Causal association between metro transits and air quality: China’s evidence

Abstract

In policy remedies, transportation infrastructure such as metro transit is widely considered to be an important and effective means to reduce air pollution. However, the policy prediction that metro transits reduce air pollution depends on driver responses. China provides an appropriate context to explore the answer since its major cities have expanded their metro transit systems in recent years, which enables us to exploit a natural experiment. Accordingly, a sharp regression discontinuity is employed to evaluate the impact of 112 metro lines (with an accumulation of 3286 km) on air quality. Evidence shows that the opening of metro transits has a negative causal effect on air pollution. The results are robust to several alternative specifications. Furthermore, a heterogeneity analysis demonstrates that metro openings have a time-varying impact on air pollution, which is larger during rush hours. We anticipate that the air quality improvement in China caused by metro transits can generate large welfare gains.

Appendices

Appendix 1. OLS

Although OLS cannot address endogeneity, it can provide an intuitive sense of the relationship between the opening of a metro and air pollution. Second, we are curious about the information that can be obtained if only cumulative mileage and air pollutants are considered in the regression.

The ordinary least squares (OLS) method is used to estimate the time series model:

$${\mathrm{y}}_{\mathrm{it}}={\upbeta }_{0}+{\upbeta }_{1}{\mathrm{length}\_\mathrm{acum}\_\mathrm{meter}}_{\mathrm{it}}+{\upbeta }^{\mathrm{^{\prime}}{\mathrm{X}}_{\mathrm{it}}}+{\mathrm{\alpha }}_{\mathrm{i}}{+{\uplambda }_{\mathrm{t}}+\upvarepsilon }_{\mathrm{it}},$$

(A.1)

where ${\mathrm{y}}_{\mathrm{it}}$ is the log of air quality in city $i$ at time $t$. ${\mathrm{length}\_\mathrm{acum}\_\mathrm{meter}}_{\mathrm{it}}$ is the accumulated kilometers (km) of the mileage of new metro lines opened in China from 5/15/2014 to 4/30/2018. ${X}_{it}$ comprises the control variables that include the holiday and weekend dummies, city and date dummies, city and date dummies, temperature, precipitation, and ventilation and these variables’ quartic. ${\alpha }_{i}$ is the city fixed effect, ${\lambda }_{t}$ is the time fixed effect, and ${\upvarepsilon }_{\mathrm{it}}$ is the error term. The coefficient of interest is ${\beta }_{1}$, which reflects the relationship between the metro’s accumulated length and air quality. As traffic jams become more frequent, people are becoming more inclined to use the metro to save travel time. We expect that ${\beta }_{1}$ will reflect the negative correlation between the metro accumulated length and air pollution.

The estimation results of Equation (A.1) are provided in Table 7. As shown in Table 2, the cumulative mileage was negatively correlated with NO₂, PM_2.5, and the AQI. NO₂ is significant at the 5% level. There is a positive but not statistically significant correlation between cumulative mileage and PM₁₀.

Table 7 The effect of cumulative mileage of metros on pollutants: basic OLS estimates

Appendix 2. Smoothness test

The weather, weekends, and holidays are the observable covariates that may affect the estimation results. Therefore, we further fit model (2) with OLS. As shown in Table 8, the wind speed, weekends, holidays, and temperature change smoothly on the opening date, but ventilation does not change smoothly. Note that the holiday dummy variable performs well in Table 8, with no discontinuity points, which provides evidence that the metro opening was not chosen on a special date. This discussion is necessary considering the importance of the forced variable.

Table 8 Weather, weekend, and holiday dummy control smoothness tests

Dell et al. (2018) suggest that when using RD for empirical analysis, if the absolute value of the change in a variable is small compared to the mean value, then the results of the RD analysis would not be affected by the change in this variable, even when the change is statistically significant. Noting the descriptive statistics in Table 1, the mean wind speeds before and after the opening of the metro are 7.0969 and 6.3879, respectively, while the full sample mean is 6.7674. That is, the wind speed after the opening of a metro is lower than the wind speed before the opening of the metro during the sample period. The wind speed tends to be negatively correlated with the pollution level, and the reduction in wind speed weakens the diffusion of pollutants to a certain extent. Therefore, if we can observe the reduction of pollutants at the discontinuity point, then the actual treatment effect should be greater; thus, the discontinuity of the wind speed will not diminish the credibility of our analysis results.