Does technology innovation reduce haze pollution? An empirical study based on urban innovation index in China

Haze pollution is one of the most concerning environmental issues, and controlling haze pollution without affecting economic development is of immense significance. Using the panel data composed of PM2.5 concentration and other data from 278 cities in China between 2003 and 2016, this paper empirically investigates the impact of urban innovation on haze pollution and its transmission mechanism. Based on the fixed effect model, the research finds that increasing urban innovation significantly reduces haze pollution. Even after dealing with possible endogenous problems, the result still holds. Energy consumption and industrial agglomeration are two important transmission channels through which urban innovation affects haze pollution. Furthermore, time heterogeneity analysis shows that the negative effect of urban innovation on haze pollution increases with time. Spatial heterogeneity analysis shows that urban innovation has a more significant mitigation effect on haze pollution in eastern cities than in central and western cities in China. This paper indicates that technological innovation, as the main driving force for development, can provide vital support to China to improve the ecological environment.


Introduction
Haze pollution is a phenomenon of air pollution that occurs all over the world. As a common "urban disease," haze not only affects economic growth but also endangers human health. In recent years, large-scale fog and hazy weather have frequently appeared in Chinese cities. Although China has experienced the miracle of rapid economic growth for decades, the factories that led to this growth are emitting a large number of pollutants and construction sites are generating an immense amount of dust. According to the most recent statistics from the Ministry of Ecology and Environment of China, about one-third of the cities in China have PM 2.5 concentrations that do not meet the national secondlevel standard, causing a severely polluted atmosphere in 2020. China is committed to transforming the economic development pattern to environment-friendly development, and both government and society attach great importance to air pollution and other environmental problems.
The fundamental requirements for improving air quality are reducing the frequency of haze problems by setting pollutant discharge standards, enhancing environmental regulations, and, more importantly, addressing the root causes of air pollution. One of the pollution prevention approaches to reduce air pollution at its source is to enhance energy and resource efficiency through technological innovation to reduce pollutant emissions. Innovation is now a city-based phenomenon (2thinknow, 2006). 1 Cities can gather research and development (R&D) resources and form scale effects. Everywhere in the Western world, we can see the rise of cities calling themselves "innovative cities" (Hospers, 2008). Since 2008, China has gradually promoted the construction of "innovative cities" and proposed to improve the urban innovation level. It is of great significance to study whether urban agglomeration of innovation activities affects haze pollution.
In the context of balancing environmental pollution and economic growth, this paper applies the fixed effect model to examine the effect of urban innovation on haze pollution in China based on the data of prefecture-level cities. The regression results suggest that if the urban innovation level increases by 1%, the PM 2.5 concentration will decrease by 1.030. It also attempts to address the transmission mechanisms of urban innovation on PM 2.5 concentration and the heterogeneity of the effect of different periods and geographical regions. We also tested robustness through different models to check whether the effect is robust in China.
This paper contributes to the literature in two aspects. First, our paper contributes to a strand of literature that connects innovation and the environment. Given the availability of data, most of the current studies have used R&D expenditure and patents granted to measure the level of innovation and have looked mainly at the impact of technological progress within enterprises and provinces on pollution. This paper uses the urban innovation index that considers the market value of patents to measure the level of innovation and makes the first attempt to analyze the impact in the framework of the city. Second, a growing body of papers has documented the role of technological innovation in carbon emissions and sulfur dioxide emissions. However, the effect of innovation on haze pollution and the heterogeneity of different periods and geographical regions have not yet received sufficient attention. Only Liu (2018) has evaluated the link between provinces' innovation and PM 10 . Compared with PM 10 , PM 2.5 is smaller in size and richer in many toxic and harmful substances that are more harmful to human health and air environment quality. For these reasons, the present paper studies the effect of urban innovation on PM 2.5 concentration and takes other important components of haze pollution-sulfur dioxide and nitrogen dioxide-into consideration to study how they are affected by urban innovation. This paper also adds to the existing literature in performing a robust estimation check. We explore the interactive fixed effect model to take the multidimensional time shock into account and use the dynamic panel model to control the effect of a previous period. Moreover, we rely on data covering a wide city area and with a long observation period in China. We exploit the non-point source data of PM 2.5 concentration measured by satellites observed by Columbia University's NASA Social Economic Data and Application Center. In comparison, the lack of long-term monitoring data in developing countries where air quality monitors are sparse and have only been established in several cities has affected the accuracy of identification of other related studies. Our estimates of the effect are more reliable and contribute to current policy discussions on haze governance. The paper is organized as follows. "Literature review and hypotheses" describes the literature review and hypotheses. "Model and data" presents the model and data, and "Empirical results and discussion" reports empirical results and discussion. Finally, "Conclusions and policy implications" provides conclusions and policy implications.

Literature review and hypotheses
Researchers have always been concerned about the impact of technological innovation on environmental pollution. The IPAT model proposed by Ehrlich and Holdren (1971) suggested that technological advances can alleviate environmental pollution caused by population growth. Grossman and Krueger (1991) argued that economic growth means the continuous development of high technology, which reduces environmental pollution. The methods and models provided by these two papers are still widely used today, and many studies have confirmed their research conclusions. For instance, Prakash and Potoski (2006) and Bhupendra and Sangle (2015) indicated that innovative capabilities have promoted the successful implementation of pollution prevention, pollution control, and clean technology strategies. Johnstone et al. (2017) and Valentin and Elena (2020) found that the advancement of environmentally sound technology in production is conducive to reducing the discharge of pollutants and improving the efficiency of pollution control, thereby helping to suppress environmental pollution. Based on the above theoretical analyses, much empirical research has been conducted over a growing body of literature on the impact of innovation and pollution.
Some empirical studies explore the relationship between innovation and pollution from the perspective of technological advancement and pollution emissions at the micro-level. Levinson (2009) studied data from the US Environmental Protection Agency to show that the overall pollution reduction of the US manufacturing industry comes mainly from changes in technology. Baniak and Dubina (2012) believed that domestic, independent technology innovation, foreign technology import, and domestic technology transfer improved the eco-efficiency of industrial enterprises. Wan et al. (2015) focused on the industrial enterprises of China and confirmed the positive role of the three modes of technological innovation regarding the environment. Zhang et al. (2019) proposed an index system to calculate technological innovation efficiency. They verified that technological innovation is conducive to improving the capacities of industrial enterprises to deal with local environmental pollutant emissions, thereby reducing the environmental pollution. Ge (2019) also believed that enterprise technology innovation is conducive to reducing pollution emissions, including wastewater, waste gas, and solid waste. Xu et al. (2020) used the panel data of 28 sub-sectors of China's manufacturing industry from 2011 to 2017. They found that innovation capabilities have a positive effect on the suppression of environmental pollution.
Past research has also examined associations between national technological advancement and pollution emissions from a macro-level. Dinda (2018) researched the USA's technological progress and believed that technological progress is the central force that causes a reduction in CO 2 emissions. Mensah et al. (2018) found that technological advancement in OECD (Organization for Economic Co-operation and Development) countries plays a key role in mitigating CO 2 emissions. Ibrahiem (2020) and Nguyen et al. (2020) also reached similar conclusions in Egypt and 13 selected G-20 countries. In addition, researchers have also studied the relationship between innovation and environmental pollution at the provincial level. Wang and Xie (2014) used the total R&D expenditure of large and medium-sized enterprises at the provincial level in China to evaluate the technological innovation level. They found that technological innovation is beneficial to reducing SO 2 emissions. Liu (2018) used a similar approach to measure the province's innovation level in China and found that technological innovation can reduce annual average concentrations of PM 10 . Ma et al. (2020) used the patents granted to measure the innovation level of the province and discovered that technological innovation could reduce pollution between 0.167% and 0.415% under different intensities of water pollution. Wu (2020) used a similar method to measure the technological innovation level in China's provinces and reached similar conclusions.
Although the above research generally believes that technological innovation reduces pollution, a few studies hold different views. For example, Acemoglu et al. (2012) gave theoretical evidence of endogenous technological progress and suggested that new technologies can be divided into clean and pollution-based categories. Therefore, different directions of technological progress led to different effects on environmental pollution. Giuliani (2018) found that innovation-induced industrial activities have had important negative consequences for the environment. Demir et al. (2019) discovered that the relationship between CO 2 emission level and the number of domestic patents depicts an inverted U-shape curve for Turkey. Given that most theoretical and empirical studies believe that technological innovation is conducive to reducing environmental pollution from micro and macro perspectives, hypothesis 1 is proposed.

Hypothesis 1: Urban innovation has a negative effect on haze pollution
One of the pathways through which urban innovation affects haze pollution is energy consumption. On the one hand, an increase in energy efficiency and a reduction in energy consumption will directly reduce pollutant emissions, thereby reducing haze pollution. Researchers have reached a consensus on the negative effects of energy consumption on environmental quality. In terms of energy consumption, Apergis and Payne (2014) used cross-continental data to undertake their research, finding a significant correlation between energy consumption and environmental quality. Hafeez et al. (2019) believed that energy consumption is one of the main determinants of environmental degradation. In terms of energy structure, Yang et al. (2018) considered that the improvement of energy structure had made a significant contribution to improving environmental quality. On the other hand, innovation is a driving force for improving energy efficiency and reducing energy consumption. For instance, Cagno et al. (2015) took Italian foundry companies as the research object, and Subrahmanya and Kumar (2011) took small and medium-sized enterprises in the Indian machine tool industry as the research object; they concluded that technological innovation activities had promoted the improvement in energy efficiency. Studies by Ramirez-Portilla et al. (2014), Herrerias et al. (2016), and Zeng and Li (2020) also confirmed the important role of innovation in improving energy efficiency and reducing energy consumption. Therefore, we expect that urban innovation reduces haze pollution by improving energy efficiency and reducing energy consumption. Based on this, hypothesis 2 is obtained.

Hypothesis 2: Urban innovation can improve energy conversion efficiency and reduce energy consumption, thus reducing haze pollution
Another important channel through which urban innovation affects haze pollution is industrial agglomeration. On the one hand, the discharge and treatment of pollutants have the characteristics of economies of scale (Lu and Feng, 2014), and agglomeration of economic activities is found to reduce environmental pollution. Daddi et al. (2017) provided evidence that improving environmental pollution reduction efficiency could be achieved through cooperation and infrastructure sharing between enterprises. From the perspective of the positive externalities of industrial agglomeration, Porter (1998), Chertow et al. (2008), and Hosoe and Naito (2006) believed that the technology spillover effect of industrial agglomeration could promote the emergence and development of environmentally related industry clusters and positively affect the spread of clean technology. Copeland and Taylor (1994) also confirmed that the scale effect brought by industrial agglomeration can increase the return to scale of pollution control technologies across the whole industry, thus improving the environmental quality.
On the other hand, innovation is conducive to integrating ingredients and is an important force to promote industrial agglomeration. Marshall (1890) proposed that inter-firm technological spillovers can promote the spatial agglomeration of manufacturing production. Researchers have reached similar conclusions in other studies. For instance, Forman et al. (2016) found that technological innovation is important for inducing industrial space agglomeration. Sultan and Dijk (2017) believed that innovation is necessary to foster the development of industrial clusters. Chung and Alcácer (2002), Guastella and Van Oort (2015), and Goldman et al. (2016) argued that regional space agglomeration of innovation is an important source for industrial agglomeration. Based on the above analysis, the following hypothesis is obtained.

Hypothesis 3: Industrial agglomeration plays a mediating role between urban innovation and haze pollution
In summary, the past evidence on the effect of the region's innovation on pollution focuses primarily on national or provincial innovation. Although innovation is now a city-based phenomenon (2thinknow, 2006), little attention has been given to urban innovation. Moreover, it is unclear based on the previous studies whether innovation has an overall positive or negative effect on environmental pollution. The estimates in these studies may be biased due to inaccurate measurement of innovation level (the total R&D expenditure of enterprises in the region is considered to have statistical errors; the number of patent authorizations in the region could not reflect the market value of patents). Therefore, more evidence is necessary to understand the impact of urban innovation on pollution, especially haze pollution that is extremely harmful but receives little attention. This evidence can help inform policy regarding establishing innovative cities and achieving green development.

Model
To investigate the impact of urban innovation on haze pollution, the basic econometric model can be specified as follows: In Eq. (1), PM 2.5 is the concentration of fine particles in the air, indicating the level of haze pollution. LnUrba-nInno indicates the logarithm of urban innovation, and its coefficient α 1 measures the impact of urban innovation on haze pollution, which is the parameter of primary focus. X represents the set of control variables, including economic development level, government science and technology investment, government environmental regulation, industrial land occupancy, urban informatization level, human capital status, and the level of opening-up. μ i denotes the city fixed effect, and ε it is the random error term.
This paper also employs a two-way fixed effect regression model to determine the effect of urban innovation on haze pollution; the model is proposed as below: As the current haze pollution will not affect the historical level of urban innovation, we also use the lagged explanatory variables in Eq. (1) as explanatory variables and perform a regression analysis to deal with possible reverse causality bias.
Additionally, we calculate a difference between two years for each variable in Eq. (1) and form a new difference model for the robustness check. The difference model is set as Eq. (3): As the PM 2.5 in the last period may affect the current period, this paper further adds the lagged PM 2.5 on the right side of Eq. (1) to construct a dynamic panel model for empirical research. The dynamic panel model is set as Eq. (4): Referring to the mechanism analysis method of researchers such as Chen and Chen (2018), this paper investigates the transmission mechanism of urban innovation on haze pollution. The method is divided into two phases. In the first phase, we examine the effects of urban innovation on energy consumption (EnCsu) and industrial agglomeration (lnduAgg) using Eq. (5). In the second phase, we examine the influence of these effects of urban innovation on haze pollution using Eq. (6).

Independent variable
The indicators of urban innovation used in this study are from the innovation index of 338 cities over the 14-year period, 2003-2016, in the "China Urban and Industrial Innovation Report 2017" (hereinafter referred to as "Report") of Fudan University Industrial Development Research Center. The Report uses the invention patents granted by China National Intellectual Property Administration (CNIPA) to construct an innovation index. Compared with using the total number of patents to measure a city's innovation level in the previous research, the calculation method of innovation index in the Report is better.
First, the Report only uses the invention patents that best represent innovation capabilities as a statistical basis. Other forms of patents only need to be practicable or novel to a certain degree, while invention patents need to meet the three characteristics of practicability, novelty, and creativity; thus, they can best represent innovation capabilities.
Second, the value difference of patents is taken into consideration when the innovation index is calculated. Patent holders need to pay patent renewal fees to maintain their granted patents. Generally speaking, the longer the duration of a patent, the greater the value. Therefore, existing studies that directly use the number of patents to measure innovation are inaccurate and unreasonable. The Report exploits a patent update model to estimate the average value of patents of a different term. Based on this, the value of each patent is added to the city level to obtain the city innovation index.

Dependent variable
The dependent variable in this paper is urban haze pollution, measured by the concentration of PM 2.5 in the air. Haze pollution in China typically encompasses a large geographical area, and in recent years, millions of people have suffered from haze weather. There were extreme haze conditions in January 2013, when hazardous dense haze covered more than 1 million km 2 of China. The number of serious haze days in the central and eastern regions was generally more than 5 days, and it reached 10 to 20 days in parts of these areas. In the face of severe air pollution, the Chinese State Council issued the "Air Pollution Prevention Action Plan" and set a goal of reducing the concentration of PM 2.5 in the air. Since then, the overall air quality has improved a little, but haze pollution is repetitive and difficult to control. Every autumn and winter, many cities in some provinces such as Hebei, Shanxi, Shandong, and Henan are covered by haze, causing heavy pollution that lasts for a long time. Inhalation of the pollutants by residents can irritate the respiratory system, thereby inducing and exacerbating related diseases. The low visibility weather also leads to closures of high-speed road and flight delays. The factories have to cut or stop production, and cities suffer huge economic losses. Based on the Chinese central and local Government Work Report and key areas in the work plan of the Ministry of Ecology and Environmental in recent years, controlling haze is still one of the core purposes of environmental protection. PM 2.5 concentration has been the primary haze pollutant, and the PM 2.5 data are obtained from the NASA Socioeconomic Data and Application Center of Columbia University. Multiple satellite-mounted devices measure the aerosol optical depth (AOD) of aerosol systems in the air, and geographically weighted regression (GWR) is combined with global ground measurements to estimate the annual mean PM 2.5 concentration for various cities in the world from 1998 to 2016 (Van et al. 2018). This belongs to the non-point source data and has wider coverage than the ground point source monitoring data. Also, it can fully reflect the regional PM 2.5 concentration and its variation characteristics; therefore, the data has been widely used in various studies. In addition to PM 2.5 , sulfur dioxide and nitrogen oxides are also important components of haze. Therefore, the concentration of SO 2 and NO 2 in the air is also used as a dependent variable. Figure 1 displays the kernel density estimation of PM 2.5 for the year 2016 in the eastern, central, and western regions of China. As can be seen from the figure, the level of haze pollution in the eastern coastal areas of China is lower than that in the central regions on the whole. This is closely related to the fact that the eastern cities at the forefront of opening-up have always been committed to developing high-tech industries and optimizing the environment for innovation in recent years. Of course, a small number of cities in eastern China still have serious haze pollution due to the large size and the high population density of the city. China's western regions are relatively underdeveloped and less industrialized; therefore, the overall level of haze pollution is low. economic growth, China has sacrificed its environmental quality to some extent. However, in recent years, China has embarked on the path of green and sustainable development and has tried to drive economic growth by optimizing resource allocation, promoting technological advancement, and enhancing the levels of urban innovation.
Some control variables are controlled in Eq.
(1) to deal with the problem of omitted variables. The names and construction methods of the variables are as follows. The per capita GDP is used to measure economic development, and the data are deflated to strip out the effects of inflation. Governmental science and technology investment takes the logarithm value of per capita government science and technology expenditure. The degree of government environmental regulation is measured by the green coverage rate of the built-up area. The human capital is represented by the logarithm of the number of college students per 10,000 people. The level of informatization is measured by the logarithm of the number of internet users in the city. The FDI is measured by the proportion of foreign direct investment in GDP. In Eqs. (5) and (6), which are used to study the transmission mechanism of urban innovation on haze pollution, energy consumption is measured by the logarithm of electricity consumption per capita; the location quotient is used to measure industrial agglomeration. The location quotient index can reflect the spatial distribution of geographical factors more realistically and eliminate the regional scale difference factors (Li and Zhang, 2013;Yang, 2013). The above data are from the China National Bureau of Statistics. Table 1 shows descriptive statistics of variables. The mean value of PM 2.5 is 37.104, which is higher than the standard specified in the "Air Quality Guidelines" 2 set by WHO and the "Ambient Air Quality Standards" 3 set by the Chinese Research Institute of Environmental Sciences, indicating that China's haze pollution problem is still prominent. The mean value of LnUrbanInnov is − 0.273, implying that innovation capability in China still has much room for improvement. The standard deviation of LnUrbanInnov is 1.893, indicating a huge gap in the innovation level of Chinese cities.

Empirical results and discussion
Baseline results PM 2.5 is the main component of haze pollution. This part first explores the impact of urban innovation on PM 2.5 concentration in the air. Column (1) of Table 2 reports the baseline regression results of Eq. (1). In controlling the urban characteristics such as the level of economic development, industrial land occupancy, the city government's scientific and technological expenditure and environmental regulation, and other factors that may lead to omitted variable bias, and considering city fixed effect, urban innovation is seen to be significantly negatively correlated with haze pollution. With respect to the possible reverse causation between haze pollution and urban innovation, we use lagged explanatory variables in Eq. (1) as explanatory variables and perform a regression analysis to deal with possible reverse causality bias. The regression results reported in column (5) of Table 2 suggest a negative association between urban innovation and haze pollution, thus verifying hypothesis 1. Moreover, this paper adds the time fixed effect and constructs a two-way fixed effect model for regression analysis. As shown in column (2) of Table 2, the coefficient of the logarithm of the urban innovation index is − 1.030, indicating a significant negative correlation between urban innovation and PM 2.5 concentration. Assuming that the urban innovation level increased by 1%, the PM 2.5 concentration would be decreased by 1.030. Likewise, the lagged explanatory variables were used in the two-way fixed effect model for regression analysis to deal with the reverse causal bias. The regression coefficient of the lagged urban innovation has not changed fundamentally.
The evidence from our empirical studies suggests that technological progress in cities is environmentally friendly. Indeed, China proposed to act on achieving sustainable development goals at the beginning of the twenty-first century. Furthermore, in recent years, it has continued to strengthen environmental regulations. All these efforts make the technological innovation of enterprises develop in an environment-friendly direction, and the benefits of innovation offset the cost of environmental protection. Moreover, innovation-driven development is currently China's core development strategy. Many cities are trying to establish innovative practices to reduce pollution in economic growth with innovation. For example, Suzhou introduced the system of total amount control of pollutant emission and emission trading and devoted funds from the transaction to independent R&D and the introduction of environmental technologies. Although the empirical evidence in this paper shows that innovation has played a role in reducing haze pollution, there are still about one-third of the cities in China whose PM 2.5 concentration cannot meet the national second-level standard and severe pollution often occurs. Therefore, it is necessary to exploit the potential of the city's innovative development and improve the environment with technological advances in the future.
Among other control variables, the coefficient of industrial land occupation is significantly positive, demonstrating that industrial pollution is an important cause of haze pollution in China. The significantly negative coefficients of government expenditure on science and technology and the informatization level indicate that these variables have an inhibitory effect on haze pollution. The coefficient of environmental regulation is significantly negative; that is, China's environmental regulations have played a role in promoting industrial transformation, optimizing resource allocation, and ultimately promoting emission reduction. The coefficient of FDI is not significant, and the coefficient of human capital is significantly positive. These suggest that China should import foreign capital with high scientific and technological content and develop high-tech talents.
We also add the interaction term between city fixed effect and time trend to control the individual time trends in each city; the results are displayed in column (3) and show that urban innovation is significantly negatively correlated with haze pollution. The shocks over time may have different effects on different cities and may be multidimensional. Therefore, this paper also constructs an interactive fixed effects model for regression analysis. The pollution reduction effect of urban innovation still exists, as shown in column (4).
The findings of our study imply that urban innovation counts for the reduction of haze emissions, which are consistent with past findings (Baniak and Dubina, 2012;Zhang et al., 2019;Ge, 2019;Liu, 2018;Ma et al., 2020). However, the present analysis studies innovation and pollution in the city framework and uses the urban innovation index that considers the market value of patents. In addition, the dependent variable in this paper is haze pollution that has been paid little attention to by other studies. Moreover, this paper uses an advanced econometric model that has not been applied in previous related studies-the interactive fixed effect model to ensure the robustness of the findings.
In addition to fine particles, sulfur dioxide and nitrogen oxides are also major components of haze. Burning coal and oil not only causes smoke pollution but also emits sulfur dioxide. Sulfur dioxide endangers human health and causes corrosion of buildings and metal materials. Nitrogen dioxide is the most toxic of all the nitrogen oxides and easily causes acute and chronic poisoning. Moreover, nitrogen dioxide is likely to be suspended in the air for a long time and is more likely to be inhaled. Therefore, this paper also uses the content of sulfur dioxide and nitrogen dioxide in the air as the dependent variable to study the impact of urban innovation on haze pollution. 4 Similar to the results of PM 2.5 ,  (1) and (2) of Table 3 illustrate that urban innovation is negatively associated with sulfur dioxide and nitrogen dioxide pollution. Likewise, we also study the effect of lagged urban innovation on sulfur dioxide and nitrogen dioxide pollution to deal with reverse causality. The coefficients of lagged urban innovation are also statistically significant. The results imply that the improvement of urban innovation is associated with lower levels of haze pollution. In the existing research on technological innovation and environmental pollution, only a few researchers, such as Liu (2018), have paid attention to haze pollution. However, Liu (2018) used annual average concentrations of PM 10 to measure the haze pollution of each province. Compared with PM 10 , PM 2.5 is smaller and rich in many toxic and harmful substances, which are more harmful to human health. Therefore, this study mainly uses annual average concentrations of PM 2.5 to measure haze pollution. Moreover, we employ other components of haze pollution-the content of sulfur dioxide and nitrogen dioxide in the air as the explained variables for research. We also confirm that urban innovation is beneficial to reducing haze pollution. Therefore, this study fills the gap in the existing literature on innovation and haze pollution.

Robustness checks
To check the robustness of the effect of urban innovation on haze pollution, this paper also exploits a difference model to examine if urban innovation is a determinant of haze pollution, as shown in Eq. (3). The difference model is based on calculating a difference between two years for each variable in Eq. (1), eliminating the influence of city fixed effects that do not change with time. Moreover, because the urban innovation variable is in logarithmic form, the difference model uses the growth rate of urban innovation (instead of absolute values) to perform regression analysis, reducing the endogenous problem of Eq. (1). Urban innovation is also found to reduce haze pollution, as shown in column (1) of Table 4.   Since the PM 2.5 concentration in the previous period may affect the current haze pollution level, this paper also exploits the dynamic panel model in Eq. (3), which adds the explanatory variable, the PM 2.5 concentration of the last year to assess the influence of urban innovation on haze pollution. The dynamic panel model is estimated using the difference generalized method of moments (GMM). The results reveal that urban innovation has a negative effect on haze pollution (column (2) of Table 4). In addition, we perform robustness checks by examining the influence of urban innovation on the lowest or highest concentrations of PM .2.5 in each city in a year. The coefficients and significance of the core explanatory variables do not change much (columns (3)-(6) of Table 4), which is consistent with the previous research results. Overall, the results provide robust evidence that urban innovation is conducive to reducing haze pollution.

Analysis of the transmission mechanism of urban innovation on haze pollution
In this part, according to the mechanism testing method of Chen and Chen (2018), the pathways through which urban innovation affects haze pollution are discussed. First, urban innovation may affect haze pollution by reducing energy consumption. On the one hand, a large number of suspended particles, which are an important source of haze pollution, are generated during the energy utilization process. The improvement of energy utilization efficiency and the use of clean energy can help alleviate pollution. On the other hand, the improvement of the level of urban innovation has led to an increase in energy efficiency (Subrahmanya and Kumar, 2011;Ramirez-Portilla et al., 2014;Cagno et al., 2015). To examine this mechanism, this paper selects the per capita electricity consumption as the proxy variable of energy consumption. The results of the corresponding regression analysis are presented in Table 5. Columns (1) and (2) show that the regression coefficient of energy consumption is significantly positive, indicating that energy consumption exacerbates haze pollution. The coefficient of urban innovation in columns (3) and (4) is negative and statistically significant, confirming that urban innovation can effectively reduce energy consumption. Therefore, hypothesis 2, that the increase of urban innovation will decrease energy consumption that mitigates haze pollution, is verified. Second, we investigate whether the relationship between urban innovation and haze pollution is mediated by industrial agglomeration. On the one hand, there are economies of scale of environmental pollutant emissions and treatment (Lu and Feng, 2014), and agglomeration of economic activities are found to have a reducing effect on environmental pollution. On the other hand, urban space agglomeration of innovation is an important source of industrial agglomeration (Chung and Alcácer, 2002;Guastella and Van Oort, 2015;Goldman et al. 2016). The regression coefficients of the industrial agglomeration indicate that industrial agglomeration has a significant negative association with PM 2.5 concentration in the air (columns (1) and (2) of Table 6). At the same time, the estimated coefficient for urban innovation suggests a significant and positive association with industrial agglomeration (columns (3) of Table 6). In addition, the coefficient of lagged urban innovation is also statistically significant, implying that the advancement of urban innovation is associated with higher levels of industrial agglomeration after addressing the reverse causality bias. Therefore, the transmission mechanism of urban innovation affecting haze pollution through industrial agglomeration exists, thus verifying hypothesis 3.
Past research found that technological innovation improved energy efficiency and reduced energy consumption, and thus reducing pollution (Sohag et al. 2015;Miao et al. 2018). This paper tests this transmission mechanism based on the research of city-level data in China. Moreover, based on research on innovation and industrial agglomeration, we also found that innovation reduces haze pollution by influencing industrial agglomeration. Indeed, innovations in production processes and improvements in machinery and equipment are an important part of China's current technological innovation, which has led to an increase in energy conversion efficiency. Moreover, in recent years, China has also achieved technological breakthroughs in wind energy and solar energy and has effectively increased the utilization of clean energy. All of these have greatly reduced the discharge of pollutants, thereby alleviating haze pollution. In addition, China's vigorous innovations in transportation, information technology, and data sharing in recent years are all conducive to breaking through the bottleneck in the development of industrial agglomeration. At the same time, the agglomeration of enterprises is also beneficial to achieve cost savings in pollution control and sharing pollution control facilities among enterprises, which contributes to the reduction of pollution.

Heterogeneity analysis of urban innovation affecting haze pollution
The baseline regressions assume that urban innovation has the same impact on different time periods and regions. However, urban innovation and haze pollution in China show great differences in different time periods and different regions. The sample is then divided into sub-samples based on the year 2013 to examine if the role of urban innovation differs among different periods. Columns (1) and (2) of Table 7 show that urban innovation significantly affects air pollution over time, and the reducing effects tend to increase as the year advances. The results are consistent with the reality that China has implemented and enforced environmental standards and regulations and encouraged technological  (3), (4), and (5) of Table 7, respectively, implying that urban innovation has a significant negative impact on haze pollution in the eastern, central, and western cities; however, the effect of impact decreases sequentially. This may be attributable to the fact that the eastern cities can provide better human resources, and financial and infrastructure support for technological innovation, and the resource allocation and utilization efficiency are higher; thus, urban innovation can better exert a pollution reduction effect.

Conclusions and policy implications
Haze pollution is a serious environmental issue, and it is of great significance to control pollution without affecting economic development. Innovation is integral to achieving the aims of improving the ecological environment. In this paper, we use PM 2.5 concentration data from 2003 to 2016 of Chinese cities based on fixed effect models to study the relationship between urban innovation and haze pollution. We find that the improvement of urban innovation level significantly reduces PM 2.5 concentration in the air. This conclusion remains unchanged after dealing with endogenous problems such as reverse causality. Moreover, urban innovation is also conducive to reducing the concentration of SO 2 and NO 2 in the air, thus alleviating haze pollution. Analysis of the transmission mechanism shows that urban innovation can alleviate haze pollution by reducing energy consumption and promoting industrial agglomeration. Moreover, the negative effects of urban innovation on haze pollution are increasing with time, and urban innovation has a greater mitigation effect on haze pollution in eastern cities than in central and western cities of China. The empirical results of this paper suggest that technological innovation and the urban agglomeration of innovation are important driving forces for reducing haze pollution. Preventing pollution at its source is the key to reducing haze weather and improving air quality. The joint efforts of governments of cities, businesses, and the public will give an impetus to reducing air pollutants.
The government should strengthen the protection of intellectual property rights and improve the external environment for the innovation of enterprises. It can also encourage enterprise innovation by offering tax incentives and subsidies to high-tech enterprises. Additionally, government procurement policies could be used to support innovative and environmentally friendly products. At the same time, the government can build a platform for universities, research institutions, and enterprises to promote collaboration so as to enhance the capacity of R&D in pollution control. Pollutant emission standards and enterprise environmental information disclosure systems should be implemented to force enterprises to conduct technological innovation and upgrade. Finally, because urban innovation can alleviate haze pollution through promoting industrial agglomeration, the government could promote industrial agglomeration by optimizing urban planning and developing industrial parks, thereby achieving economies of scale in pollution control.
Business enterprises can enhance their ability for independent innovation and introduce advanced technology to strengthen innovation in production processes, thus improving energy conversion efficiency and pollution control during production. At the same time, companies can enhance their waste disposal capacity through technological innovation, such as using advanced technology to separate and absorb harmful substances in the exhaust gas. With the rapid development of information technology, big data and artificial intelligence can be applied to pollution control to improve efficiency. For instance, interconnected energy conservation and emission reduction data collection and information monitoring platforms could be set up. In addition, companies can also improve the efficiency of environmental pollution reduction through cooperation and facility sharing.
Funding This work was supported by the China National Social Science Fund (Grant No. 20FJLB023).

Availability of data and materials
The dataset used in this study was obtained from the China Urban and Industrial Innovation Report 2017 and the China National Bureau of Statistics.

Declarations
Ethics approval and consent to participate Not applicable.

Competing interests
The authors declare no competing interests.