Do rail transits improve local air quality? Take Chengdu-Nanchang for example

Many cities in China have invested in the city’s rail transit system to reduce urban air pollution and traffic congestion. Earlier studies rarely compare the effects of rail transit on urban air quality in different cities, providing little guidance to urban planners in solving traffic congestion and air quality. Due to the same opening date, we regard Chengdu Metro Line 4 and Nanchang Metro Line 1 as case studies. This paper attempts to examine the effects of the opening of rail transit on local air quality on the same opening date. Data were collected from 17 monitoring stations distributed along the chosen rail transit lines in both cities from 2015 to 2016 and analyzed using the regression discontinuity design to address the potential endogenous location of subway stations. The results show that subway opening in Nanchang has a better reduction from automobile exhaust than that in Chengdu. Specifically, carbon monoxide pollution, one key tailpipe pollutant, experienced a 10.23% greater reduction after Nanchang Metro Line 1 opened. However, the point estimate for carbon monoxide in Chengdu is 22.42% and statistically significant at the 1% level. Nanchang Metro Line 1 does play an important role in road traffic externalities, but the benefit is not huge enough to change the overall air quality. On the contrary, the opening of the Chengdu Metro Line 4 is unlikely to yield improvements in air quality.


Introduction
China's remarkable economic growth during the past four decades has been accompanied by environmental challenges, air pollution in particular (Zhang et al. 2020;Li et al. 2019;Wu et al. 2019). Traffic has been considered a significant factor for ambient air pollution (Zhang et al. 2021). The evidence shows that vehicular emissions contribute to around 33% to 57% of ambient air pollution (Chen et al. 2020). The environmental effect of traffic may be hardly negligible; air pollution has posed a threat to sustain high-quality economic development in China. To reverse the trend of nationwide environmental deterioration, the Chinese government decided to transform its economic pattern to a new one featured with technology innovation since the Ninth Five-Year Plan by supporting the implementation of the Air Pollution Control Action Plan and setting detailed emission reduction target for each Five-Year Plan (Wang et al. 2019).
The most commonly adopted policy measures for improvements in terms of traffic pollution are routinely focusing on limiting transportation demand, such as imposing road pricing (Gibson and Carnovale 2015), parking controls (Melia and Clark 2018), driving restrictions (Zhang et al. 2017(Zhang et al. , 2020, and traffic calming (Akbari and Haghighi 2020) or increasing transportation supply, including increasing capacities of transportation facilities and investments. Compared with other external transportation infrastructures such as river transport, road traffic, or air transport, urban rail transits featured with safety and comfort improve the accessibility within cities, which effectively ease urban traffic congestion and air pollution Zhang et al. 2021). The Chinese government has been investing heavily in transportation infrastructure. As of 2019, 46 cities in mainland China had opened urban rail lines, with 5680.84 km in total route length. The rapid subway expansion is still ongoing at the national level: the network consists of another 266 subway lines, serving 3,872 subway stations with a total length of 6282.6 km. Recently major cities in China have also been investing heavily in green infrastructure to mitigate air pollution. However, existing studies rarely compare the effects of rail transit on air quality in different cities. Nanchang Metro Line 1, with the opening date of December 26, 2015, runs from Shuanggang in the north to Yaohu West in the east, serving a total of 24 stations with a total length of 28.84 km. On the same date, the first phase of Chengdu Metro Line 4 opened, 1 running from intangible cultural heritage park to Wannianchang Station. Owing to the same opening date, this paper focuses on the variations in concentrations of various air pollutants by using urban air pollutant monitoring stations data from Nanchang and Chengdu during the period 2015 to 2016. The literatures in subway infrastructure incorporate two countervailing forces; one possible channel that may affect traffic pollution is the possibility that households near the subway stations may substitute for their driving after the rail transit opened; this traffic diversion effect may relieve traffic congestion and thus alleviate air pollution (Mohring 1972). The other possible channel is that improved traffic conditions may provide some people near the subway stations with more potential choices and induce additional travel demand using private cars (Vickrey 1969). The traffic creation effect would result in the increase of the traffic volume in the streets and the prolongation of peak-hours.
This analysis is relevant to currently high-profile discussions about whether the opening of rail transit has positive effects on air quality in different cities. Our empirical analysis leverages discontinuity in the concentration of air pollutant to control for possible confounding factors. The data on air quality come from hourly air quality readings from 17 monitoring stations. Figure 1 and Fig. 2, respectively, show the spatial distribution of air pollutant monitoring sites and rail transit stations; eight of these are located in Chengdu, and The central empirical challenge in scrutinizing the effects of rail transit on air quality is an identification concern. First, it is hard to give sound results when confounding variations in rail transit with other factors affecting air pollution (Chen and Whalley 2012;Li et al. 2019). For example, subway stations are more likely to be located in areas with high population growth, which induces travel demand and increases air pollution. Second, it is a mistake to confound the variation in the demand for public transit with other covariates that affect air quality (Rivers et al. 2017). The demand for public transport is associated with the overall price, income, speed, and frequency elasticities (Toro-González et al. 2020). Public transportation is not a sufficiently good alternative for those who travel a lot (Lalive et al. 2018), and the demand for public transport decreases given an increase in per capita income (Toro-González et al. 2020).
The paper is structured as follows: Sect. 2 provides the research hypothesis. Section 3 presents the material and methods. Section 4 describes the empirical analysis. Section 5 presents the discussion. Section 6 concludes.

Research hypothesis
The subway is a good choice as a travel mode in the modern urban green transportation system. The leading environmental effects of subway infrastructure have been practiced in many cities (e.g., Taipei, Changsha, Mexico City). However, empirical evidence on its effect on air quality is unclear. Whether properly invested in subway infrastructure can lead some commuters to prefer to travel using subways that to a certain extent may improve air quality (the so-called traffic diversion effect or "Mohring effect") or otherwise induce residents to switch to cheaper communities far away from the railway station (the so-called traffic creation effect) has triggered traffic congestion and deteriorated air pollution (Vickrey 1969;Mohring 1972). According to the earlier literature, the subway network could create two countervailing forces that could affect air quality, i.e., the traffic diversion effect and the traffic creation effect. We present these effects in the form of mathematical models through an analysis of the associated parameters. The traffic diversion effect means rapid subway expansion could reduce the trip frequency of travelers using private cars and the daily flow of motor vehicles; then we believe the impact of new subway opening on air quality seems to be positive, and it may relieve traffic congestion and then improve air quality.
While the traffic modes of resident turn to diversity, such as bicycle, taxi, bus, private vehicle, train, subway, and so on, travel mode selected depends much on the relative prices charged on them. We assume the total demand of traffic remains virtually unchanged, the price of green transportation isP G , and the price of non-green transportation is P NG (see Fig. 3). The increase in the supply of green traffic infrastructures in China has led to a decline in the relative prices of green transportation and non-green transportation. Hence, we have . The improved traffic coverage could lead some commuters to trade down to green traffic infrastructures which are cheaper. Thus, a falling-off in relative price shifts the budget line outward in a parallel fashion. The traveler's purchasing power increases because the set of baskets from which he can choose is increased.
In Fig. 3, the commuters initially face budget line KK 1 . With the indifference curve U 1 pictured, the optimal basket will be at point P, where the commuters could purchase B units of off-green transportation and E units of green transportation. When the price of green transportation falls short of off-green transportation's price, the budget line rotates outward to JJ 1 . That new budget line JJ 1 is tangent to indifference line U 2 at point M, where the commuters could purchase C units of off-green transportation and G units of green transportation. Thus, the overall effect of the price change on the quantity of green transportation chose is E-G. The traveler realizes a higher level of utility as the falling-off in relative price, as shown by the fact that the initial basket P lies inside the new budget line JJ 1 . The dashed line AA 1 has the same slope as the budget constraint KH but is drawn to be tangent to the initial indifference curve U 1 . The substitution effect leads to an increase in the amount of green transportation consumed from E to F (so the substitution effect is E-F). The income effect also leads to an increase in green transportation, from F to G (so the income effect is F-G). Notice in Fig. 3 that the income and substitution effects work in the same direction and cut down the demand for green transportation in response to a decrease in its relative price.
On the other hand, investment in highways could in some cases worsen traffic congestion (Ding and Song 2012). For cities with large populations, demand for subway infrastructure is usually greater than supply (Zhang et al. 2021), which leads commuters to shift from public transport to private cars or shift to the upgraded road from other congested roads. Meanwhile, faster increases in housing price near the subway station foster the fact that traffic costs of off-green transportation seem to be cheaper, which induces additional travel demand using private cars and causes the traffic creation effect.
In the following figure, we perform this analysis in relation to the price of green transportation increases. When the price of green transportation is higher than that of off-green transportation, hence, we have . As shown in Fig. 4, when the initial budget line is KK 1 , the commuters realize maximizes utility by choosing basket P on indifference curve U 1 , where the commuters could purchase D units of off-green transportation and F units of Fig. 3 The traffic diversion effect. The vertical axis is the non-green traffic demand. The horizontal axis is the green traffic demand, the price of green transportation is P G , and the price of non-green transportation is P NG substituation effect green transportation. As the price of off-green transportation decreases, the income effect leads to an increase in the amount of off-green transportation from D to C (so the substitution effect is D-C). The substitution effect leads to an increase in the amount of off-green transportation consumed from B to C (so the substitution effect is B-C). Thus, the income and substitution effects work in the same direction and increase the demand for non-green transportation in response to a reduction in its relative price.

Site description
Chengdu, the capital of southwestern Sichuan Province, is one of the fastest-growing regions in China. The number of passenger vehicles in Chengdu increased dramatically in the past decade; its stock of passenger vehicles increased from about 416,900 units in 2004 to 4.2 million units in 2018. With the rapid development of city economy and citifying course, the tendency of the increasing quantity of private vehicles seems to lead to traffic flow uprising and congested road. Chengdu has made many efforts in the public transportation system, such as building intelligent transportation systems, traffic restrictions based on the license plate, and heavily investing in rail transit. As of December 2018, urban rail lines have been in operation, with 329.8 km in total route length and 190 subway stations. The network consists of six lines (numbered 1, 2, 3, 4, 7, 10), serving 156 stations with a total length of 226.017 km. Transportation planners also consider bus routes and their layout based on passenger flow and integrate them with the Chengdu Metro system. Chengdu Metro Line 4 runs from Wansheng in the west to Xihe in the east, which consists of 5 major transfer points with the other public transport modes: Chengdu Tram Line 2, Chengdu Metro Line 7, Chengdu Metro Line 3, the city expressway, and local buses.
Nanchang is one of the most important regional economic centers in Jiangxi Province in eastern China. It ranked the 42nd in one hundred better cities in China, and per capita GDP was about 9,825 RMB in 2018. By the end of 2018, Nanchang had a total population of 5.545 million, with a density of approximately 771 people per square kilometer, roughly half of that found in Chengdu. The top 5 areas with the highest population density include Nanchang County, Jinxian County, Xinjian District, Donghu District, and Qingshanhu District (Statistics Bureau of Nanchang, 2018). Nanchang Metro Line 1, with its opening on December 26, 2015, runs from Shuanggang in the north to Yaohu West in the east, serving a total of 24 stations with a total length of 28.84 km. It passes through four administrative regions, including Qingshanhu District (west), Donghu District, Qingshanhu District (east), and Nanchang County. Every day, 0.38 million people travel on urban rail transit systems in Nanchang. The Nanchang Metro network operates daily from 6:00 am to 11:20 pm, with an interval of 2-4 min.

Data
Our research relies on three major datasets, as presented in Table 1. We obtained these air pollution data from the Ministry of Ecology and Environment of the People's Republic of China, 2 which records the daily and hourly air quality index (AQI) as well as the concentration of six related pollutants of 1497 monitoring stations widely distributed in more than 300 cities. The number of pollutant monitoring stations included in this paper is 8 and 9, for Chengdu and Nanchang, respectively. We take the average at the monitoring station level on an hourly basis.
The second dataset contains the independent variable. The variable subway it is the independent variable to measure the effects of the opening of rail transit on air quality. It is represented by a dummy variable showing whether city i opens a new rail transit line at time t. It takes a value of one for all hours after the city's Metro is operational and a value of zero before the city's Metro is operational. Our data source for the construction of rail transit is the China Association of Metros. 3 As our empirical analysis is based on regression discontinuity design, observations local to the opening date may be not affected by macro factors. Weather controls are often considered in the analysis of the effects of air quality (Chen and Whalley 2012;Li et al. 2019;Zhang et al. 2020). Thus, we include control variables: temperature, air pressure, wind speed, and precipitation. Meanwhile, in the central analysis, we also consider station fix effects and time fix effects to  Fig. 4 The traffic creation effect. The vertical axis is the non-green traffic demand. The horizontal axis is the green traffic demand, the price of green transportation is P G , and the price of non-green transportation is P NG control for unobserved factors that may vary across locations and time, respectively. The National Meteorological Science Data Center 4 provides daily data for weather monitoring stations. Due to the lack of the original data of climate variables, this chapter is based on the principle of proximity in the data processing. Specifically, we use the air quality index at time 0-2 to estimate climate variable at time 0 and the air quality index at time 3-6 to assess climate variable at time 3 and so on.

Methodology
Many documents and papers nowadays have used three approaches to incorporate a sequence of effects of rail transit on air quality. The first method they adopted is ordinary least squares based on the growth of public transportation subsidy to value rail transit infrastructure. This method is used to establish the regression model of rail transit, and its results usually show that the endogenous deviation occurred. A second approach uses the difference-indifference (DID) to analyze cost-benefit estimates. A classic example of this kind of approach is Li et al. (2019), who studied the effects of the expansion of the Beijing subway system and found a 7.7 percent reduction in air quality within 2 km of the subway station than the monitor within a radius of 20 km 60 days after the opening date of rail transit. A third approach uses regression discontinuity design to estimate the potential effects of rail transit. For example, Rivers et al. (2017) provided a series of studies of 39 cities worldwide that opened subways from August 2001 to July 2013, concluding that the opening of a new rail transit line was conducive to the improvement of overall social welfare.
We measure the effect of rail transit on air quality during the period 2015-2016 by using the sharp regression discontinuity design to control for possible confounding factors. To eliminate endogenous problems, we leverage three datasets to study the causal relationship between rail The unit of observations summarizes the daily air quality and hourly air pollutant concentration from 2015 to 2016. CO is carbon monoxide, NO 2 is nitrogen dioxide, PM 2.5 is particulates smaller than or equal to 2.5 µm in diameter, O 3 is ozone, PM 10 is particulates smaller than or equal to 10 µm in diameter, SO 2 is sulfur dioxide. All pollutants are expressed in parts per million, the temperature is expressed in degrees Celsius, the air pressure is expressed in hPa's, wind speed is expressed in meters per second, precipitation is expressed in millimeter. The entries in brackets in columns (5), (10) report the standard deviation of the variable *** Significant at the 1 percent level * Significant at the 10 percent level transit and air quality in Chengdu and Nanchang. The key assumption is that the only reason for the discontinuous change of air quality on the opening date is the opening of the Metro itself. After the introduction of the dummy variable of subway it , if we can observe that air quality changes suddenly around the cutoff, and other covariates affecting air quality change smoothly in the neighborhood of the opening date, it is reasonable that the implementation of the dummy variable results in the discontinuity in the concentration of air pollutant. Specifically, we use the regression discontinuity (RD) design to estimate the following model: where Air it stands for air quality in the city i at time t, including air quality index (AQI), CO, NO 2 , PM 2.5 , O 3 , PM 10 , and SO 2 . The variable subway it is a dummy variable that takes a value of one for all hours after the city's Metro is operational and a value of zero before the city's Metro is operational. T x is forcing variable that indicates the number of days since the opening of the rail transit. It takes a value of zero on the opening day, negative before it and positive after it. The vector f(T x ) is a polynomial function on T x . We also include interactions between the forcing variable and the polynomial time trend to allow the time trend in air pollution to differ on either side of the opening date. The control variables X t include temperature, air pressure, wind speed, and precipitation. The vector t indicates year, month, week, day of the week, hour, and holiday fixed effects. t is the error term. The coefficient of interest is 1 which can directly reflect the changes of pollutant concentration and air quality index (AQI) after the opening of rail transit in each city. (1)

Descriptive statistics analysis
Descriptive statistics of air quality and weather variables are summarized in Table 1. The concentrations of most air pollutants are noticeably higher in the post-Chengdu Metro period than those in the pre-Chengdu Metro period. Furthermore, the concentrations of PM 10 and PM 2.5 are noticeably higher than those of other pollutants, suggesting that fine-particle pollutant is in the prominent place among the various pollutants. Then, the concentrations of various air pollutants in Chengdu are higher than those in Nanchang, except for ground-level ozone (O 3 ) and sulfur dioxide (SO 2 ).

Main results: RD estimation
Chengdu Metro Line 4 opened on December 26, 2015. Taking the opening date of rail transit as the cutoff, we analyzed the changing trend of air pollutant within the observation period of 40 days. We first provide a graphical presentation of the effect of Chengdu Metro opening on air quality (see Fig. 5).
We can see a sharp jump discontinuity in the concentration of air pollutant within 40 days before and after the opening date. Specifically, after the cutoff, AQI leaps upward and increases sharply and then it decreases slightly; finally, it presents the trend of increasing. So the visual analysis method of the data in the narrow window of the Chengdu Metro opening date allows for further statistical analysis. Next, we present the results of models that include controls for weather conditions and time fix effect and station fixed effects in Table 2 to account for variations in air pollution in both cities. The results in Table 2 show a model that controls the relevant variables, indicating that most ambient air pollutants, except for O 3 , have the positive relationship with the opening of the Chengdu Metro Line 4, which is significant at the 1% significance level. The estimates of Chengdu Metro opening reported in Table 2 are consistent with the discontinuities indicated in Fig. 5. In this specification, there is a 22.42 percent increase in CO, which crudely reflects the evidence of pollution. In the scenario, variation in the concentrations of air pollutant increases is more likely to be affected by additional travel demand after the opening of the new transit line, resulting in the traffic creation effect.
Then, taking the time when Nanchang Metro Line 1 opened as the cutoff, we consider the graphical presentations of the concentrations of air pollutant near the cutoff in Nanchang using 40 days window of data (see Fig. 6).
As shown in Fig. 6, we found that sharp discontinuity in the concentration of air pollutant around the opening date of the Nanchang Metro Line 1 is secondary polynomial fitting. The changing trend of AQI appears a clear rising trend before the cutoff, and there is a downward trend for AQI after the cutoff. We estimate the results of Nanchang based on control variables when the second-order polynomial was used to fit the time trend in a shorter 40-day window around the Nanchang Metro opening date in Table 2. There is a negative relationship between subway opening and vehicle emission pollutants across column (2) to (3), while a positive correlation between subway opening and some air pollutants  Each column reports the results from one regression with a second-order polynomial time trend. The sample for all regression is 40 days before and 40 days after the Metro opening date. The main entries in columns (1), (2), (3), (4),(5), (6),(7) report the coefficient estimate from the model (1). CO is carbon monoxide, NO 2 is nitrogen dioxide, PM 2.5 is particulates smaller than or equal to 2.5 µm in diameter, O 3 is ozone, PM 10 is particulates smaller than or equal to 10 µm in diameter, SO 2 is sulfur dioxide. Specifications include weather covariables and indicator variables for month of the year, day of the week, and hour of the day *** Significant at the 1 percent level  (5) to (7). The contradiction of environmental effects of Nanchang subway opening may boil down to dual influences of the traffic creation effect and the traffic diversion effect. Overall, our results have confirmed the conjecture of differential effects of the Metro's opening on the average level of air pollution for both cities.

Robustness test
In this result, we use a 40-day window to examine the continuity of the weather variables. There is no jump of the control variables within narrow window around the opening date, which could verify the effectiveness of the results.
The meteorological conditions, such as precipitation, wind speed, and temperature, have been main factors that affect the levels of ambient air pollutant (Chen and Whalley 2012). We first show the plots for major meteorological parameters before and after the opening of metro line to make sure the difference is not driven by meteorological differences. In Fig. 7 and Fig. 8, we find little evidence for a large discontinuity in control variables on the opening date in Chengdu and Nanchang, except for precipitation. The graphical representations of weather controls provide intuitive impression, but there are still some limitations. We then report estimates of the Eq. (1) with control variables as the outcome. The results in Table 3 show that the coefficients of air pressure and precipitation are insignificant on the opening date in  Nanchang and Chengdu. However, temperature does not change smoothly on Nanchang Metro opening date; similarly wind speed on Chengdu Metro opening date does not change smoothly. As the results in this context would have less explanatory power for whole robustness test, we present further estimates to address the potential concern. Second, we examine RD estimates with varying window widths. The bandwidth is adjusted to 15 days, 20 days, 25 days, 30 days, and 35 days, respectively; these samples are selected as the bandwidth for regression to be carried out for robustness test. Limiting the sample period to a narrow window could to some extent help disentangle the effect of rail transit from the effect of other time-varying factors that affect air quality (Davis 2008). The results in Table 4 show the main results of five different bandwidths in Chengdu. It can be seen that when the bandwidth is 15, the coefficient of AQI is statistically positive at the 10% level. The opening of rail transit has positive effect on NO 2 , O 3 , PM 2.5 , and SO 2 , and the results are statistically significant. It can be seen that the regression results in different bandwidths are consistent with the above discussion.
Similarly, Table 5 presents the estimated results of the narrower window before and after the opening date of Nanchang Metro in different bandwidth. We can see that the opening of rail transit has negative effect on CO, NO 2 , and PM 2.5 at the 1 percent significance level. However, when the bandwidth is 15, the coefficients of PM 10 and O 3 are statistically positive. It can be seen that the results in five different bandwidths are basically consistent with the above analysis.
We may get a significant effect on the estimations if we choose the treatment effect at the real cutoff, which could verify that it is the real cutoff, not other cutoffs, at work. A common approach to probe it is to estimate treatment effects at other cutoffs and compare them with the results at the real cutoff. In this subsection, we employ placebo checks for the assumed data in 2014 as the treatment groups to test the treatment effects at other cutoffs. Placebo checks, reported in Table 6 for the main outcomes, document that the Metro opening does not impact the air pollution statistically on December 26, 2014; that is, the treatment effect at the assumptive cutoff is ineffective in improving air quality.

Heterogeneity test
The results thus far have demonstrated the differential effect of the Metro's opening on local air pollution in Chengdu and Nanchang. In this subsection, we examine the concentrations Each column reports the results from one regression with controls for a second-order polynomial time trend, year, month of year, week, hour of day, holiday and stations based on the Regression discontinuity designs (RD). The sample for all regressions is 40 days before and 40 after the opening date *** Significant at the 1 percent level ** Significant at the 5 percent level * Significant at the 10 percent level [-15,15] [ -20,20]   of air quality responds to the Metro opening differently depending on whether the monitor time is peak hours. Peak hours involve morning peak (6:00-9:00 am) and evening peak (5:00-8:00 pm). We present evidence for heterogeneous results of ambient air pollutant in Tables 7 and 8, respectively. We see very large differences in the concentrations of ambient air pollutant between peak and off-peak hours in Chengdu and Nanchang. The results in Table 7 indicate that Chengdu Metro Line 4 opening is positive and statistically significant at the 1% level for the six pollutants (CO, NO 2 , PM 2.5 , O 3 , PM 10 , SO 2 ). Our analysis of the effects of the opening of the Nanchang Metro Line 1 reveals two findings. First, compared with the regression results in the off-peak hour in Nanchang, the coefficients of air quality index AQI and some air pollutants, such as CO, NO 2 , PM 2.5 , and PM 10 , are lower in peak hour. Second, results indicate significantly negative effects on air quality index (AQI) and three for six pollutants (CO, NO 2 , PM 2.5 ). We find that the opening of the Nanchang Metro significantly reduced nitrogen dioxide from − 27.489 (off-peak) to − 29.137 (peak). Overall, the traffic creation effect coexists with traffic diversion effect in Nanchang Metro Line 1, which is consistent with the evidence that subway opening encourages some people to take the subway and thus reduce vehicle exhaust emissions, and at the same time, the improved traffic conditions induce some people to travel more frequently and increase air pollution to some extent.

Discussion
This paper focuses on the variations in concentrations of various air pollutants of Chengdu and Nanchang by using the regression discontinuity design. For an easily understandable discussion, the changes in air quality of the Chengdu and Nanchang from 2015 to 2016 are depicted in Fig. 5 and Fig. 6, respectively. Consistent with statistical results (shown in Table 2), very different results can be observed in both cities. Furthermore, as presented in Fig. 9, mountain terrain surrounds Chengdu city, and many counties and towns lie in the Sichuan Basin, which is unfavorable for pollutant dispersion while facilitating pollution accumulation under stagnant meteorological conditions. As presented in Fig. 10, Nanchang city is mainly to the plains, susceptible to monsoon climate, which are conducive to the diffusion of pollutants.
In this study, we empirically analyzed the influence of the opening of rail transit on the levels of AQI as well as the concentrations of six related pollutants from Chengdu and Nanchang over the period 2015 to 2016.We use the estimates of the effect of Metro opening on air quality in Table 3. By dividing these estimates by the corresponding averages 1 year before Metro opening date from Table 1, we can obtain estimates of the effect of rail transit lines opening on air quality. There are quite a few differences in statistical results between Chengdu and Nanchang city. Firstly, the total population in Chengdu has surpassed that of in Nanchang; particularly, the population aged from 15 to 64 makes Table 7 Effect of Chengdu Metro opening on pollution levels during peak and non-peak hour Each column reports the quadratic term results from Regression discontinuity designs (RD). The sample for all regression is 40 days before and 40 days after the Chengdu Metro opening date. Specifications include weather covariables, time fixed effects and monitoring station effects *** Significant at the 1 percent level ** Significant at the 5 percent level Compared with Chengdu, both the population density and pollution levels display discernible differences in Nanchang. By the end of 2018, Nanchang had a total population of 5.545 million, with a density of approximately 771 people per square kilometer, roughly half of that found in Chengdu. After the opening of Nanchang Metro Line 1, the average concentrations of CO, NO 2 , and PM 2.5 , reduced by 10.23%, 0.48%, and 22.24%, respectively, but O 3 , PM 10 , and SO 2 increased by 14.79%, 1.56%, and 33.70%, respectively. It is worth mentioning that compared with Chengdu Metro Line 4, the opening of Nanchang Metro Line 1 decreased the concentrations of pollutants associated with vehicle exhaust. During the study, the automobile industry in Nanchang rose much more slowly than that in Chengdu. Nanchang's number of civil vehicles increased from 618,086 units in 2014 to 861,045 units in 2016. Meanwhile, Nanchang Metro Line 1 covers the geographical areas with the highest population density. On the one hand, the expansion of subway network in Nanchang could lead some commuters living near the subway stations to prefer to subway. This traffic diversion effect may be conducive to alleviate people's travel pressure and thus reduce automobile exhausted gas pollution. On the other hand, a completely new subway line's opening could offer more convenient mobility options and induce additional travel demand, resulting in the traffic creation effect. In addition, the evidence of the robustness test cannot be ignored, which is discussed in detail below.  Table 3 reveal that not all of the control variables are smooth on the opening date, we present further estimates to address the potential concern.
The study however has a couple of avail aments. First, we cannot provide enough evidence for the efficiency of detection based on the limited number of rail transit lines. Thus, future work is needed to explore if a transport plan at the city scale is possible for the impacts of rail transit on urban air pollution. Furthermore, air pollution is known to be sensitive to many factors such as the delay time of congestion, space range, amount of gasoline pumped, and so on. We cannot confirm the precision of the results of statistical analysis since we find it hard to obtain the above information by official statistics. Lastly, although identifying one specific line could provide useful information to guide policymakers to the local traffic problems, put concrete environment concrete analysis, a single line could not map to the city's transport network and present temporal and spatial variation in traffic patterns.

Conclusions
Air pollution and traffic congestion are serious threats to health worldwide. The existing literature upon the precise relationship between the opening of rail transit and air pollution is highly debated with some argue traffic creation effect while others emphasize traffic substitution effect. When examining air quality effects of rail transit, existing studies rarely consider traffic creation effect and traffic substitution effect simultaneously. However, this major question has been shown to be increasingly important for fast-growing developing countries, especially for countries like China. As the rapid economic growth, many cities in China have been investing heavily in transportation infrastructures to meet the different travel demand of various kinds of people commute and to combat traffic congestion and air pollution, such as rail transit construction. This paper aims to fill the gap by using hourly pollutant concentration data combined with weather controls in Chengdu and Nanchang for the period of 2015-2016.
Using the regression discontinuity design based on the combinations of weather controls (i.e., temperature, wind speed, volume, precipitation), air pollution (CO, NO 2 , PM 2.5 , SO 2 , O 3 , PM 10 ), time fixed effect, and station fixed effect, this study seeks to empirically examine the impact of the opening of rail transit on local air quality. We find that the opening of rail transit in different cities may have different effects on local air pollutants. The findings of this paper are contrary to much of the existing literature on the relationship between rail transit and air quality. Firstly, the concentrations of the main pollutant in the automobile exhaust reduced significantly after Nanchang Metro Line 1 opened. More specifically, carbon monoxide experienced a 10.23% greater reduction, but other atmospheric pollutants such as O 3 , PM 10 , and SO 2 may produce adverse environmental effects. The opening of Nanchang Metro Line 1 leads to two contradictory results. On the one hand, the expansion of Nanchang's subway network could lead some commuters living near the subway stations to prefer to subways. This traffic diversion effect may be conducive to alleviate people's travel pressure and thus reduce automobile exhausted gas pollution. On the other hand, a completely new subway line's opening could offer more convenient mobility options and induce additional travel demand, resulting in the traffic creation effect. Secondly, carbon monoxide pollution led up to 22.42% after Chengdu Metro Line 4 opened, and other forms of air pollution had also risen in varying degree. The opening of the Chengdu Metro Line 4 can provide no benefits in terms of air quality and even deteriorate air pollution. To understand how the concentrations of air pollutant change over time, we divided the sample into two sub-periods based on peak and off-peak hours. The results show that air quality turns out to be better because of the traffic substitution effect during the rush hour in Nanchang. Nevertheless, our empirical findings suggest that the influence of rail transit relies on population density and climatic and topographical conditions.
Understanding the traffic diversion effect and the traffic creation effect could shed further light on policy implications for rapid expansion of transport networks in China as well as other developing countries. The opening of rail transit could indeed affect local air quality, apart from population density as well as climatic and topographical conditions which have a different influence on air quality. Furthermore, even though the Chengdu Metro Line 4 and the Nanchang Metro Line 1 opened on the same date, their air quality effects of rail transit may vary because of different domination effect. To achieve the national targets for air quality improvements and smooth traffic, central and local governments should consider the traffic creation effect and the traffic substitution effect at the same time. Future research could examine the impact of subway opening on urban spatial structure, location value, and urban planning. Data availability All data generated or analyzed during this study are included in this published article.

Declarations
Ethics approval and consent to participate Not applicable.

Consent for publication
The authors of this manuscript grant the publisher of this manuscript the sole exclusive license of the full copyright, which the publisher hereby accepts.

Conflict of interest
The authors declare no competing interests.