Subway Openings and Urban Air Pollution Mitigation: Pathways to Sustainable Development in China

Abstract

Air pollution not only affects urban production and residents’ lives but also threatens human health, and the construction of the subway is considered an important measure to improve urban traffic congestion and air quality. To test whether the construction and opening of subways with high operating costs can effectively improve urban air quality, this paper takes the opening of subways as a quasi-natural experiment. Based on panel data from 282 cities in China from 2014 to 2021, the difference-in-differences method is used to estimate the impact of subway openings on urban air pollution. It is found that (1) the subway opening significantly reduces urban air pollution, and this conclusion remains valid after a series of robustness tests. (2) Mechanism analysis indicates that subway openings mostly mitigate urban air pollution through the traffic substitution effect, which is generated through the substitution of private car travel. (3) Heterogeneity analyses show that the abatement effect of subway openings on air pollution is more significant when the city is a non-Yangtze River Economic Belt city, or when the city’s economic level is lower, or when the city is a non-resource-based city. Therefore, this paper puts forward targeted policy recommendations, such as optimising the subway layout, promoting the transit-oriented development mode based on the subway system, and implementing differentiated development strategies, with a view to providing certain references for promoting sustainable urban development.

1. Introduction

Air pollution is a common challenge in the global urbanisation process, and its effects have transcended geographical boundaries, posing a threat to the health of the worldwide population. According to the World Health Organization [1], 99% of the world’s population lives in areas with substandard air quality, and ambient air pollution is responsible for 4.2 million premature deaths worldwide. Several studies have confirmed that air pollution is a key environmental factor triggering respiratory diseases [2,3]. For example, particulate matter concentrations are highly correlated with the number of hospitalisations for respiratory diseases [4]. For every 1 μg/m³ increase in the mean mass concentrations of 2.5-micrometer particulate matter (PM_2.5), inhalable particles (PM₁₀), nitrogen dioxide (NO₂) and sulphur dioxide (SO₂) in the ambient air, the incidence of respiratory diseases increases by 0.2–1.4%, 0.7–1.6%, 3.7–8.2%, and 0.5–2.3% [5]. According to the China Health Statistics Yearbook 2022 [6], respiratory diseases ranked as the fourth cause of death among urban residents in 2021, with a share of 8.45%. How to improve air pollution and enhance people’s well-being has become a common challenge in global urban governance.

As the world’s largest developing country, China’s experience in air pollution control is an important reference for emerging economies. The Chinese government has contributed to the improvement of air pollution in China through a series of measures such as the formulation of the Action Plan for Prevention and Control of Air Pollution [7] and the Three-Year Action Plan for Winning the Battle for the Blue Sky [8], and the requirement to “strengthen the synergistic control of pollutants” [9]. According to the 2021 China Ecological Environment Status Bulletin [10], the air quality compliance rate for 339 cities at the prefecture level and above nationwide reached 64.3%, an improvement of 3.5 percentage points from 2020. However, pollution levels, including PM_2.5 and ozone (O₃), remained high at 30 μg/m³ and 137 μg/m³, respectively. This far exceeds the air quality guideline values set by the World Health Organisation [1], where the safety threshold for the annual average concentration of PM_2.5 is 5 μg/m³, and the threshold for the maximum 8-h daily average concentration of O₃ is 100 μg/m³. Pollution control still has a long way to go.

Among the sources of urban air pollution, emissions from fast-growing vehicles are especially concerning. In Europe, nitrogen oxides (NO_x) and PM₁₀ are strongly influenced by road traffic, which contributes 73% of NO_x emissions and 42% of PM₁₀ emissions in Paris and 50% of NO_x and PM₁₀ emissions in London [11]. In Africa, vehicles are the largest source of urban outdoor air pollution [12]. In Asia, particulate matter concentrations in South Korea are mainly influenced by traffic [13], and vehicle emissions are one of the important sources of air pollution in China [14]. Data from the Ministry of Public Security shows that China’s vehicle fleet will reach 395 million in 2021, with 79 cities exceeding the one-million mark, ranking first globally in scale [15]. Therefore, reducing vehicle exhaust pollution is an urgent problem.

As a high-capacity, low-emission green transport infrastructure, the subway is regarded as a “double key” to solving urban traffic congestion and air pollution [16]. By replacing private car travel, the subway not only directly reduces transport carbon emissions [17], improves air quality, reduces the incidence of diseases caused by air pollution [18] and social costs [19,20], but also promotes compact development by optimising the spatial layout of cities [21]. The Chinese government attaches great importance to the construction of subway networks. As shown in Figure 1. The left y-axis indicates the length of new operating lines of China’s subway (km), and the right y-axis indicates the total length of China’s subway operating lines (km). From 2014 to 2021, the total length of subway operating lines in China increased from 2361 km to 7209.7 km. By 2021, the number of cities with subway lines in operation had reached 42, forming the world’s largest urban rail transit system. However, the high cost of subway construction has led to an ongoing debate about its environmental benefits. Does the opening of a subway actually reduce air pollution by changing the travel structure? Does this effect vary across cities with different geographic locations, levels of development, and resource endowments? Answering these questions is of great practical significance for optimising transport infrastructure investment decisions and promoting sustainable urban development.

To answer the above questions, this paper employs the difference-in-differences (DID) model to systematically assess the impact of subway openings on urban air pollution, to deeply analyse the path of its action, and to explore the possible influence of urban heterogeneity on this effect. The potential marginal contributions of this paper are as follows: firstly, this research empirically examines the effect of subway openings on urban air pollution by utilising data from 282 prefecture-level cities in China, spanning from 2014 to 2021, through the DID methodology. On this basis, slope data is used as an instrumental variable for subway openings to effectively control the endogeneity problem of metro opening and obtain more robust research results. Secondly, combined with the existing literature, this study summarises the logical framework of the effect of subway openings on air pollution and investigates the mechanism of this effect. Thirdly, the effects of regional heterogeneity, economic level heterogeneity, and urban heterogeneity are further discussed with the aim of exploring the reasons for the possible differential impact of the air pollution improvement effect of subway openings.

2. Literature Review

A large number of studies have estimated the impact of subways on housing markets [22,23,24], urban economic activity [25,26,27] and the environment [28,29,30], and the factors influencing air pollution, such as economic growth [31,32], transport infrastructure [33,34] and environmental regulatory policies [35,36].

Focusing on the environmental effects of the opening of the subway, researchers at home and abroad have widely discussed the impact of subway openings on air pollution. However, no consensus has been reached so far. Most researchers believed that the opening of the subway, i.e., the first opening of a new subway line in operation, significantly reduced air pollution. Chen and Whalley [19] studied the effect of the opening of the Taipei Metro on air pollution based on a discontinuity regression design (RDD) and found that the opening of the subway could reduce the concentration of carbon monoxide (CO). Using daily data from 29 Chinese cities from 2013 to 2018, Lu et al. [17] used the RDD and found that the newly opened subway significantly reduced PM_2.5 concentration. Li et al. [37] found that opening 14 new underground lines in Beijing could significantly improve air quality, with the extent of improvement correlating positively with the density of the station network. Zheng et al. [38] discovered that the opening of the Changsha Metro in Hunan Province significantly reduced CO concentrations while exerting minimal influence on O₃ concentrations. Xiao et al. [39] found that urban air quality significantly improved after the opening of the metro, based on daily data from 2000 to 2012 for prefecture-level cities in China. Li et al. [40] employed the sharp regression discontinuity method to assess the impact of 112 newly opened subway lines in 26 Chinese cities on air quality and found that the opening of the subway resulted in a significant reduction in PM_2.5 and NO₂ concentrations. Leirião et al. [41] used an interrupted regression model and found that the opening of the metro improved the local air quality in the city of São Paulo. Xie et al. [20] used a machine learning approach to study the impact of subway line openings on air quality in China between 2013 and 2018, greatly improving the precision of the estimates, and they found that subway openings improved air quality.

However, some researchers also believed that the opening of the subway did not have a significant impact on reducing air pollution. O’Toole [42] argued that new rail transport reduces social welfare and may produce more greenhouse gases. Beaudoin et al. [43] reviewed relevant studies since 2000 and concluded that whether subway opening improves air quality is still open to discussion. Sun et al. [44] conducted a study based on quarterly data from 28 subway-opening cities in China from 2013–2016 and found that subway opening exacerbates air pollution in the short term. Rivers et al. [45] explored the impact of traffic strikes on air quality in 18 Canadian cities. They found that traffic stoppages did not affect levels of carbon monoxide or fine particulate matter concentrations. Gendron-Carrier et al. [46] used a multi-period DID model to analyse 58 cities around the world with subway openings between February 2000 and December 2014 and concluded that, on average, particulate matter concentrations do not change as a result of subway openings.

At present, more researchers have conducted a lot of research and discussion on the opening of the subway and air pollution, respectively. This provides an important reference value for subsequent research, but there is still some room for improvement. Firstly, most of the existing researchers used a single city [37,38] or several specific cities [40,45] as samples for their studies. Although this can explain the impact of subway openings on air pollution in specific cities, whether the findings can be generalised to all cities in China requires further research and discussion. Secondly, some of the literature uses regression discontinuity [17,19,40] to study the impact of metro opening on urban air pollution. The window periods of such studies are generally short, making it difficult to examine the medium- to long-term environmental effects of metro opening. RDD estimation results are sensitive to bandwidth selection, and it may be difficult to exclude confounding factors near the breakpoints, whereas the use of DID mitigates such problems by designing a control group. Finally, existing relevant studies have yet to reach a consensus on the impact of the opening of the subway on air pollution. This is based on the fact that, on the one hand, the opening of the subway can reduce air pollution emissions through the traffic substitution effect [19], and on the other hand, it can create more traffic demand and aggravate air pollution [47].

Therefore, based on the panel data of prefecture-level cities in China from 2014 to 2021, this paper empirically examines the impact of subway openings (i.e., new subway lines opening and operating for the first time) on urban air pollution, and introduces the slope data as an instrumental variable for the endogeneity issue to control for more credible findings. The study not only makes up for the lack of empirical evidence at the national level but also provides a reference for optimising subway layout policies, promoting synergistic development of transport and environment, and promoting sustainable urban development by discussing the mechanism of action and heterogeneous impacts.

3. Theoretical Analysis and Research Hypothesis

This section synthesises the current literature to outline the logical framework regarding the impact of subway openings on air pollution and to examine the mechanisms via which subway openings influence urban air quality (Figure 2).

3.1. The Effect of Subway Opening on Urban Air Pollution

As one of the backbone systems of passenger transport, the urban rail transit system occupies a dominant position in the urban transport system due to its powerful functions and transport advantages. Among them, the subway, as an important part of the urban rail transport system, has the advantages of high passenger capacity, punctuality, energy saving, safety, and comfort, etc., which has made a significant contribution to alleviating urban traffic congestion and air pollution [16].

However, there is no academic consensus on the relationship between rail transport and air impacts [43]. It has been argued that the opening of the subway significantly reduces urban air pollution and improves air quality [19,40]. Others argue that the opening of the subway creates new commuting demands and increases urban air pollution [42]. Whether the opening of the subway can significantly reduce air pollution depends on the total effect of the traffic substitution effect and the traffic creation effect, which act in opposite directions. Most of the views believe that the substitution effect of the opening of the subway exceeds the creation effect [20,39,44]. Accordingly, this paper proposes Hypothesis 1.

 Hypothesis 1: 

The opening of the subway significantly reduces the concentration of air pollutants and improves urban air quality.

3.2. Mechanisms of the Impact of Subway Opening on Urban Air Pollution

As a low-pollution, high-capacity urban rail transport system, the subway is regarded as an important management tool for relieving urban traffic congestion and air pollution. Currently, there are two contrasting mechanisms in academia regarding the impact of subway openings on air pollution: one is the traffic substitution effect, and the other is the traffic creation effect.

The transport substitution effect suggests that the opening of the subway replaces people’s previous highly polluting road-based travel modes and reduces motor vehicle emissions, thereby reducing air pollution concentrations and improving air quality [19]. On the one hand, the opening of the subway provides people with a fast and safe way to travel by public transport. This reduces the commuting cost and replaces people’s previous non-public transport travel mode, thus reducing motor vehicle exhaust emissions and air pollution [39,48]. Huang et al. [48] revealed that the opening of the Xi’an subway significantly reduced car and e-bike ownership among nearby residents, which in turn had an emission reduction effect. Xiao et al. [39] found that the mechanism by which the opening of the subway improves air quality is the substitution effect of the subway for taxis. On the other hand, the opening of the subway has increased the supply of transport infrastructure, optimised the urban transport system, and enriched people’s mode of travel. This can, to some extent, reduce people’s reliance on non-public transport modes of travel such as private cars and taxis, thereby reducing traffic congestion and air pollution [16]. Mei et al. [49] showed that under a transit-oriented development (TOD) model based on the subway system, low-carbon connections such as walking and shared bikes replace high-carbon connections such as private cars, significantly reducing carbon emissions. Therefore, this paper proposes Hypothesis 2.

 Hypothesis 2: 

The opening of the subway replaces some type of traditional road travel and reduces urban air pollution levels.

The transport creation effect challenges the transport substitution effect. It is argued that the substitution effect of the opening of the subway would be limited and would create more traffic demand, increasing commuter stress and air pollution levels [47]. The Brexit paradox states that adding a new line to a transport network may lead to longer travel times [50]. On the one hand, new urban railways make surrounding land prices rise [51], commercial activities increase [52], and housing prices rise [24]. The increase in housing costs leads people to move to the suburbs, generating higher commuting demand with longer commuting distances, counteracting the transport substitution effect, which in turn aggravates urban air pollution levels [53]. On the other hand, as the opening of the subway replaces some of the travelling modes, such as private cars and taxis, it improves the road travel conditions [16]. This makes some people choose private vehicles to travel again, the use of vehicles on the road will not be significantly reduced [54], and may aggravate air pollution. In addition, convenient transport can drive the city’s economic development [55], achieving green and innovative development of the city [56,57]. This will draw the foreign population to move in, increasing people’s activities, causing more commuting pressure and leading to a decrease in air quality. Therefore, this paper proposes Hypothesis 3.

 Hypothesis 3: 

The opening of the subway creates more commuting demand and increases urban air pollution.

4. Methods

4.1. Model Set-Up

To examine the impact of the subway openings on urban air pollution, this paper quantitatively analyses the changes in urban air pollution before and after the opening and operation of the subway lines using a DID model. The baseline regression model is as follows:

$${AQI}_{it}={\beta }_0+{\beta }_1{subway}_{it}+\gamma C_{it}+{\mu }_i+{\nu }_t+{\epsilon }_{it}$$

(1)

where i represents city, t denotes year. AQI represents the air quality index of a city. Subway represents whether or not the city has a subway. C denotes the set of control variables containing variables affecting urban air quality other than the opening of the subway. ${\mu }_i$ denotes the city fixed effect (time FE) that controls for city-level characteristics that do not vary over time, such as geographic location. ${\nu }_t$ denotes the time fixed effect (city FE), which is used to control for time trends common to all cities, such as macroeconomic fluctuations, uniform adjustment of environmental policies, etc. ${\epsilon }_{it}$ represents the error term.

${\beta }_0$, ${\beta }_1$ and $\gamma$ are parameters to be estimated, of which the most critical parameter to be estimated is ${\beta }_1$. It measures the impact of the opening of the subway on urban air pollution. If ${\beta }_1$ is significantly greater than 0, it indicates that the opening of the subway has increased urban air pollution. If ${\beta }_1$ is significantly less than 0, it indicates that the opening of the subway has reduced urban air pollution. If ${\beta }_1$ is not significant, it indicates that the opening of the subway has no significant impact on urban air pollution.

4.2. Variable Selection

${AQI}_{it}$ is the explained variable of this study, denoting the air quality index of city i in year t. A higher air quality index (AQI) indicates more severe air pollution. Since the Ministry of Environmental Protection issued the announcement of the Technical Regulations on Ambient Air Quality Index (on trial) [58], it is stipulated that the official publication of the air quality index (AQI) will replace the air pollution index (API), which was the air pollution index before 2012, from 2013 onwards. Most of the existing studies on air quality select the AQI after 2013 or the API before 2013 separately. Due to the limited number of cities that reported the AQI in 2013, and in light of data completeness and availability, this study encompasses the period from 2014 to 2021. The AQI is incorporated into the model after applying the logarithm to the AQI.

${subway}_{it}$ is the core explanatory variable, indicating whether city i has a subway line in operation in year t. The opening of the subway in the first half of the year (30 June and before) is classified as occurring in that year, while an opening in the second half of the year is classified as occurring in the subsequent year. If city i opens the subway in year t, then it takes the value of 1, and vice versa is 0. The experimental group in this paper includes 27 cities that opened subways between 2014 and 2021, removing 15 cities that opened subways before 2014. The control group consists of cities in China that do not have subways, totalling a sample of 282 cities (Hong Kong, Macau and Taiwan are excluded for data availability reasons).

Established literature suggests that the level of urban economic development, industrial structure, population density, and natural climatic factors are important factors affecting the level of urban air pollution [59]. Therefore, the following control variables are included in this paper: (1) Gross city product (lnGDP), measured by the city’s gross domestic product. The economic status of a city influences its transportation infrastructure, which subsequently impacts the city’s air quality. (2) Industrial structure (industrial), measured by the proportion of the city’s gross domestic product accounted for by the secondary industry. The higher the proportion of secondary industry, the more likely it is to affect urban air quality. (3) Population density (lnPD), measured by the number of people per square kilometer of area. Population activity influences transportation congestion, which then impacts air quality. (4) Average annual temperature (lnTEM), measured as the average of daily temperatures in the city. (5) Average annual precipitation (lnrain), measured as the average of the city’s daily precipitation. (6) Average annual wind speed (lnwind), measured as the average of the city’s daily wind speed. The above three natural climatic factors may influence pollution deposition and, consequently, air quality.

4.3. Data Description

This paper utilises a sample of 282 cities in China from 2014 to 2021. The explanatory variable AQI is from the China Air Quality Historical Data Network. The core explanatory variable subway is from Baidu Encyclopedia. The control variables gross city product, industrial structure and population density are from the China Urban Statistical Yearbook and the EPS database of past years. Urban meteorological data, including annual average temperature, rainfall, and wind speed, are from the National Meteorological Science Data Sharing Service Platform.

Table 1. Results of descriptive statistics.

Variable	Variable Meaning	Obs	Mean	Std.	Min	Max
lnAQI	Annual Mean of City’s AQI	2225	4.313	0.272	3.322	5.167
subway	Dummy variable for whether the city opens a subway or not	2225	0.054	0.227	0	1
lnGDP	Gross domestic product of the city	2225	7.337	0.866	4.424	9.588
industrial	Industrial structure	2225	42.760	10.340	8.125	84.910
lnPD	City population density	2225	5.593	1.104	−0.959	7.882
lntem	Average annual temperature	2225	2.588	0.487	−1.469	3.255
lnrain	Average annual precipitation	2225	6.795	0.623	3.796	8.046
lnwind	Average annual wind speed	2225	0.855	0.344	0.089	2.104

gives the descriptive statistics of the basic variables.

5. Results and Discussion

5.1. Baseline Regression Analysis

Table 2. The impact of subway openings on urban air pollution: baseline regression results.

Variable	(1)	(2)	(3)	(4)	(5)	(6)	(7)
subway	−0.025 *	−0.034 **	−0.036 **	−0.037 ***	−0.037 **	−0.040 ***	−0.042 ***
	(0.015)	(0.014)	(0.014)	(0.014)	(0.014)	(0.014)	(0.014)
lnGDP		0.144 ***	0.121 ***	0.123 ***	0.124 ***	0.101 ***	0.107 ***
		(0.014)	(0.016)	(0.016)	(0.016)	(0.016)	(0.016)
industrial			0.002 ***	0.002 ***	0.002 ***	0.002 ***	0.001 ***
			(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
lnPD				−0.041	−0.037	−0.041	−0.029
				(0.025)	(0.025)	(0.025)	(0.025)
lntem					0.108 **	0.099 **	0.104 **
					(0.043)	(0.042)	(0.042)
lnrain						−0.081 ***	−0.086 ***
						(0.011)	(0.011)
lnwind							−0.144 ***
							(0.026)
Time FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	2225	2225	2225	2225	2225	2225	2225
R-squared	0.542	0.565	0.568	0.568	0.570	0.582	0.588

Note: *** p < 0.01, ** p < 0.05, * p < 0.1, standard errors in parentheses.

presents the baseline regression results. Column (1) presents the regression results without the inclusion of control variables, accounting just for time-fixed and city-fixed effects. The coefficient of the explanatory variable subway is significantly negative, generally in line with the paper’s anticipation that the opening of the subway reduces urban air pollution. Columns (2) to (7) present the regression results after adding the six control variables sequentially. The coefficient of the explanatory variable subway is −0.042 and passes the significance test at the 1% level. This shows that, all other things being equal, the opening of the subway reduced the city’s annual average AQI by 4.2%, compared to the city without the subway. The opening of the subway has a significant inhibiting effect on the rise of the air quality index, significantly reducing the level of air pollution in the city and improving urban air quality. This may be because the opening of the subway supplants certain vehicle transport modes and partially alleviates urban traffic congestion. It efficiently diminishes motor vehicle exhaust emissions, hence decreasing air pollution concentration and improving air quality. Therefore, Hypothesis 1 is verified: the opening of the subway can significantly reduce the concentration of air pollutants and improve urban air quality.

5.2. Robustness Tests

5.2.1. Parallel Trend Test

The normative DID model needs to satisfy the premise of the parallel trend assumption, i.e., the experimental and control groups have the same trend before the policy intervention. Referring to Beck et al. [60], to verify whether the experimental and control groups have the same trend of change in air pollution before the opening of the subway, the model is set as follows:

$${AQI}_{it}={\alpha }_0+\sum _{k=-5^{+},k\ne -1}^{k=4^{+}}{\alpha }_k{subway}_{it}^k+\gamma C_{it}+{\mu }_i+{\nu }_t+{\epsilon }_{it}$$

(2)

where k denotes the number of years between time t and the year in which the city subway first opened for operation, taking values of −5+, −4, −3, −2, 0, 1, ..., 4+. Negative, 0 and positive values indicate the year before, the year of, and the year after the opening of the subway, respectively. −5+ denotes the 5th year before the opening of the subway and before, and 4+ denotes the 4th year after the opening of the subway and after. The dummy variable ${subway}_{it}^k$ denotes the policy shock of opening a subway in the city. In order to avoid the covariance of the time dummy variables, the dummy variable ${subway}_{it}^{-1}$ in the year before the opening of the subway is not put into the model as the base period. The definitions of other variables and symbols in Equation (2) are the same as those in Equation (1).

The results of the parallel trend test are shown in Figure 3. The horizontal axis denotes the year of the policy shock, current denotes the year the subway opened, prek denotes the kth year before the subway opened, and postk denotes the kth year after the subway opened. The vertical axis is the estimated coefficients, the hollow origin in the figure indicates the policy effect of the subway opening in a city in a given year, and the dotted line indicates the 90% confidence interval. It can be seen that the regression coefficients are not significant in all periods before the implementation of the policy, indicating that there is no significant difference between the air pollution of the experimental group and the control group before the opening of the subway, i.e., the DID model passed the parallel trend test.

5.2.2. Placebo Test

To verify whether the results of the baseline regression are affected by other random factors, this paper uses random sampling for the placebo test [18], and the cities that open subways and the opening time of the subway are randomly set and then put into the model for regression. The results of the regression based on 500 random samples are shown in Figure 4. The circles located at the bottom of the figure are the (estimated coefficients, p-values) combination points, the dashed line is the baseline regression results for the real policy, and the solid line is the kernel density estimate of the regression coefficients of the DID term after randomising the treatment group. The estimated coefficients obtained based on the random sample have a mean of 0, follow an approximately positive normal distribution, and are significantly different from the baseline regression results of the real policy. Most of the scatters lie above the dotted line with a p-value of 0.1, indicating that they are not significant at the 10% level. Thus, the placebo test is passed, and other random factors can be excluded from interfering with the results, reflecting the robustness of the baseline regression findings.

5.2.3. Counterfactual Test

In addition to the shock of the opening of the subway in the city, some other policies or unobserved factors may also affect urban air quality. However, this impact is not associated with the opening of the subway, which in turn leads to the conclusion that the previous baseline regression is not valid. Therefore, this paper conducts a counterfactual test by changing the timing of the policy shock, the opening of the city’s subway. If the regression result is significantly negative, it indicates that the improvement of urban air quality may stem from the impact of other policies rather than the opening of the subway. If the regression result is not significantly negative, it indicates that the urban air quality improvement originates from opening the subway. The regression results of the counterfactual test are shown in

Table 3. Regression results of varying the timing of policy shocks and excluding outlier cities: robustness test.

Variable	(1)	(2)	(3)
subway_bf2	−0.024
	(0.016)
subway_bf3		0.002
		(0.017)
subway			−0.037 ***
			(0.014)
lnGDP	0.106 ***	0.105 ***	0.101 ***
	(0.016)	(0.016)	(0.016)
industrial	0.001 ***	0.001 ***	0.001 **
	(0.001)	(0.001)	(0.001)
lnPD	−0.027	−0.026	−0.005
	(0.025)	(0.025)	(0.030)
lntem	0.104 **	0.105 **	0.096 **
	(0.042)	(0.042)	(0.042)
lnrain	−0.085 ***	−0.084 ***	−0.084 ***
	(0.011)	(0.011)	(0.011)
lnwind	−0.144 ***	−0.142 ***	−0.135 ***
	(0.026)	(0.026)	(0.027)
Time FE	Yes	Yes	Yes
City FE	Yes	Yes	Yes
Observations	2225	2225	2104
R-squared	0.587	0.586	0.608

Note: *** p < 0.01, ** p < 0.05, standard errors in parentheses.

. Column (1) shows the regression results for the opening of the subway 2 years in advance, and column (2) shows the regression results for the opening of the subway 3 years in advance. The coefficients of the explanatory variables subway_bf2 and subway_bf3 are not significant, which suggests that the improvement of urban air quality is not caused by other factors but originates from the opening of the subway, verifying the robustness of the baseline regression results.

5.2.4. Exclusion of Outlier Cities

As this article removes the 15 cities that opened their subways before 2014, it includes Beijing, Shanghai, Guangzhou and Shenzhen. This has excluded the four first-tier cities in China, avoiding any impact on the reliability of the baseline regression results due to the large differences between these cities and the others. To further enhance the comparability between the experimental and control groups, 17 cities within the provinces (Hainan Province, Tibet Autonomous Region, Qinghai Province, and Ningxia Hui Autonomous Region) that did not have subways during the period 2014 to 2021 have been deleted. The results are shown in column (3) of Table 3. After the removal of the 17 outlier cities, the coefficient of the subway is still significantly negative and is pretty much the same as the coefficient in Table 2. This means that the conclusions of the baseline regression in this paper are still valid.

5.2.5. Alteration of Air Pollution Proxy Variables

In previous studies, this paper used the air quality index as a comprehensive indicator of air pollution levels. This method can objectively reflect the overall level of air quality by integrating the concentration information of multiple pollutants, and has been widely used in environmental effect studies [61,62]. However, relying on a single composite indicator, AQI, may not be comprehensive enough in assessing the impact of subway opening on air pollution. Based on this, this paper will further refine the dimensions of the analysis by taking six air pollution indicators, namely nitrogen dioxide (lnNO₂), sulphur dioxide (lnSO₂), carbon monoxide (lnCO), ozone 8-hr concentration (lnO₃), inhalable particles (lnPM₁₀), and 2.5-micrometer particulate matter (lnPM_2.5), as the explained variables to conduct robustness tests, and also to investigate the impact of subway opening on different pollutants.

The regression results are presented in

Table 4. The impact of subway openings on six pollutants: robustness test.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	lnNO2	lnSO2	lnCO	lnO3	lnPM10	lnPM2.5
subway	−0.057 ***	−0.089 **	0.021	0.032 **	−0.059 ***	−0.046 **
	(0.020)	(0.045)	(0.024)	(0.016)	(0.019)	(0.021)
lnGDP	0.070 ***	0.085 *	0.068 **	0.049 ***	0.049 **	0.065 ***
	(0.022)	(0.051)	(0.027)	(0.018)	(0.022)	(0.024)
industrial	0.003 ***	−0.001	−0.000	−0.001	0.001 *	0.001
	(0.001)	(0.002)	(0.001)	(0.001)	(0.001)	(0.001)
lnPD	−0.063 *	−0.081	0.050	−0.058 **	−0.106 ***	−0.098 ***
	(0.033)	(0.075)	(0.040)	(0.027)	(0.032)	(0.035)
lntem	−0.149 ***	0.114	0.161 **	0.153 ***	0.112 **	0.065
	(0.058)	(0.131)	(0.070)	(0.046)	(0.056)	(0.062)
lnrain	−0.026 *	−0.122 ***	0.002	−0.045 ***	−0.097 ***	−0.067 ***
	(0.015)	(0.034)	(0.018)	(0.012)	(0.015)	(0.016)
lnwind	−0.032	0.017	−0.046	−0.063 **	−0.012	−0.060
	(0.037)	(0.084)	(0.045)	(0.030)	(0.036)	(0.039)
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	2072	2072	2072	2072	2072	2072
R-squared	0.412	0.730	0.636	0.289	0.686	0.727

Note: *** p < 0.01, ** p < 0.05, * p < 0.1, standard errors in parentheses.

. Columns (1), (2), (5), and (6) demonstrate that the coefficients of the subway are significantly negative, indicating that the opening of the subway significantly reduces the NO₂, SO₂, PM₁₀ and PM_2.5 concentrations, and improves the air pollution level in the city. The coefficient of the subway in column (4) is significantly positive, indicating that the opening of the subway significantly contributes to the production of O₃. This may be related to the secondary pollutant properties of O₃. Therefore, these regression results align closely with the baseline regression result, verifying the robustness of the baseline regression result.

5.3. Endogeneity Test

The results of the previous baseline regression and the robustness test show that the air pollution reduction effect from the opening of the subway is robust. However, the construction planning of the subway may be affected by the air pollution situation, and the endogeneity problem of reverse causality may occur, leading to biased parameter estimates. In addition, the inability to control for omitted variables that simultaneously affect the opening of the subway and urban air pollution may also have an impact on the parameter estimates.

Therefore, the instrumental variable approach will be used to mitigate the endogeneity problem. Suitable instrumental variables should satisfy both high correlation with endogenous variables and ensure sufficient exogeneity, i.e., the instrumental variables affect the explained variables only through endogenous variables. This paper selects the interaction term between slope data and time trend as an instrumental variable, based on the following considerations. Firstly, the geographical gradient is a composite of the topographic variation in a given area and can indirectly measure the cost of opening a subway. Since the slope directly affects subway opening, the correlation between the effective instrumental variable and the endogenous variable can be satisfied. Secondly, slope, as one of the indicators of urban topographic variation, is an inherent geographic information variable of the city. It has a more natural exogenous characteristic, which well satisfies the exogenous assumption of effective instrumental variables.

Since the slope data is a cross-section data of the urban interface, both the endogenous and explanatory variables are panel data that contain information about the city and time. The effect of slope data on subway openings may differ because of different years. Therefore, this paper adopts ArcGIS version 10.2 software to calculate the geographic slope data of each region, and the interaction term between slope data and time trend is used as an instrumental variable. The two-stage least squares method (2SLS) is used to estimate the model. The results are shown in

Table 5. The impact of subway openings on urban air pollution using an instrumental variable approach: endogeneity test.

	Explained Variable: lnAQI
VARIABLES	IV_2SLS
subway	−0.397 **
	(0.165)
lnGDP	0.101 ***
	(0.026)
industrial	0.003 ***
	(0.001)
lnPD	0.142 ***
	(0.007)
lntem	−0.092 ***
	(0.015)
lnrain	−0.277 ***
	(0.009)
lnwind	−0.168 ***
	(0.022)
Time-fixed effect	Yes
City-fixed effect	Yes
LM statistic	23.806
	(0.000)
Cragg-Donald Wald F statistic	28.11
Stock-Yogo weak ID test: 10% maximal IV size	16.38
Observations	1963
R-squared	0.564

Note: *** p < 0.01, ** p < 0.05, standard errors in parentheses.

.

The p-value of the Kleibergen-Paap rk LM statistic is less than 0.01, which means that the original hypothesis of “insufficient identification of instrumental variables” is rejected at the 1% level. The selection of instrumental variables is reasonable and valid. In the first stage estimation results, the Cragg-Donald Wald F statistic is greater than the standard value at the 10% level (28.11 > 16.38), indicating that there is no weak instrumental variable problem. The coefficients of the explanatory variables are negative at the 10% level of significance, which is consistent in direction with the baseline regression results, verifying that there is a significant improvement effect of subway opening on urban air pollution.

5.4. Mechanism Test

At present, there are two opposite mechanisms in the academic community for the impact of the subway opening on air pollution: one is the traffic substitution effect, and the other is the traffic creation effect. The traffic substitution effect is that the opening of the subway replaces people’s traditional road-travelling modes, such as private cars and taxis, and reduces the number of motor vehicles, which in turn reduces motor vehicle emissions and air pollution [19]. The traffic creation effect suggests that the subway’s substitution effect is limited and that it generates more traffic demand, increasing commuting pressure and air pollution levels [47]. Therefore, to investigate the mechanism through which the opening of the subway influences air pollution levels, this paper replaces AQI with the ownership of private cars (lncar), buses (lnbus), and taxis (lntaxi) (in logarithmic terms), together with the growth rates of these vehicle categories (rcar, rbus, rtaxi). The data on bus and taxi ownership were obtained from the China Urban Statistical Yearbook, and the data on private car ownership were obtained from local statistical bureaus.

Table 6. The impact of subway openings on vehicle ownership and growth rates: mechanism test.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	lncar	lnbus	lntaxi	rcar	rbus	rtaxi
subway	−0.065 ***	−0.027	0.061	1.267	6.433	3.343
	(0.020)	(0.048)	(0.044)	(1.377)	(41.161)	(9.957)
lnGDP	0.259 ***	0.312 ***	0.243 ***	−1.046	−29.664	4.219
	(0.026)	(0.054)	(0.051)	(1.729)	(49.149)	(11.889)
industrial	−0.004 ***	0.002	−0.002	0.017	0.615	0.161
	(0.001)	(0.002)	(0.002)	(0.056)	(1.478)	(0.358)
lnPD	0.043	0.160 **	−0.146 **	3.023	−5.808	0.340
	(0.045)	(0.076)	(0.071)	(2.867)	(62.200)	(15.046)
lntem	0.255 **	0.086	0.077	15.152 *	56.745	21.118
	(0.128)	(0.143)	(0.134)	(8.304)	(133.499)	(32.293)
lnrain	−0.029 *	0.023	0.140 ***	−0.749	−11.012	0.329
	(0.016)	(0.036)	(0.034)	(1.002)	(29.170)	(7.056)
lnwind	0.067 *	0.027	−0.058	−0.512	−17.451	−16.625
	(0.039)	(0.089)	(0.083)	(2.688)	(76.331)	(18.464)
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	1726	2233	2233	1498	1953	1953
R-squared	0.901	0.154	0.037	0.317	0.005	0.006

Note: *** p < 0.01, ** p < 0.05, * p < 0.1, standard errors in parentheses.

presents the results of the mechanism test. Column (1) shows that the coefficient of the explanatory variables is −0.065, which passes the test of significance at the 1% level, while columns (2) and (3) show that the coefficients lack significance. This suggests that the opening of the subway has led to a significant decline in private car ownership but has no significant effect on bus and taxi ownership, which shows that the opening of the subway has only had a substitution effect on private cars. This may be attributed to the subway offering a rapid and secure mode of public transportation, which leads to some commuters switching from using private cars to using the subway [63], resulting in a decrease in the number of private cars used. The decrease in private car use directly reduces motor vehicle emissions, which are the main source of NO_x and particulate matter, which corroborates the previous robustness test that “the opening of the subway significantly reduces the concentration of NO_x and particulate matter”. However, the number of buses and taxis is not significantly affected, reflecting the complementary functions and rigidity of demand for different modes of transport. As the “capillary” of urban transport, buses can cover areas not covered by the subway, and their demand is rigid. It still plays an irreplaceable and important role in meeting people’s daily travelling needs. Taxis, on the other hand, mainly cater for immediate travelling needs, with a low degree of overlap with the subway service group, which is difficult to replace with the opening of the subway. Consequently, the opening of the subway mitigates air pollution through the substitution of private cars. Thus, Hypothesis 2 is confirmed: the opening of the subway supplants certain conventional road transportation methods and decreases urban air pollution levels.

The coefficients of the explanatory variables in columns (4)–(6) are not significant, indicating that the opening of the subway does not significantly affect the growth rate of private cars, buses, and taxis. This may be attributed to the current situation where most cities in China with operational subways are tiny and lack comprehensive transit routes throughout the city. In addition, as of 2021, a total of 42 cities in mainland China have opened subways, the subway network is relatively small, and has not yet formed a “network effect”. Therefore, it is difficult to have a significant impact on people’s travelling patterns and cannot significantly influence the growth rate of vehicle ownership. Therefore, it is not possible to test Hypothesis 3: the opening of the subway creates more demand for commuting and exacerbates urban air pollution. In summary, the improvement effect of the subway opening on urban air pollution stems mainly from the dominance of the traffic substitution effect, while the influence of the traffic creation effect is relatively limited.

5.5. Heterogeneity Test

5.5.1. Regional Heterogeneity

The Yangtze River Economic Belt spans 11 provinces and cities in eastern, central and western China. Relying on the Yangtze River as a golden waterway, cities along the river (the Yangtze River Economic Belt cities) have a favourable geographic location, and are often different from other cities in terms of factor resources and governance modes. This may make the opening of the subway have a differential effect on air pollution levels. Therefore, this paper divides the cities into Yangtze River Economic Belt cities (YEB cities) and non-YEB cities (non-YEB cities) for the heterogeneity test. The regression results are shown in column (1) and column (2) of

Table 7. The impact of subway openings on different types of cities: heterogeneity test.

	(1)	(2)	(3)	(4)	(5)	(6)
Variable	Yangtze River Economic Belt Cities	Non-Yangtze River Economic Belt Cities	Low Economic Level Cities	High Economic Level Cities	Resource-Based Cities	Non-Resource-Based Cities
subway	−0.018	−0.050 ***	−0.075 ***	−0.002	−0.013	−0.034 **
	(0.024)	(0.017)	(0.025)	(0.019)	(0.054)	(0.014)
lnGDP	0.249 ***	0.119 ***	0.146 ***	−0.007	0.138 ***	0.082 ***
	(0.043)	(0.019)	(0.019)	(0.036)	(0.023)	(0.023)
industrial	−0.003 **	0.002 ***	0.000	0.006 ***	0.000	0.003 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
lnPD	−0.232 *	−0.017	−0.036	−0.037	−0.153 **	0.002
	(0.121)	(0.025)	(0.041)	(0.032)	(0.065)	(0.026)
lntem	0.569 ***	0.073 *	0.104 **	0.166	0.032	0.387 ***
	(0.143)	(0.044)	(0.046)	(0.150)	(0.051)	(0.086)
lnrain	−0.064 ***	−0.086 ***	−0.084 ***	−0.043 **	−0.087 ***	−0.079 ***
	(0.021)	(0.013)	(0.013)	(0.021)	(0.020)	(0.013)
lnwind	−0.103 **	−0.146 ***	−0.174 ***	−0.144 **	−0.203 ***	−0.112 ***
	(0.044)	(0.032)	(0.031)	(0.062)	(0.047)	(0.031)
Time FE	Yes	Yes	Yes	Yes	Yes	Yes
City FE	Yes	Yes	Yes	Yes	Yes	Yes
Observations	816	1409	1701	524	912	1313
R-squared	0.632	0.584	0.582	0.603	0.550	0.634

Note: *** p < 0.01, ** p < 0.05, * p < 0.1, standard errors in parentheses.

.

The coefficient of the explanatory variables for the YEB cities is negative but not significant, and the coefficient for the non-YEB cities is −0.05, which passes the significance test at the 1% level. This indicates that the improvement effect of subway opening on air pollution in non-YEB cities is significantly stronger than that in YEB cities. This may be because the cities in the Yangtze River Economic Belt, as the core region of China’s economic development, are generally characterised by high population density and dense industrial clusters. Although the opening of the subway can replace some private car trips, the economic activities and population movements along the route may increase the demand for commuting, partially offsetting the substitution effect. In addition, the share of industrial emissions in cities in the Yangtze River Economic Belt is generally higher than the national average, and the impact of industrial pollution on air quality may overshadow the marginal substitution effect of the subway. In contrast, non-YEB cities have relatively little economic and human activity. At the same time, cities have relatively lighter industrial pollution and a higher share of transport pollution in overall pollution, making the air pollution improvement effect of subways more readily apparent.

5.5.2. Economic Level Heterogeneity

A city normally builds a subway mainly taking into account the level of the economy. The varying economic levels of cities may result in a disparate impact of subway openings on air pollution levels. This paper measures the city’s economic status using its per capita GDP. This article categorises the sample cities into two groups based on their median per capita GDP: low economic level cities and high economic level cities. The regression outcomes are presented in columns (3) and (4) of Table 7.

The coefficient of the explanatory variable for cities with a low level of economy is −0.075, which is significant at the 1% level, while the coefficient for cities with a high level of economy is negative but not significant. It shows that the effect of subway opening on air pollution improvement in low-level cities is significantly stronger than that in high-level cities. This may be because the subway network in low economic level cities is mostly concentrated in the core urban areas, and the line planning matches the population distribution better, which makes it easier to achieve the private car substitution target. In contrast, the substitution effect of the subway is partly offset by the fact that high-level cities have better public transport networks and metro lines extend to the suburbs, which may stimulate the “job-household separation” phenomenon and lead to longer commuting distances, increasing the overall demand for transport.

5.5.3. Urban Heterogeneity

Resource-based cities are urban areas where the primary industry is the extraction and processing of natural resources, including minerals and forests, within the region. They have made great contributions to the establishment of China’s independent and complete industrial system and the development of the national economy. However, the prolonged and intense mining and excavation of resources have led to resource-based cities generally facing serious problems such as environmental pollution and ecological damage. This is often at variance with other cities in terms of development planning, etc., which in turn may make the opening of the subway have a differential impact on air pollution levels. Therefore, this paper categorizes cities into resource-based and non-resource-based for heterogeneity testing. The regression outcomes are presented in columns (5) and (6) of Table 7.

The coefficient of the explanatory variable for resource-based cities is negative but not significant, while the coefficient for non-resource-based cities is −0.034, which is significant at the 5% level. It shows that the opening of the subway has a significantly stronger effect on the improvement of air pollution in non-resource-based cities than in resource-based cities. This may be due to the fact that resource-based cities have long relied on resource development and heavy industry, and industrial pollution dominates in air pollution. Even though the opening of the subway has produced a transport substitution effect, the industrial emission pollution makes it difficult to show a significant improvement in the overall air pollution. In addition, resource-based cities generally face lagging industrial transformation and population outflow. The subway network is mostly laid out around traditional industrial areas, with a low degree of matching with new residential areas, failing to effectively guide the optimisation of the commuting structure and weakening the efficiency of the subway as a substitute for the private car. In contrast, the industrial structure of non-resource-based cities is relatively diversified, and the proportion of traffic pollution in the overall air pollution is relatively high. The opening of the subway can more directly reduce pollutant emissions from transport sources by replacing private car travel, thus showing a significant air pollution improvement effect.

5.6. Discussion and Limitations

In this paper, a DID model is constructed to assess the impact of subway opening on urban air pollution in China by integrating transport infrastructure and environmental quality into a unified framework. This study aims to fill the research gap on the long-term impact of subway opening on urban air pollution. The findings of the study not only provide new empirical evidence for the role of the subway in urban environmental governance, but also provide a reference for optimising subway layout policies and promoting sustainable urban development.

The findings of this paper are similar to those of existing studies [20,27,48], which concluded that the opening of the subway significantly reduces the concentration of air pollutants and improves urban air quality. This improvement effect is mainly achieved through the substitution of private cars. However, Xiao et al. [39] found that the subway favoured the substitution of taxis, thereby reducing taxi emissions. This may be due to the fact that the period of their study is 2000–2012, which does not overlap with that of this paper. In the discussion of regional heterogeneity, existing studies generally use east, central and west to classify cities [14,39], which may hide the differences in the development of different cities within the same region. In contrast, this paper adopts the Yangtze River Economic Belt for regional division, which can capture the synergistic effects of environmental policies and transport networks more precisely. In addition, this paper discusses the impact of different urban development patterns, i.e., whether it is a resource-based city or not, on the improvement effect of air pollution, which provides a new perspective to study the heterogeneous impact of subway opening.

The study also has some limitations. First, this study uses sample data at the prefecture-level city level in China for the empirical study, but due to the unavailability of national raster data samples, it is not possible to examine the improvement effect of the subway openings on air pollution within cities. In further studies, we will endeavour to collect raster data to provide more micro empirical evidence. Second, we discussed the effects of regional, economic level, and urban heterogeneity in the heterogeneity analysis section, but the spatial insights were limited to broad qualitative statements, and the spatial analyses were not well developed. In future research, we will endeavour to improve our spatial analysis, enhance the visualisation of our study, and collect more relevant data to provide a more complete spatial analysis. Finally, our study lacks a substantial contribution to urban environmental research. Our existing analyses are mainly based on air quality index and pollutant concentration data, and we have not yet directly examined the impact of the opening of the subway on the health of residents. Although there is a strong correlation between air pollution and health problems such as respiratory diseases, whether the reduction in pollutant concentrations can be translated into health benefits needs to be further verified. We will further collect health indicators at the individual level, such as the incidence of diseases and the frequency of medical consultations, in order to assess the effect of the subway openings on the improvement of residents’ health, and thus provide substantial contributions to the study of the urban environment.

6. Conclusions and Policy Recommendations

Identifying the improvement effect of subway opening on urban air pollution provides an important basis for coordinating transport infrastructure development with ecological sustainability. This study utilises panel data from 282 Chinese cities spanning 2014 to 2021 and employs the DID model to assess the effect of subway openings on urban air pollution, leading to the following results. (1) The opening of the subway significantly decreases air pollution concentrations and enhances urban air quality, and this conclusion remains valid after a series of robustness tests. (2) The mechanism analysis indicates that the opening of the subway reduces urban air pollution mostly through the traffic substitution effect, resulting from the replacement of private car traffic. (3) Heterogeneity research indicates that the abatement effect of subway openings on air pollution is more significant when the city is a non-Yangtze River Economic Belt city, or when the city’s economic level is lower or when the city is non-resource-based.

Based on the above findings, this paper puts forward the following policy recommendations: first, the government should increase financial support to expand the scale of subway construction. It is recommended that the central and local governments set up a special financial fund to give priority to supporting cities that meet the construction conditions to speed up the process of metro planning and construction. This will not only help to alleviate regional air pollution problems, but will also lead to balanced local economic development through infrastructure investment.

Second, the government should optimise the layout of the subway. It is recommended that subway lines connecting high-density residential areas, business districts and transport hubs be prioritised in subway construction planning to increase station density and accessibility. Particular attention needs to be paid to line coverage in peripheral urban areas to reduce the need for private car commuting. It is also recommended that the government implement a transit-oriented development (TOD) model based on the subway system. High-density mixed-use areas should be built around subway stations, integrating commercial, residential and public service facilities to shorten the travelling distance of residents and further enhance the substitution effect of the subway on private cars.

Third, the government should implement differentiated development strategies for different types of cities. For example, for densely populated cities with complex transport needs, such as the Yangtze River Economic Belt cities, subway construction needs to be linked to regional transport networks, such as water transport and intercity railways, to avoid the traffic creation effect offsetting the emission reduction effect. For cities with a low level of economic performance, it is recommended to focus on promoting the construction of low-cost and high-efficiency subway lines, and to simultaneously implement supporting policies such as bus fare subsidies. For resource-based cities, it is necessary to promote the construction of subway lines in synergy with industrial transformation programmes, attracting green industrial agglomeration by improving transport accessibility and reducing path dependence on traditional high-polluting industries.