in the j th year after entering the pilot list. At the endpoints, D −6 it equals one for all years that are six or more years before entering the list, while D +4 it equals one for all years that are four or more years after entering the list. After de-trending and centering the estimates on the pilot year, Fig. 2 plots the results and the 95% confidence intervals, which are adjusted for city-level clustering. If the parallel trend hypothesis holds, the coefficients of β −i will be not significantly different from zero or significantly opposite to the baseline coefficient (Tanaka 2015). Figure 2 illustrates one crucial point: the impacts are close to zero before the pilots and fall quickly after the pilots. The coefficients of the dummy variables are insignificantly different from zero for all years before the pilots, implying little difference exists between the control and treatment group before the pilot policy. Next, it is clearly noted that SO 2 and wastewater pollution emissions fall immediately after the pilots, reaching the lowest point in the fourth year. The parallel trend test shows smart city construction might have a long-term impact on SO 2 and wastewater emissions, strengthening with the construction period. Therefore, the parallel trend assumption is satisfied, indicating our baseline results are robust.

PSM-DID tests
Although the NSC pilots are based on the self-application of cities, the MOHURD might select cities with wellperformed economic and environmental conditions to conduct smart city construction firstly, causing some bias problems. To cope with possible selection biases and control for unobservable group differences, we use propensity score matching and difference-in-differences (PSM-DID) to examine the robustness of our baseline results. Following Bu and Shi (2021), we use the city-level covariates as characteristic variables to find the control group for the treatment group, and then repeat the multi-period DID estimation. Table 3 presents the estimated result using the PSM-DID method, still indicating the negative effects of the NSC pilots on environmental pollution.

Excluding other policies
Besides, the accuracy and statistical significance of the empirical results might be challenged by other policies that have environmental effects. By checking other concurrent policy disruptions, we find that the SO 2 emissions trading scheme (ETS) and the Air Pollution Prevention and Control Action Plan (Action Plan) might influence our results. 2 Hence, we eliminate these highly possible influenced cities and use subsamples to run the regression. Column (1)-(2) of Table 4 show the estimated result after deleting pilot cities of the ETS. Column (3)-(4) indicate the result after dropping cities located in the Beijing-Tianjin-Hebei region, the Yangtze River Delta and the Pearl River Delta, of which were highly regulated by the Action Plan. And Column (5)-(6) present the result eliminating all possible influenced cities by the two policies. We find that the NSC pilots still have negative effects on environmental pollution even excluding other relevant policies, further verifying the robustness of our results.

Placebo test
The placebo test in DID can further promote the robustness of our results (Hu and Shi 2021). In terms of multi-period DID model, we randomly select 91 pilot cities (the amounts of pilots in our sample) with smart city construction firstly and then select the pilot years randomly as the "virtual treatment." Using the baseline model as the regression form, the process is repeated for 500 times. Intuitively, we compare the coefficients before the DID variable between the "actual treatment" and the "virtual treatment." Figure 3 shows the distribution of estimated coefficients, along with the red dashed line denoting the actual coefficient in Table 2. The result illustrates the estimates from the "virtual treatment" are centered around zero in both SO 2 and wastewater emissions with the mean value −0.0020736 and −0.002231 respectively, while the actual estimate, beyond the 95% of the 500 estimates, stays away from zero, indicating the reduction effect on environmental pollution. These findings show that the negative effects of the NSC pilots on environmental pollution are robust.

Mechanism test
According to the theoretical analysis above, smart city construction may mitigate environmental pollution through two channels: improving green total factor productivity and accelerating technological innovation. In order to quantify the role of each channel, this paper uses the mediating effect model to analyze the corresponding mechanism empirically (Wang et al. 2021).
The first stage regression is shown in Table 5. As can be seen from the table, smart city construction does improve GTFP of pilot cities. On average, the green total factor productivity of the NSC pilots is 0.0906 higher than that of the non-pilot city, which is significant at the 1% significance level. The result shows that with the aid of information technology and network platform, smart cities change the management and operation mode of cities and improve the efficiency of resource allocation. Similarly, smart city construction promotes the innovation capacity of the city. The number of patents in the pilot city increases by 15.68% (e 0.1457 -1), which is statistically significant at the 10% significance level but not as significant as GTFP. The reason behind this finding is that smart city construction mainly integrates advanced technology and management into urban operation and promotes the coordination of social, economic, cultural, and environmental development. In this sense, the Standard errors reported in the parentheses are clustered at the city level. ***, **, and * indicate the result is significant at the 1%, 5%, and 10% significance level respectively NSC pilots contribute more to the improvement of GTFP. Therefore, this paper is cautious about the mediating role of technological innovation and will conduct additional test to further verify its effect. Overall, the first stage regression suggests that the NSC pilots can help improve GTFP and accelerate technological innovation in pilot cities.
In order to verify that smart city construction can reduce pollution by improving GTFP and promoting technological innovation, these two variables are introduced into the baseline model. The result is shown in Table 6.
For SO 2 emissions, the NSC pilots help mitigate pollution mainly by improving GTFP, which is statistically significant at the 1% significance level. The technological innovation variable is also negatively correlated with SO 2 . After introducing the two variables into the baseline model, the effect of the NSC pilots on SO 2 pollution decreases from 13.76% (1-e −0.148 ) to 11.04% (1-e −0.117 ). Since the coefficient before the pilot variable is still significant at the 10% significance level, we can draw the preliminary conclusion that green total factor productivity and technological innovation play a partial mediating role in mitigating SO 2 pollution emissions, contributing to 19.77% of the mitigation overall. The drivers of SO 2 emissions come from the combustion of fossil fuels. The NSC pilots lead to the improvement of resource allocation and reduce energy consumption comparatively. Thus, improving GTFP plays a key role in mitigating SO 2 emissions.
As for wastewater pollution, in the second-stage regression, improving GTFP and promoting technological innova- Standard errors reported in the parentheses are clustered at the city level. ***, **, and * indicate the result is significant at the 1%, 5%, and 10% significance level respectively tion lead to the mitigation of wastewater emissions, which is significant at the 1% significance level. When introducing the two variables into the baseline model, the direct effect of smart city pilots on water pollution decreases from 14.36% (1-e −0.155 ) to 12.1% (1-e −0.129 ). And the coefficient before the pilot variable is significant at the 5% significance level. The result also indicates that GTFP and technological innovation have a partial mediating effect on water pollution preliminarily, contributing to 15.73% of the mitigation. Since the mediating role of technological innovation should be argued, we conduct the Sobel test to further verify the mediating role of GTFP and innovation. The statistics calculated are shown in Table 7.
The test statistics indicate that GTFP plays a partial mediating role for smart city construction mitigating SO 2 and wastewater emissions, while technological innovation plays a partial mediating role in mitigating wastewater emissions rather than SO 2 emissions. In conclusion, the result of the mediating effect model could support hypothesis 1, hypothesis 2, and hypothesis 3 as a whole. The NSC pilots can mitigate environmental pollution directly by choosing low-polluting enterprises and giving differential treatments regarding fiscal policies. Besides, the NSC pilots help improve green total factor productivity and accelerate innovation. Among them, GTFP has significant impacts on both SO 2 and wastewater emissions, while technological innovation plays a partial mediating role in wastewater emissions. Comparatively, the mediating effect of GTFP is more pronounced and the NSC pilots contribute to the improvement of GTFP.

Heterogeneous effects of the NSC pilots
Due to different economic bases, social cultures, and environmental backgrounds, the impacts of the NSC pilots  Standard errors reported in the parentheses are clustered at the city level. ***, **, and * indicate the result is significant at the 1%, 5%, and 10% significance level respectively may differ among these pilot cities. Previous findings could not capture the difference. In order to analyze the impact of smart city construction of different cities, this paper mainly classifies cities according to their size, geographical locations and city characteristics, like human capital and fiscal support. And taking SO 2 emissions as an object, we analyze the impact of the NSC pilots on pollution emissions.

City size and location
Intuitively, large cities tend to have an agglomeration effect and higher efficiency of resource allocation (Lu and Feng 2014). For example, the popularity of online carhailing is higher in large cities, contributing to bringing more efficiency improvement and reducing environmental pollution. According to the Adjusting Standards for Classifying Cities issued by the State Council, cities with an urban population of more than 3 million, between 1 million and 3 million, and less than 1 million are classified as large, medium, and small cities respectively. Column (1)-(3) of Table 8 shows the mitigation effect for SO 2 emissions of smart city construction performs well in large cities, while for small cities the reduction effect is not significant. This might indicate the fact that big cities have advantages in alleviating environmental pollution through the agglomeration of population and economic activities (Dong et al. 2020). Generally, big cities tend to exhibit broad markets and therefore, they might benefit from economies of scale greatly. Besides, according to their geographical locations, we divide cities into three parts: east

Fiscal support, human capital, and R&D expenditure
Since fiscal support, human capital, and R&D expenditure are essential for productivity and efficiency, we are concerned about the heterogeneous effect of the NSC pilots on SO 2 emissions as well as the indirect effect of GTFP.
According to the size of human capital (measured by the number of employees), fiscal support (measured by fiscal expenditure) and R&D expenditure, the prefecturelevel cities are divided into three equal parts. Among them, we consider cities with highest and lowest indicators respectively. Column (1)-(2) of Table 9 indicate that the mitigation effect on SO 2 emissions is much remarkable in cities with higher fiscal support, enabling them to subsidize enterprises to adopt advanced management and high-tech applications. Hence, the mediating role of GTFP is more significant in cities with strong fiscal support. Similarly, Column (3)-(4) shows that cities with abundant human capital have more salient mitigation effect on pollution. The mediating effect of GTFP is more pronounced, implying the agglomeration of talents could greatly improve economic efficiency. Besides, Column (5)-(6) illustrate that R&D expenditure also contributes to the improvement of GTFP and thereby help mitigate environmental pollution. All in all, the mitigation effect of the NSC pilots on SO 2 emissions is more pronounced in cities with strong fiscal support, cities with abundant human capital and cities with more R&D expenditure. The effects of GTFP are also pronounced. In addition, the parallel trend test indicates the NSC pilots might have a long-term mitigation effect on environmental pollution. As smart city construction progresses, the pollution reduction effects are gradually increasing. Hence, mitigating pollution cannot happen overnight, which requires long-term planning and construction. And the policy of the NSC pilots in China is an excellent attempt to alleviate environmental pollution systematically.

Conclusions and implications
Besides, further analysis reveals the underlying mechanism through which mitigates environmental pollution, specifically, green total factor productivity drives both SO 2 and wastewater mitigation, while technological innovation only drives wastewater mitigation. In detail, the first stage regression shows that the NSC pilots have significantly positive impacts on improving green total factor productivity and promoting technological innovation, while the latter is not significant as the former. On average, the NSC policy helps improve GTFP by 0.0906 and increase the invention patents by 13.56% (e 0.1457 -1). However, combined with Sobel test, the second stage regression indicates the partial mediating effect of GTFP on both SO 2 emissions and wastewater emissions mitigation, while technological innovation only takes a mediating role in mitigating wastewater emissions. Hence, the NSC pilots mitigate environmental pollution mainly through improving GTFP rather than innovation. As for the heterogeneous effect of the NSC pilots on environmental pollution, we find that the mitigation effect is more pronounced in big cities, cities with abundant human capital, cities with strong fiscal support and cities with more R&D expenditure, indicating the agglomeration of human capital and R&D expenditure are important for cities to control their pollution emissions. Meanwhile, the mediating effect of GTFP is more salient, contributing to mitigating environmental pollution.
Based on these novel findings, this study provides some important policy implications. First, governments should encourage the movement of people to cities and improve their education attainment, as the agglomeration of human capital could lead to economies of scale and improve green total factor greatly. China has some unreasonable restrictions on population and urban sprawl, including household registration system and construction land restriction, which is not conducive to urban efficiency. There is room for removing institutional barriers and making cities more intelligent. And the concentration of human capital could promote the smooth implementation of relevant policies and exhibit well-performed mitigation effect. Second, fiscal subsidies and other incentives are essential and important for enterprises to conduct innovative activities and promote their operating efficiency. Governments could use fiscal tools to guide the direction of corporate innovation and development, like clean energy substitution, end-of-pipe treatment and improving operating efficiency. Additionally, governments should create a relaxed atmosphere for innovation and encourage new-born technologies or industries, which are usually environmentally friendly. Third, due to the appreciable reduction effect on pollution of the NSC pilots, governments should expand the scope of pilot cities and promote the development of new-type urbanization in China. As evidence indicates, smart city construction can effectively mitigate environmental pollution at a low expense. Compared with restricting industrial production, improving the operating efficiency of cities could deal with the dilemma between economic growth and ecological protection (Lin and Du 2015). Therefore, the central government should apply the smart city experience to the rest of China, changing their governance mode with advanced technology, management and visionary planning. Besides, the implementation of the NSC pilots requires cooperation from the public. Fostering environmental awareness among citizens and businesses is also an important dimension.
Author contribution GX and ZY developed the conceptual framework, ZY prepared and processed the data. GX and ZY wrote part of the manuscript. GX supervised the final manuscript. All authors read and approved the final manuscript.
Funding This work was financially supported by Shanghai University of Finance and Economics (CXJJ-2019-415).

Data availability
The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

Consent for Publication Not applicable
Competing interests The authors declare no competing interests.