Urban Spatial Development Mode and Haze Pollution in China: From the Perspective of Polycentricity

Abstract

Increasing urbanization in China has caused severe haze pollution in Chinese cities in recent years. This study investigates the impact of urban spatial development mode on haze pollution from the leapfrog spatial dimension. We constructed a dataset containing LandScan population dynamic statistical data, PM2.5 surface concentration data, and urban economic statistical data and adopted ordinary least squares (OLS) and instrumental variable methods. The findings indicate that the polycentric spatial structure within the city increases the PM2.5 levels, indicating that the urban monocentric spatial development mode is conducive to alleviating haze pollution. The use of the core explanatory variables, lagged by one and two periods, and of the number of Qing Dynasty walls as the instrumental variable confirm the robustness of the benchmark results. The heterogeneity analysis demonstrates that, in cities with underdeveloped public rail transit, the monocentric spatial development mode has a more obvious effect on reducing haze pollution. The results of the influence mechanism test show that the urban polycentric spatial development mode aggravates urban haze pollution, mainly by encouraging residents to travel by private transport, thereby increasing energy consumption. The obtained research results provide a sufficient basis for taking appropriate measures to govern haze pollution in Chinese cities from the perspective of polycentric spatial development mode.

1. Introduction

With increasing urbanization in China, Chinese cities have faced severe haze pollution in recent years [1]. Studies reveal that approximately 28% of China’s territory (approximately 2.72 million square kilometers) was exposed to severe haze pollution in 2010 [2,3]. Poor air quality significantly affects the public [4,5,6,7], with incalculable effects on the health of infants and the elderly [8,9,10,11,12]. Related studies have shown that in the European Union, air pollution is estimated to cause over 400,000 premature deaths annually, mainly in cities [13,14]. In addition, air pollution can lead to heavy economic costs [15]. In this context, China’s haze pollution problem has been increasingly addressed by government departments and scholars worldwide [16]. In the field of urban economics, it is generally believed that urban spatial structure is closely related to haze pollution. Specifically, the polycentric spatial structure of cities has led to a reduction in the intensity of land use and a centrifugal distribution of population and economic activities in cities, which has resulted in a series of environmental problems such as deterioration of air quality and loss of agricultural space [17]. Therefore, the adoption of reasonable spatial development modes by cities requires attention.

Urban spatial development modes involve aspects such as population distribution within the city, mixed land development, and the use of public transportation and other public facilities. Currently, much literature measures the urban polycentric spatial structure from the leapfrog spatial dimensions [18], reflecting the urban leapfrog spatial development mode. Different from the continuity of population distribution emphasized in vertical or horizontal spatial development, the polycentric spatial structure emphasizes “leapfrog” development in a sub-central way, which is agglomeration in dispersion or centralized dispersion [19]. Specifically, related studies use the coefficient of variation term to measure the urban vertical spatial development pattern [20]. Although it can capture the differentiated distribution of population and economic activities within the city, the level of the urban coefficient of variation term cannot reflect monocentric or polycentric spatial structure because of the inability to identify the main/sub-center. Additionally, some scholars have used the urban geometric form index to measure the spatial development mode [21], but it is still impossible to judge the urban internal spatial structure from the rules of urban geometry. Therefore, the internal spatial scale of the city, mainly from the perspective of form, emphasizes the existence of multiple centers within the city, and the main center, secondary center, and population or economic scale are uniformly distributed. Further exploration of the influence of the urban polycentric spatial development mode on haze pollution, reasonable optimization of the urban spatial structure, and promotion of a green urban economy have strong theoretical and practical significance.

Based on the research background of this study, the following questions are proposed: First, from the perspective of urban leapfrog spatial development, will the multi-decentralized distribution of population and economic activities produce positive or negative environmental externalities? What is the impact of polycentric spatial development on haze pollution? Second, what is the main mechanism by which polycentric urban spatial development affects haze pollution? Third, the development of an urban polycentric space will increase commuting distance and time. In addition, relevant studies have shown that sustainable urban transport has a strong impact on urban development and the environment [22]. Therefore, the difference in the level of urban public transportation facilities will lead to heterogeneity in the impact of the polycentric spatial development mode on haze pollution. In view of this, this study first uses the LandScan global population spatial distribution database to identify the main and sub-centers within each city in China, measures the polycentric spatial structure by calculating the number of urban sub-centers and the proportion of the population of all centers, and represents the urban polycentric spatial development mode. At the same time, the average surface concentration of PM2.5 over the years was extracted to represent the level of urban haze pollution. Second, OLS and instrumental variable methods were used to identify the causal relationship between the urban multicentric spatial structure and haze pollution. Subsequently, the urban rail transit, represented by the subway, was used to reflect the level of public transport facilities, the difference in which was employed to analyze the impact of heterogeneity. Finally, according to the relevant theoretical analysis, the main mechanism by which the urban polycentric spatial development mode affects haze pollution was empirically tested.

The main contributions of this paper are as follows: First, based on the perspective of the polycentric spatial structure within the city, it explores the influence of urban polycentric development mode on haze pollution from the dimension of leapfrog space development. It helps to clarify the environmental externalities of the multi-decentralized distribution of population and economic activities within the city and is a useful supplement to the existing literature. Second, selecting the number of Qing Dynasty walls as the tool variable of polycentric spatial structure, this study not only addresses the problem of potential endogeneity in the regression model but also identifies the causal relationship between the urban polycentric spatial development mode and haze pollution more accurately. It also provides new ideas for the selection of tool variables of urban spatial structure. Third, the heterogeneity of the influence of haze pollution on the level of urban polycentric spatial development mode is comprehensively tested to facilitate the targeted formulation of relevant policies. Fourth, based on the theoretical analysis, it tests the main action mechanism of the urban polycentric spatial development mode affecting haze pollution, providing empirical evidence for the relevant theories. The remainder of this paper is organized as follows. Section 2 presents the theoretical analysis and research hypotheses. Section 3 then introduces the model, variable, and data. Section 4 focuses on the empirical results and discussion. The conclusion and policy implications are provided in Section 5.

2. Theoretical Analysis and Research Hypotheses

The polycentric urban spatial development mode is closely related to the spatial distribution of the urban population. Specifically, the degree of polycentricity of the urban spatial structure is high, which is manifested in the weak continuity of population distribution, and economic activities are more likely to spread by “leapfrogging” within the city. In contrast, low centralization of the urban spatial structure indicates a more concentrated distribution of the population and economic activities. A large population concentration in the core areas of the city causes a high degree of spatial matching between the residence and employment places of citizens, which is conducive to reducing people’s demand for private car travel. Therefore, in theory, polycentric spatial development mainly affects haze pollution by changing the transportation modes of residents. First, a monocentric spatial structure shortens commuting distance and time within the city, resulting in people relying less on private cars for travel [17,23]. The decline in citizen demand for private cars will relieve the pressure on urban roads, which is conducive to reducing traffic congestion to some extent. As improvements in traffic congestion can significantly reduce the fuel consumption and exhaust emissions of motor vehicles [24], they are conducive to reducing urban haze pollution [25]. Second, there is an alternative relationship between urban public transportation and private car travel; a more monocentric spatial structure encourages people to choose public transport. Compared with private cars, urban public transport is undoubtedly greener. Therefore, enhanced use of urban public transportation for travel can reduce urban haze pollution.

Further, multi-decentralized development within cities will also have an impact on energy efficiency and the sharing of environmental infrastructure [26]. Specifically, cities with a high level of centralized development can contribute to the dissemination and sharing of cleaner production technologies, which can reduce haze pollution by improving the energy utilization efficiency of enterprises. Simultaneously, the concentration of industrial activity space is more conducive to the sharing of environmental protection infrastructure, which increases the significance of its haze reduction effect. Based on the above analysis, the first research hypothesis is proposed.

Hypothesis 1:

From the perspective of leapfrog spatial development, the urban polycentric spatial development mode mainly aggravates urban haze pollution levels by encouraging residents to travel by private car and increasing energy consumption.

Hypothesis 1 proposes that residents adopting a less environmentally friendly transportation mode is one of the main mechanisms by which urban polycentric spatial development affects haze pollution. Owing to the evident gap in the level of rail transit and other infrastructure in Chinese cities, the heterogeneity of the influence of urban rail transit on the urban polycentric spatial development mode is worth an in-depth discussion. For example, Beijing, Shanghai, Nanjing, and other cities have many rail transit lines and wide coverage, making them some of the main travel routes. However, by the end of this study, in 2016, only 31 cities in China had built and opened rail transits (the statistics related to urban rail transit are from the China Urban Construction Statistical Yearbook), such as subways, indicating that most urban residents could not avail of it, leading to the heterogeneous impact of polycentric spatial development modes in different cities on haze pollution. Specifically, the more developed the urban rail transit, the more residents use it, rather than private cars, to travel. Owing to the characteristics of green and environmental protection of urban rail transit facilities, traffic congestion will be relatively alleviated, and the effect of haze pollution on the deterioration of the urban polycentric spatial development mode will be weakened. Therefore, the second research hypothesis is as follows.

Hypothesis 2:

The higher the development level of urban rail transit, the lower the effect of the urban polycentric spatial development mode in exacerbating haze pollution.

3. Model, Variable, and Data

3.1. Benchmark Model

Based on the previous analysis, the benchmark model described in this section is as follows:

$$\mathrm{P}\mathrm{M}{2.5}_{\mathrm{i}\mathrm{t}}={\mathsf{\alpha }}_0+{\mathsf{\alpha }}_1{\mathrm{P}\mathrm{o}\mathrm{l}\mathrm{y}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\alpha }}_{\mathrm{x}}{\mathrm{C}\mathrm{o}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{o}\mathrm{l}\mathrm{s}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_{\mathrm{x}}{\mathrm{D}}_{\mathrm{i}}+{\mathsf{\gamma }}_{\mathrm{t}}+{\mathsf{\epsilon }}_{\mathrm{i}\mathrm{t}}$$

(1)

In the above Equation (1), i represents the city (city refers to the municipal districts of prefecture-level cities (including four municipalities directly under the Central Government) divided by administrative district boundaries. With the continuous progress of urbanization, as of 2016, China has cities with a population size of more than 10 million, such as Shanghai, Chongqing, Beijing, Tianjin, Guangzhou, Shenzhen, and Chengdu, as well as cities with a population size of less than 300 thousand, such as Jiayuguan, Jinchang, Yingtan, Lijiang, Haidong, Heihe, Xianning, and so on) and t represents the year. PM2.5 characterizes the level of haze pollution, the explained variable in the model, which is measured by the average annual PM2.5 surface concentration. PM2.5 takes the log form in the regression. Poly represents the urban polycentric spatial structure and is the core explanatory variable of this model. It is measured by the number of urban sub-centers (Poly_a) and the proportion of urban sub-center population in all centers (Poly_b), respectively. Poly_a takes the logarithmic form in the regression; in the model, Controls is the control variable at the city level over time and includes the proportion of secondary industry (Second), level of economic openness (Pfdi), road area per capita (Proad), GDP per capita (Pgdp) and the square of GDP per capita (Pgdp²). D_i represents a series of urban characteristic variables that do not change with time, including whether it is a provincial capital city (Capital), tourist city (Travelcity), number of historical sites (Historicalsite), average summer temperature (Summertemp), and geographical latitude (Latitude). Furthermore, controlling for year-fixed effect γ_t, α₀, α₁, α_x, β_x is the estimated coefficient, and ε is the random interference term.

Considering that the core explanatory variable—the urban polycentric spatial structure—has little variation in the time dimension, the variation in samples is mainly due to cities, and the fixed effect of cities will affect the accuracy of the estimation results. Therefore, we borrowed Harari’s (2020) practice to stop controlling for city fixed effect in the benchmark model [21] but solved the bias of possible missing variables by adding a set of urban characteristic variables (D_i) that do not change over time [27,28]. In the regression, logarithms were taken for GDP per capita, number of historical sites, average summer temperature, and geographical latitude.

3.2. Variable and Data Sources

3.2.1. Urban Polycentric Spatial Structure

From the dimensions of urban leapfrog spatial development, we measured the urban polycentric spatial structure based on LandScan global population spatial distribution grid data provided by the Oak Ridge National Laboratory, Tennessee, USA (source: https://www.eastview.com/resources/e-collections/landscan/ (accessed on 16 August 2024)). First, because the LandScan population distribution data consider all population economic activities, the population is estimated to be 1 km² in the grid within the scale, and the distribution of population and economic activity within the city can be accurately captured. Related studies have used different methods to identify the main centers and sub-centers of cities [29,30,31,32]. We followed Li and Liu (2018) to identify the main centers and sub-centers of Chinese cities by setting the screening criteria for population centers [33]. Second, the number of urban sub-centers (Poly_a) was calculated as the core explanatory variable to characterize the urban polycentric spatial development mode. Further, the proportion of the urban sub-center population in the central population (Poly_b) was measured to determine the urban polycentric spatial structure. The larger the number of urban sub-centers and the higher their proportion in the population of all centers, the more polycentric the urban spatial development mode.

3.2.2. Haze Pollution

A large number of existing studies have used suspended particulate matter concentrations, represented by PM2.5, as a measure of haze pollution in the literature compared to air quality data from ground-based monitoring stations [34,35,36,37]. The study uses the global PM2.5 annual average surface concentration data (1998–2016) provided by the Socioeconomic Data and Applications Center (SEDAC) of Columbia University (source: https://sedac.ciesin.columbia.edu/data/sets/browse (accessed on 16 August 2024)). These PM2.5 concentration data were obtained by combining AOD inversions from multiple satellite instruments, including NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS), Multi-Angle Imaging Spectroradiometer (MISR), and Sea View Wide Field of View Sensor (SeaWiFS) [38]. The PM2.5 annual average surface concentration data can not only solve the problem of missing historical data on PM2.5 concentration but also effectively avoid the statistical errors caused by artificial factors. It is more objective and accurate, and the coverage of this satellite remote sensing data is also wider. Hence, the PM2.5 annual average surface concentration of Chinese cities, extracted over the years, constituted the explained variable. However, the global PM2.5 annual average surface concentration data were available up to 2016. Since the focus of our paper is to identify the impact of urban polycentric spatial structure on PM2.5 concentration, we set the study period to 2000–2016. A sample size of 17 years was sufficient to support the analysis of economic laws in this study.

The temporal evolution trend of urban PM2.5 concentration in China is shown in Figure 1. It can be found that the annual average surface PM2.5 concentration was in an increasing trend from 2000 to 2007, and the PM2.5 concentration reached its highest point in 2007, indicating that the haze pollution in Chinese cities was gradually increasing in this period, and the haze pollution in cities was the most serious in 2007. The PM2.5 concentration declined in the period of 2007–2012, indicating that the problem of urban haze pollution was somewhat alleviated in this period. The PM2.5 concentration increased in 2013 compared to 2012 but then was in a decreasing trend until 2016. Overall, urban haze pollution levels in Chinese cities experienced an inverted U-shaped evolution from 2000 to 2016.

3.2.3. Control Variables

This study selects a series of control variables, which are introduced as follows. (1) The proportion of secondary industry (second) was measured by the proportion of urban secondary industry output to GDP. Because industrial production is the main source of gas pollution, the expected influence coefficient is positive [39]. (2) The level of economic openness (Pfdi) was measured as the ratio of foreign direct investment (FDI) to GDP, representing the dependence of urban economic development on foreign investment. Enterprises in developed countries will treat developing countries as “pollution shelter” [40], and high-polluting industries with foreign investment will increase the emission of polluting gases and deteriorate air quality, and the expected impact coefficient is positive. (3) Road area per capita (Proad) is calculated as the proportion of the total urban road area to the total urban population, which represents the carrying capacity of urban road traffic. Cities with a higher road area per capita have wider roads, which can ease motor vehicle congestion in the city and reduce motor vehicle exhaust emissions [41]; therefore, the coefficient of expected influence is negative. (4) GDP per capita (Pgdp) is measured by the ratio of the urban gross regional product (GDP) to the total population and represents the level of economic development of the city; higher levels of economic development are often accompanied by more severe haze pollution [42]. Therefore, the expected impact coefficient of the GDP per capita is positive. (5) Whether it is a provincial capital city (capital): the provincial capital city virtual variable is set to reflect the city’s administrative level. (6) Whether it is a tourist city (Travelcity): a list of excellent tourist cities based on the Sohu tourism network and set the tourist city virtual variable, which is used to represent the development characteristics of the city. (7) The number of historical sites (historical site) refers to the number of key national cultural relics under protection units approved and announced by the State Council of China. In 1961, 1982, 1988, 1996, 2001, and 2006, the State Council of China announced the first seven batches of national key cultural relics protection units in 2013, respectively, and the cultural relics protection units were placed in urban districts and counties across the country. In this study, the first seven batches of cultural sites were the number of historic sites, reflecting the actual number of local sites and representing the historical and cultural characteristics of the city. (8) Summer average temperature (summer), based on climate data from 743 conventional sites, was provided by the National Climate Center from 1970 to 2010. For cities with weather stations, the station data were used directly as the climate standards of the city. For cities without these weather stations, we referred to the method of Hanson (2005) and used the data of the nearest weather station [43]. (9) The geographical latitude (Latitude) was extracted from the national electronic map provided by the National Center for Basic Geographic Information, using ArcGIS 10.1 to represent the geographical location characteristics of the city. The above data were obtained from the China Urban Statistical Yearbook (2001–2017), the National Center for Basic Geographic Information, and the National Climate Center.

Table 1. Description of the variables and the descriptive statistics.

Variables	Variable Description	Observations	Mean	Standard Deviation
LnPM2.5	Natural logarithm of annual average surface concentration of PM2.5	5925	3.554	0.594
LnPoly_a	Natural logarithm of the number of urban sub-centers	3571	0.998	0.703
Poly_b	Population of urban sub-centers as a proportion of the population of all centers	4130	0.326	0.218
Second	Proportion of secondary industry	4746	49.842	12.438
Pfdi	Level of economic openness	4346	2.783	4.153
Proad	Road area per capita	4738	9.792	11.306
LnPgdp	Natural logarithm of GDP per capita	4773	10.075	0.825
Capital	Whether it is a provincial capital city	4849	0.113	0.317
Travelcity	Whether it is a tourist city	5002	0.462	0.499
LnHistoricalsite	Natural logarithm of the number of historic sites	4157	0.351	0.470
LnSummertemp	Natural logarithm of average summer temperature	4849	3.198	0.132
LnLatitude	Natural logarithm of geographic latitude	4849	3.472	0.206

presents the variable descriptions and descriptive statistics.

4. Empirical Results and Discussion

4.1. Benchmark Regression Results

In this section, the OLS method was used to regress the relationship between the polycentric urban spatial structure and haze pollution. The benchmark regression results are presented in

Table 2. Benchmark regression results.

	LnPM2.5
	(1)	(2)	(3)	(4)	(5)	(6)
LnPoly_a	0.072 ***	0.067 ***	0.049 ***	0.050 ***
	(0.009)	(0.010)	(0.009)	(0.009)
Poly_b					0.076 ***	0.076 **
					(0.029)	(0.030)
Second				0.004 ***		0.004 ***
				(0.001)		(0.001)
Pfdi				0.005 **		0.004 *
				(0.003)		(0.002)
Proad				−0.002 **		−0.002 ***
				(0.001)		(0.001)
LnPgdp				0.289 ***		0.259 ***
				(0.079)		(0.075)
(LnPgdp)2				−0.019 ***		−0.017 ***
				(0.004)		(0.004)
Capital			0.057 ***	0.106 ***	0.058 ***	0.112 ***
			(0.017)	(0.020)	(0.018)	(0.020)
Travelcity			−0.018	0.011	−0.025 *	0.002
			(0.014)	(0.015)	(0.013)	(0.014)
LnHistoricalsite			0.142 ***	0.154 ***	0.147 ***	0.157 ***
			(0.013)	(0.014)	(0.012)	(0.013)
LnSummertemp			2.117 ***	2.062 ***	2.398 ***	2.346 ***
			(0.098)	(0.096)	(0.076)	(0.088)
LnLatitude			0.777 ***	0.740 ***	0.881 ***	0.849 ***
			(0.049)	(0.048)	(0.044)	(0.047)
Year FE	NO	YES	YES	YES	YES	YES
Observations	3571	3571	3042	2721	3537	3160
R2	0.016	0.050	0.362	0.377	0.428	0.415

Note: ***, **, and * denote the 1%, 5%, and 10% significance levels, respectively; values in brackets below the coefficients are robust standard errors; coefficients on constant terms are not reported because of space limitations.

. The number of urban sub-centers (Poly_a) is used in the first to fourth columns as the core explanatory variable to measure the polycentric spatial structure of the city. Specifically, the first column did not add any control variable or year-fixed effect, the second column added the year-fixed effect based on the first column, the third column added a series of city characteristic variables that do not change over time on the basis of the second column, and the fourth column added the control variables at city level on the basis of the third column. The results of the first to fourth columns in Table 2 show that the coefficient of urban sub-centers (Poly_a) is significantly positive at the 1% level, indicating that the polycentric spatial structure can increase the concentration of PM2.5; that is, the polycentric spatial development mode will exacerbate urban haze pollution. In order to more intuitively reflect the relationship between urban spatial development mode and haze pollution in the baseline regression, we plotted a scatter plot, as shown in Figure 2.

In the fifth and sixth columns of Table 2, the core explanatory variables in the regression are replaced, using the proportion of the urban sub-center population to the population of all centers (Poly_b) to characterize the urban polycentric spatial structure. In the fifth column, a series of urban characteristic variables and year-fixed effects over time are controlled, and the sixth column adds control variables at the city level over time based on the fifth column. The results of the fifth and sixth columns in Table 2 show that the coefficient of the proportion of urban sub-center population in all central populations (Poly_b) is positive and significant at the 1% and 5% levels, respectively, indicating that the higher the polycentric degree of the urban spatial structure, the more serious the haze pollution. To sum up, it can be seen that the urban polycentric spatial development mode will aggravate the urban haze pollution level, and adopting the monocentric spatial development mode is more conducive to alleviating haze pollution.

The regression results of the control variables in the fourth and sixth columns of Table 2 are presented below. The coefficients of the proportion of secondary industry (second) and the level of economic openness (Pfdi) are significantly positive, indicating that the proportion of secondary industry and the improvement in the level of economic openness will aggravate the haze pollution. The coefficient of road area per capita (Proad) is significantly negative, indicating that an increase in road area per capita will alleviate haze pollution levels. The coefficient of GDP per capita (Pgdp) is significantly positive at the 1% level, and the coefficient of the square of GDP per capita (Pgdp²) is significantly negative at the 1% level, indicating an inverted U-shaped relationship between economic development and haze pollution. Further, to avoid the bias of missing variables caused by the characteristics of cities that do not change over time, a series of urban characteristic variables was added to the regression, and the coefficients and significance of these variables were not explored in this study.

4.2. Robustness Check Results

As the lag treatment of the core explanatory variables can alleviate the endogeneity problem caused by mutual causality, this section mainly adopts a robustness test of the core explanatory variables lagging in the first and second periods. In the first and second columns of

Table 3. Results of the robustness check.

	LnPM2.5
	(1)	(2)	(3)	(4)	(5)
L1. LnPoly_a	0.044 ***
	(0.009)
L2. LnPoly_a		0.042 ***
		(0.010)
L1. Poly_b			0.057 *
			(0.031)
L2. Poly_b				0.055 *
				(0.033)
LnCompact					−0.144 ***
					(0.012)
Second	0.005 ***	0.005 ***	0.005 ***	0.005 ***	0.004 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Pfdi	0.010 ***	0.013 ***	0.007 ***	0.010 ***	0.005 **
	(0.002)	(0.002)	(0.002)	(0.002)	(0.002)
Proad	−0.001	−0.001	−0.002 **	−0.003 ***	−0.003 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
LnPgdp	0.338 ***	0.320 ***	0.296 ***	0.268 ***	0.221 *
	(0.087)	(0.083)	(0.078)	(0.073)	(0.130)
(LnPgdp)2	−0.022 ***	−0.021 ***	−0.020 ***	−0.018 ***	−0.012 *
	(0.005)	(0.004)	(0.004)	(0.004)	(0.007)
Capital	0.119 ***	0.112 ***	0.116 ***	0.113 ***	0.180 ***
	(0.021)	(0.022)	(0.021)	(0.023)	(0.019)
Travelcity	0.009	0.008	0.001	0.002	0.021
	(0.016)	(0.017)	(0.015)	(0.016)	(0.014)
LnHistoricalsite	0.149 ***	0.148 ***	0.153 ***	0.147 ***	0.147 ***
	(0.014)	(0.015)	(0.013)	(0.014)	(0.012)
LnSummertemp	2.088 ***	2.054 ***	2.378 ***	2.369 ***	2.118 ***
	(0.104)	(0.108)	(0.094)	(0.099)	(0.093)
LnLatitude	0.716 ***	0.705 ***	0.827 ***	0.825 ***	0.828 ***
	(0.051)	(0.054)	(0.050)	(0.052)	(0.044)
Year FE	YES	YES	YES	YES	YES
Observations	2468	2243	2889	2645	3160
R2	0.385	0.369	0.424	0.417	0.449

Note: ***, **, and * denote the 1%, 5%, and 10% significance levels, respectively; values in brackets below the coefficients are robust standard errors; coefficients on constant terms are not reported because of space limitations.

, the number of urban sub-centers (Poly_a) is treated in lagged periods one and two, respectively. The results show that the numbers of urban sub-centers (Poly_a) in lagging periods one and two were 0.044 and 0.042, respectively, and both were significant at the 1% level. The above results show that with an increase in the number of urban sub-centers, the centralization of the urban spatial structure will cause an improvement in the haze pollution level. In the third and fourth columns of Table 3, the proportion of the urban sub-center population in all central center populations (Poly_b) is lagged and regressed as core explanatory variables. The results show that the coefficient of the proportion of all centers (Poly_b) is still significantly positive, which further supports the regression results obtained in the first and second columns of Table 3. It can be seen that after treating the urban polycentric spatial structure lagging behind, it is still found that adopting the polycentric urban spatial development mode will aggravate the haze pollution, which is consistent with the conclusions obtained in the previous paper, indicating that the benchmark results have strong robustness.

In addition, the urban spatial compactness index is widely used in the quantitative study of urban spatial structure, usually from a macro and qualitative point of view, to measure the compact urban form, and commonly used indexes include the Richardson index, Cole index, Gibbs index, and so on. The higher the urban spatial compactness, the more concentrated the urban population distribution. In this paper, we calculate the ratio of the population in the main center of the city to that in the sub-centers as a measure of urban spatial compactness. Then, we adopt the strategy of replacing the core explanatory variable with the urban spatial compactness index to conduct a robustness test. In the fifth column of Table 3, a series of control variables and year-fixed effects over time are added. The results show that the coefficient of the natural logarithm of urban spatial compactness (LnCompact) is negative and significant at the 1% level. Based on the above results, we found that the compact spatial development of cities is beneficial for reducing PM2.5 concentration, which further supports the benchmark results.

4.3. Endogeneity Treatment Results

Although the previous benchmark model added a series of city-level changes with time control variables, and the characteristics of the city variables do not change with time and control for the year-fixed effect, to identify the causal relationship between urban polycentric spatial structure and haze pollution, it is still difficult to avoid missing variable bias, causing OLS estimator deviation. At the same time, changes in haze pollution may also lead to the transfer of population and economic activities within the city, thereby affecting the urban spatial structure. Therefore, the potential endogeneity problem resulting from missing variables and reverse causation must be addressed. It is generally believed that the instrumental variable method is effective in solving the endogeneity problem. Because instrumental variables must meet the principles of both correlation and exogeneity, it is often difficult to find suitable instrumental variables. According to the method of Ioannides and Zhang (2017) [44], this study collected data from all governments and county governments during the Qing Dynasty from 1644 to 1911 and calculated the number of city walls at the current prefecture level and above as the tool variable of the polycentric spatial structure of the city. Data on the Qing Dynasty city wall were mainly obtained from the Qing Dynasty city wall database compiled by Skinner (1977), which contains information on various dimensions of city walls for more than 1600 cities [45].

Next, we analyze whether the number of Qing Dynasty walls meets the exophytic and relevance principles of instrumental variables. On the one hand, because of its long age, the city wall of the Qing Dynasty will not be affected by the current economic activity variables, and it can rarely be retained up to now; it will not directly affect the haze pollution level in the sample period of this study, so it meets the principle of exogenous nature. On the other hand, prefectures and counties with city walls during the Qing Dynasty often had a certain population size that belonged to the population center of the Qing Dynasty. Related research showed that a higher density of economic activity is associated with the existence of walls in the past [44]. Hence, the larger the number of city walls, the higher the degree of polycentricity. The geographical location and population size of the city have a certain historical continuity; therefore, the number of city walls during the Qing Dynasty is related to the current urban spatial structure, and the instrumental variables meet the principle of correlation. The principles of relevance of instrumental variables are further discussed in the following paragraphs.

Table 4. Results of endogeneity treatment.

	LnPoly_a	LnPM2.5	Poly_b	LnPM2.5
	(1)	(2)	(3)	(4)
	First Stage	Second Stage	First Stage	Second Stage
LnPoly_a		0.445 ***
		(0.042)
Poly_b				1.872 ***
				(0.196)
LnCitywall	0.283 ***		0.065 ***
	(0.018)		(0.005)
Second	−0.004 **	0.005 ***	−0.001 **	0.005 ***
	(0.001)	(0.001)	(0.000)	(0.001)
Pfdi	0.002	0.002	0.002 **	−0.002
	(0.003)	(0.002)	(0.001)	(0.002)
Proad	0.003	0.002	0.003 ***	−0.002
	(0.003)	(0.002)	(0.001)	(0.002)
LnPgdp	−0.345 *	0.106	0.119 **	−0.287
	(0.198)	(0.126)	(0.060)	(0.183)
(LnPgdp)2	0.033 ***	−0.013 **	−0.005	0.012
	(0.010)	(0.007)	(0.003)	(0.009)
Capital	−0.164 ***	0.120 ***	−0.203 ***	0.425 ***
	(0.045)	(0.029)	(0.013)	(0.047)
Travelcity	0.114 ***	−0.054 ***	−0.028 ***	0.042 *
	(0.033)	(0.021)	(0.009)	(0.023)
LnHistoricalsite	−0.041	0.109 ***	−0.039 ***	0.145 ***
	(0.033)	(0.019)	(0.010)	(0.021)
LnSummertemp	−1.098 ***	2.184***	−0.259 ***	2.434 ***
	(0.149)	(0.122)	(0.043)	(0.110)
LnLatitude	−0.217 **	1.062 ***	−0.096 ***	1.247 ***
	(0.086)	(0.063)	(0.025)	(0.061)
Year FE	YES	YES	YES	YES
F-statistic	242.43	——	163.95	——
Observations	2388	2388	2733	2733

Note: ***, **, and * denote the 1%, 5%, and 10% significance levels, respectively; values in brackets below the coefficients are robust standard errors; coefficients on constant terms are not reported because of space limitations.

presents the results of the IV-2SLS regression using the number of Qing Dynasty walls (Citywall) as the instrumental variable. The number of urban sub-centers (Poly_a) in the first and second columns is used to represent the urban polycentric spatial structure. The first column shows the results of the first stage of the 2SLS regression, which shows that the coefficient of the Qing Dynasty wall number (Citywall) is significantly positive at 1%, indicating that the tool variable has a strong correlation with the number of urban sub-centers (Poly_a). The first-stage F-statistic is 242.43, much higher than the empirical value of 10, indicating that the tool variables selected in this study are not weak. The second column reports the second-stage results of the 2SLS regression, showing that the number of urban sub-centers (Poly_a) is 0.445 and significant at the 1% level, which is consistent with the benchmark regression.

The third and fourth columns of Table 4 use the proportion of the entire center population (Poly_b). Among them, the coefficient value of the number of Qing Dynasty walls (Citywall) was 0.065 and significant at 1%, and the first-stage F-statistic was 163.95, which indicates that the instrumental variables were positively correlated with the proportion of the urban sub-center population in all central populations (Poly_b), and there were no weak instrumental variables. In addition, the fourth column shows that the second-stage regression results indicate that the city center population proportion (Poly_b) coefficient is still significant at the 1% level, which further shows that after considering possible endogenous problems, the results still support the original conclusion that the polycentric spatial development mode will improve the level of urban haze pollution.

4.4. Heterogeneity Analysis Results

The regression results of cities that did not consider the heterogeneity of urban rail transit in previous studies show that the polycentric spatial development mode in cities aggravates the level of haze pollution. According to the relevant theoretical analysis in a previous study, the heterogeneous influence of the polycentric spatial structure on haze pollution was investigated from the perspective of urban leapfrog spatial development.

A series of control variables and year-fixed effects are added to each column in

Table 5. Results of heterogeneity analysis.

	LnPM2.5
	(1)	(2)	(3)	(4)	(5)	(6)
LnPoly_a	0.054 ***	0.162 ***	0.053 ***
	(0.010)	(0.028)	(0.011)
Poly_b				0.085 ***	0.219 ***	0.078 ***
				(0.031)	(0.070)	(0.030)
LnPoly_a×Subway	−0.054 ***
	(0.020)
LnPoly_a×LnSubway_road		−0.158 ***
		(0.033)
LnPoly_a×LnSubway_station			−0.042 ***
			(0.007)
Poly_b×Subway				−0.225 ***
				(0.072)
Poly_b×LnSubway_road					−0.202 **
					(0.081)
Poly_b×LnSubway_station						−0.047 **
						(0.021)
Subway	0.080 **			0.152 ***
	(0.035)			(0.035)
LnSubway_road		0.387 ***			0.088 ***
		(0.062)			(0.024)
LnSubway_station			0.097 ***			0.021 ***
			(0.014)			(0.007)
Controls	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	2721	3176	3176	3160	3160	3160
R2	0.378	0.080	0.079	0.418	0.415	0.415

Note: *** and ** denote the 1% and 5% significance levels, respectively; values in brackets below the coefficients are robust standard errors; coefficients on constant terms are not reported because of space limitations.

, where the first to third columns use the number of urban sub-centers (Poly_a). Specifically, the first column in Table 5, on the basis of the benchmark regression, added the dummy variable of whether the city has opened rail transit (Subway) and the interaction term between this dummy variable and urban polycentric spatial structure (Poly_a × Subway). It shows that the coefficient of the interaction term at 1% is significantly negative. It can be seen that public rail transit has a negative adjustment effect on the impact of the urban polycentric spatial structure on haze pollution. For cities that have already built rail transit, the increase in haze pollution caused by the urban polycentric spatial structure is relatively small.

The number of urban rail transit lines (Subway_road) and the interaction term with the urban polycentric spatial structure (Poly_a × Subway_road) are included. The regression results show that the coefficient of the interaction term remains significantly negative at the 1% level. The number of urban rail transit stations (Subway_station) and the interaction term between this variable and the multi-center spatial structure of the city (Poly_a × Subway_station) are added in the third column of Table 5. It can be found that the coefficient of the interaction term is also significantly negative at the level of 1%, which is consistent with the estimated results of the interaction term in the second column. The second and third columns show that the larger the number of urban rail transit lines and stations, the less the role of the polycentric urban spatial structure in improving haze pollution.

Further, the fourth to sixth columns in Table 5 replace the core explanatory variables, using the proportion of the urban sub-center population to the entire central population (poly_b) and the urban polycentric spatial structure; then, using the same strategy as in the first to third columns, we examined the heterogeneity of urban rail transit facility level, namely, whether the urban rail transit open virtual variable (Subway), urban rail transit lines (Subway_road), and the number of urban rail transit stations (Subway_station) in the polycentric spatial structure influence the haze pollution regulation effect. The results showed that the coefficients of the interaction terms in the fourth to sixth columns were significantly negative, indicating that the estimated results remain robust after replacing the polycentric urban spatial structure measure.

In summary, the more developed the urban rail transit and rail transit construction, the lesser the role of the polycentric urban spatial development mode in aggravating haze pollution. Thus, Hypothesis 2 was verified. This result shows that cities with a high level of rail transit facilities can moderately adopt a polycentric spatial development mode to alleviate other “urban diseases,” while cities with a low level of rail transit facilities should implement a monocentric spatial development mode to avoid blind leapfrog development and expansion.

4.5. Mechanism Test Results

As mentioned in a previous study, from the perspective of urban leapfrog spatial development, the urban polycentric spatial development mode mainly affects haze pollution levels by changing the transportation mode of residents and increasing energy consumption [17,23,25]. Based on the relevant mechanism analysis, this section first uses the per capita volume of urban public bus passengers (Pbus) and private car ownership (Car) to measure the travel mode of urban residents and empirically examines the impact of the urban polycentric spatial development mode on the travel modes of residents. On the one hand, the larger the average passenger volume (Pbus), the more residents travel by public transport; on the other hand, owing to the alternative relationship between private cars and public transport, greater private car ownership (Car) in the city means that more residents will travel by private cars, and fewer, by public transport. The data were obtained from the China City Statistical Yearbook and the China National Bureau of Statistics database. Second, because all the CO₂ emissions generated by primary and secondary energy consumption in the city can reflect the energy consumption of cities, this section uses the carbon emissions of urban residents to measure urban energy consumption (Pc) and empirically tests the impact of the urban polycentric spatial development mode on energy consumption.

The results of the influence mechanism test are presented in

Table 6. Results of mechanism test.

	LnPbus	LnPbus	LnCar	LnCar	LnPc	LnPc
	(1)	(2)	(3)	(4)	(5)	(6)
LnPoly_a	−0.199 ***		0.289 ***		0.121 ***
	(0.025)		(0.026)		(0.040)
Poly_b		−0.551 ***		0.696 ***		0.398 ***
		(0.079)		(0.096)		(0.124)
Controls	YES	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES	YES
Observations	2692	3121	532	572	3170	3673
R2	0.364	0.355	0.512	0.435	0.663	0.643

Note: *** denote the 1% significance levels, respectively; values in brackets below the coefficients are robust standard errors; coefficients on constant terms are not reported because of space limitations.

, with a series of control variables and year-fixed effects added to each column. In the first and second columns, the number of urban sub-centers (Poly_a) and the proportion of the urban sub-center population in all centers (Poly_b) were used to measure the urban polycentric spatial structure, and the relationship between the urban polycentric spatial development mode and urban per capita public bus passenger traffic (Pbus) was investigated. The regression results of the first and second columns in Table 6 show that the coefficient of the urban polycentric spatial structure is significantly negative at the 1% level, indicating that the polycentric urban spatial structure will reduce the passenger volume of public transport; in other words, this mode does not encourage residents to adopt public transportation. Second, the third and fourth columns in Table 6 use the number of urban sub-centers (Poly_a) and the proportion of their population in all centers (Poly_b) to represent the urban polycentric spatial structure to discuss the impact of the urban polycentric spatial development mode on urban private car ownership (Car). The regression results in the third and fourth columns show that the coefficient of the urban polycentric spatial structure is significantly positive at the 1% level, indicating that the polycentric spatial development mode in cities can increase private car ownership, implying that residents will be more inclined to travel by private cars. Compared with private cars, public transportation has proven to be a more energy-saving and environment-friendly way of travel; therefore, it is conducive to reducing the level of haze pollution.

Finally, the fifth and sixth columns in Table 6 use the number of urban sub-centers (Poly_a) and the proportion of the urban sub-center population in all centers (Poly_b) as core explanatory variables to study the impact of the urban polycentric spatial development mode on energy consumption (Pc). The regression results in the fifth and sixth columns demonstrate that the coefficient of the urban polycentric spatial structure is significantly positive at the 1% level, indicating that the urban polycentric spatial development mode increases urban energy consumption, further exacerbating the level of urban haze pollution. We found that research hypothesis 1 was verified. Additionally, after further analysis, it was observed that, corresponding to the urban polycentric spatial development mode, the monocentric mode encourages residents to use public transportation and is conducive to saving urban energy consumption, thus alleviating haze pollution.

5. Conclusions and Policy Implications

Based on the dimensions of the leapfrog space, this study conducted research at the city level. In contrast to examining the development mode of vertical or horizontal urban spaces, it focuses on the influence of the urban polycentric spatial development mode on haze pollution. First, this study uses the number of urban sub-centers and the proportion of their population in the total center population to measure the urban polycentric spatial structure and characterize the urban polycentric spatial development mode. The OLS method was used to investigate the influence of the urban polycentric spatial development mode on haze pollution. The number of city walls of the Qing Dynasty was selected as the tool variable for the polycentric urban spatial structure, and the causal relationship between the polycentric urban spatial development mode and haze pollution was more accurately identified using the instrumental variable method. Subsequently, the impact of heterogeneity at the urban rail transit facility level was discussed. Finally, the main mechanism by which the polycentric urban spatial development mode affects haze pollution was tested.

This study draws the following main conclusions: First, the polycentric urban spatial structure will increase urban PM2.5 concentration, suggesting that polycentric spatial development mode in Chinese cities can exacerbate haze pollution. However, a study using European data found that residents in monocentric urban areas face more severe air pollution [46]. The different results of Chinese and European studies deserve more consideration and attention. Second, the more developed the urban rail transit, the smaller the role of the polycentric spatial development mode in worsening haze pollution in cities. Third, reducing public transportation, encouraging private car travel, and increasing urban energy consumption are the main mechanisms for the polycentric spatial development mode to aggravate the haze pollution.

The conclusion of this study provides a basis for controlling haze from the perspective of the urban internal spatial development mode, and we propose the following policy suggestions accordingly. First, in Chinese cities, the rapid development of the urban population will aggravate haze pollution. It is necessary to prevent excessive population dispersion in the process of urban development and strengthen the agglomeration and distribution of urban population around the central region. This strategy may also be applicable in countries with large populations similar to China, such as the United States and India. Second, cities with underdeveloped public rail transit should adopt the single-center development mode, focus more on promoting population and economic activities in the space of agglomeration distribution, and ensure a high level of transportation infrastructure according to its dense public transport network advantage. The moderate polycentric development mode should be adopted and allowed to reach the regional extension of urban public transport development. As far as possible, the excessive population agglomeration crowding effect should be avoided and other aspects of the “urban disease” alleviated. Finally, the supporting facilities required for urban population agglomeration should be established accordingly, e.g., more efficient urban road networks and energy transportation pipelines, to achieve more environment-friendly and efficient emission reduction targets.