Exploring the Carbon-Mitigation Effect of High-Speed Railway and Its Underlying Mechanism

Abstract

Existing studies on high-speed railway (HSR) and carbon dioxide (CO₂) emissions have focused on analyzing their net effect, paying less attention to HSR’s impact mechanism on CO₂ emissions. This paper investigates the influence and underlying mechanism of HSR on CO₂ emissions. We apply a time-varying difference-in-differences approach to panel data from 283 cities in China between 2003 and 2016 to analyze HSR’s carbon-mitigation effects. A causal three-path mediation model and bootstrap method are utilized to analyze how HSR contributes to CO₂ mitigation. The results show that HSR operations have significant carbon-mitigation effects, especially in the eastern regions and large cities of China. Moreover, the carbon-mitigation effect of HSR tends to amplify over time. Mechanism analysis showed that HSR’s carbon-mitigation effects are realized through four pathways: transportation substitution, economic agglomeration, industrial structuring, and technological innovation, with the last one contributing the most. This paper provides practical policy recommendations for the next phase of carbon governance in China.

1. Introduction

Massive CO₂ emissions are causing global warming, seriously threatening people’s daily lives and production activities [1,2]. The Paris Agreement sets the goal of controlling global temperature rise at 2 °C by the end of this century while proposing to strive to achieve the goal of 1.5 °C to address the increasingly severe problem of climate change. It encourages countries to participate in the global response to climate change through nationally owned contributions. China has actively responded to global climate change as the world’s second-largest economy. In September 2020, China pledged to increase its nationally owned contributions, strive to achieve a carbon peak by 2030, and achieve carbon neutrality by 2060. However, China is undergoing fast-paced industrialization with growing demands for energy [3,4], creating tremendous pressure for emission mitigation in the long run. Achieving low-carbon development while maintaining economic growth has become a significant challenge for the Chinese government.

The transportation industry is essential for the economy but produces considerable CO₂ emissions. In China, transportation contributes to around 18% of the country’s overall CO₂ emissions [5]. Therefore, emission mitigation in the transport sector is of critical importance in the current context in China. High-speed railway (HSR) is one of the most essential long-distance transport modes and the most desirable green transport infrastructure today. In terms of energy preservation, HSR’s energy consumption per person for 100 km is only 18% of the energy consumption of an aircraft and around 50% of that of a bus. In terms of environmental conservation, HSR’s CO₂ emissions are only 6% of an airplane’s and 11% of a car’s under the same transport conditions (Data source: https://www.gov.cn/xinwen/2021-09/25/content_5639291.htm, 6 August 2023). In 2016, the operating mileage of HSR in China reached 22,000 km, accounting for approximately 65% of HSR’s global operating mileage. China’s transportation network, known as the “four vertical and four horizontal”, has begun to take shape (see Figure 1). During the year, China’s State Council updated the Medium and Long Term Railway Network Plan, which aims to build 38,000 km of HSR mileage by 2025 and form an HSR network covering most counties by 2030. Given the rapid expansion of HSR in China and the increasingly severe environmental situation, investigating the impact of HSR on CO₂ emissions is of great theoretical value and practical importance.

HSR may mitigate CO₂ emissions in four pathways. First, HSR is more energy efficient and emits fewer pollutants than traditional transportation methods [6,7]. Using HSR instead of traditional modes will enable the transportation sector to decrease its CO₂ emissions greatly. Second, HSR can enhance economic agglomeration by promoting factor mobility and reconfiguration [8], which may subsequently affect CO₂ emissions from economic activities. Third, HSR can facilitate industrial restructuring by changing the spatial association and layout of factors [9,10], accelerating industrial structures’ green and low-carbon transformation. Finally, HSR brings more knowledge and technology spillover [2], contributing to developing cleaner production technologies [11] and reducing pollutant emissions. However, previous studies have focused on HSR mainly as a construction project, examining its total lifecycle CO₂ emissions or its impact on substitute modes [12,13]. Little research has considered the impact of HSR on CO₂ emissions through, for example, its economic impact on industrial structure and technological advances. The crucial question of the primary pathways through which HSR affects CO₂ emissions has not been addressed.

To address this gap, this paper considers the operation of HSR in Chinese cities as a quasi-natural experiment. It applies the difference-in-differences (DID) approach to explore the effects and mechanisms of the operation of HSR on CO₂ emissions based on panel data from 283 Chinese cities from 2006 to 2016. We attempt to address the following issues: Has the operation of HSR successfully mitigated CO₂ emissions? If so, does the degree of carbon mitigation differ among cities of various locations and sizes? What are the specific mechanisms by which HSR contributes to carbon mitigation, and which pathway has the most significant impact? Answering the above questions will allow us to understand the influence and mechanisms of HSR on CO₂ emissions and propose practical suggestions concerning CO₂ governance in the era of HSR.

The contributions of this study are as follows. First, the results of this paper provide insights into the mechanisms by which the operation of HSR affects the cities’ carbon emissions both directly and indirectly. Most studies focus on HSR’s direct impacts on carbon emissions, while few address its indirect economic effects. Our study contributes to a better understanding of HSR’s role in mitigating urban CO₂ emissions. Second, this paper creatively combines the three-path mediation model and bootstrap approach to validate and compare mediated paths, thereby improving the robustness of the mechanism analysis results and enabling the comparison of mediating path effect. We draw an important conclusion that technological innovation is the primary way HSR mitigates CO₂ emissions in cities. Third, this paper uses CO₂ emission data based on nighttime lighting data, which are closer to the actual value than the previous urban CO₂ emissions measured through urban energy consumption. Our results reflect the actual impact of HSR on CO₂ emissions relatively accurately. Finally, this paper may shed light on the HSR line planning and urban carbon governance. We find that HSR’s carbon-mitigation effects vary by city scale and location through heterogeneity analysis. These findings imply that government should promote the development of HSR to the west region and large cities.

The remainder of the paper is structured as follows. Section 2 presents the literature on the correlation between HSR and CO₂ mitigation. The theoretical analysis is presented in Section 3. Section 4 describes the construction of the DID model and the data. Section 5 presents the empirical analysis, which includes benchmark regression, heterogeneity analysis, and robustness tests. Section 6 presents the mediation analysis. The findings and suggestions are presented in Section 7.

2. Literature Review

Current research related to CO₂ emissions falls into two main categories. One category focuses on measuring and predicting CO₂ emissions at different levels, while the other concentrates on analyzing the factors affecting CO₂ emissions. Most studies seek to identify the factors influencing CO₂ emissions regarding environmental regulation, urbanization, trade opening, and industrial structure [14,15,16,17]. Transportation infrastructure construction, operation, and maintenance emit CO₂ emissions directly, and relevant research centers on measuring the emissions of one mode of transport or comparing the emissions of several modes per unit mile [18,19]. Transportation infrastructure also indirectly affects CO₂ emissions by influencing economic activities. Thus, related studies in this area are usually based on indicator analysis to determine the economic factors affecting CO₂ emissions, exploring the overall impact of transportation and economic factors [20,21].

As an emerging transportation infrastructure, HSR may affect CO₂ emissions through direct transportation substitution effects or indirectly through a range of economic impacts. From the direct mechanism’s perspective, HSR has environmental benefits, including low energy use, pollution emissions, and transportation costs [22]. Therefore, it can mitigate CO₂ emissions by replacing existing modes of transport [7]. From the indirect mechanism’s perspective, HSR can significantly inhibit inter-regional transaction costs [23] and promote the interregional mobility of resource elements and intermediate inputs, thus optimizing resource allocation and inhibiting energy losses [24,25]. Hence, HSR can indirectly affect CO₂ emissions through its economic impacts.

However, most previous studies have conducted mainly qualitative or descriptive analyses of HSR’s effect on economic factors [20,26,27]. Limited research has explored the influence of HSR on CO₂ emissions, with a prevailing emphasis on HSR’s direct impact. For instance, reference [7] found that replacing civil air travel with HSR in China has resulted in an 18% drop in CO₂ emissions. Many researchers have proposed that the indirect impacts of transport infrastructure are more profound than their direct impacts and argued that transport infrastructure would reduce CO₂ emissions through economic effects [28]. In China’s low-carbon green development, several studies have recently explored the mechanisms by which HSR affects CO₂ emissions in terms of indirect effects but did not reach a consistent conclusion. For instance, reference [29] argued that HSR affects CO₂ emissions through industrial upgrading and economic integration. Reference [30] discovered that HSR affects CO₂ emissions only by improving the urban innovation level. These studies focus on verifying the existence of various mediating pathways, neglecting how much the mediating pathways contribute to HSR’s carbon-mitigation effects. Therefore, which mechanism has the greatest impact on HSR’s carbon reduction effects is also not clear.

To conclude, the factors influencing CO₂ emissions are numerous and complex. While many scholars have studied how transportation affects CO₂ emissions, few studies have focused on the influence of HSR on CO₂ emissions. Moreover, previous studies have mainly examined HSR’s direct impact, neglecting its influence on CO₂ emissions through economic effects. This paper finds that HSR has the mechanical characteristics of reducing CO₂ emissions and that HSR can reduce CO₂ emissions through transportation substitution, economic accumulation, industrial structuring, and technological innovation. This paper begins by studying how HSR affects CO₂ emissions. The following section of the paper offers a theoretical analysis of the four pathways.

3. Theoretical Analysis

In recent years, HSR has influenced various factors, from transportation modes to economic development, industrial structure, and urban innovation, impacting countries’ CO₂ emissions. This study analyzes the multiple transmission pathways of HSR to reduce CO₂ emissions from four aspects: traffic substitution, economic agglomeration, industrial structuring, and technological innovation. Figure 2 depicts the theoretical analytical framework.

3.1. Transportation Substitution

With obvious benefits in terms of timeliness, punctuality, safety, and comfort, HSR is becoming increasingly popular among interstate travelers and creating a direct substitution for road and air transportation [7,31]. Some studies have shown that HSR can effectively replace 25–50% of intercity transport in Europe [32]. HSR impacts CO₂ emissions through the substitution of transportation by optimizing the traffic structure and alleviating traffic congestion. Substituting HSR for existing transportation modes will optimize city traffic structures and minimize the pollution the transportation sector produces due to its low pollutant emissions [33]. The traffic substitution effect can relieve the strain on road and aviation traffic, diminishing emissions from exhaust and pollutants resulting from congestion and ultimately mitigating CO₂ emissions [34]. For instance, reference [35] found that the expansion of China’s HSR network reduced freight greenhouse gas emissions equivalent to 18.999 million tonnes of CO₂. Moreover, some papers analyze the influence of HSR on environmental pollution from a lifecycle view and find a reduction in greenhouse gas emissions when HSR replaces highways and air travel [36].

3.2. Economic Agglomeration

The factor flow and reconfiguration brought by HSR will affect economic agglomeration in cities along the route [37]. The time–space compression effect of HSR inhibits cross-regional transaction costs, strengthens economic ties between HSR-opening cities and surrounding areas, and helps attract corporate investment and population clustering [38]. For firms, the opening of HSR allows them to search for suppliers and customers at a lower cost, thereby increasing corporate profits [39]. For individuals, the time cost of commuting to major cities is lower with HSR opening, and they can enjoy improved public services and obtain more suitable jobs [23]. Similar to the obverse side of the coin, economic agglomeration plays a dual role in CO₂ emissions. It exerts a positive influence by reducing production costs and enhancing production efficiency through factor sharing [8]. It also brings negative externalities. The “congestion effect” from over-agglomeration can expand the population size and production scale, consume more energy, and produce more pollutants [40]. Therefore, the connection between economic agglomeration and CO₂ emissions is not definite.

3.3. Industrial Structuring

HSR operation weakens the barriers to factor mobility and facilitates resource allocation of production factors on a broader scale [41]. Because HSR is used mainly for passenger transportation, it significantly influences industries that rely on highly mobile production components, such as the service sector [9]. Developing services is critical to industrial restructuring and optimization [42]. Relevant literature shows that HSR promotes the growth of the tertiary sector, notably the service sector, while inhibiting the share of the secondary industry [43]. Industrial structuring has a huge impact on low-carbon development. Some scholars have found that CO₂ emissions positively correlate with the secondary sector and negatively correlate with the tertiary sector [44,45]. Therefore, industrial structuring effectively inhibits CO₂ emissions. Industrial restructuring can contribute up to 60% towards China’s carbon intensity targets [46]. Therefore, the industrial structuring effect caused by HSR can help inhibit CO₂ emissions.

3.4. Technological Innovation

HSR can drive urban innovation by promoting the clustering of innovation factors and improving the innovation environment. First, HSR may boost the number and scale of firms in cities along the route, increase the attraction of senior talent to cities, and accelerate knowledge and technology spillover between firms [47]. Second, HSR expands the market scale faced by firms, which enhances the intensity of competition and helps firms learn from those with high innovation efficiency [9]. Technical innovation influences CO₂ emissions in two ways. First, green technology-oriented innovation can boost the new energy sector, which benefits the green transformation of the energy consumption structure [48,49]. Low-carbon technological innovation can save decarbonization costs and empower pollution treatment at the end of the environment [50]. Second, as technology advances, the economy will gradually shift from a growth mode where energy is the primary factor input to knowledge is the primary factor input [51], ultimately achieving low-carbon economic growth.

4. Data and Model

4.1. Data and Sample

This paper selected data from 283 cities in China as the research sample from 2003 to 2016 (excluding Hong Kong, Macao, and Taiwan). The sample selection process was as follows. (1) Cities that underwent administrative restructuring at the prefectural level during the study period, such as Chaohu and Sansha, were excluded. The study sample still includes cities where administrative division adjustment occurred below the prefecture-level city level. (2) Cities whose economic data were not included in the yearbook or had large missing data were excluded. City-level data were obtained mainly from the statistical yearbooks of provinces.

4.2. Variable Selection

4.2.1. Dependent Variable

CO₂ emissions (CE). In most studies, CO₂ emissions are calculated based on energy consumption, leading to a lower estimate than what is realistic. This study uses the defense meteorological satellite program/operational linescan system and national polar-orbiting partnership/visible infrared imaging radiometer suite nighttime lighting data provided by the National Physical Earth Data Center. We extract county-scale emission data as the base sample and calculate the city-level data. The CE variable is given as the log amount of the city’s CO₂ emissions.

4.2.2. Key Independent Variable

HSR operation (HSR). HSR is a binary indicator for the operation of HSR. We collect the commissioning of HSR lines from the website of the China Railroad Corporation. Then, we manually gather and organize the actual year of HSR opening in different cities throughout the country. This paper uses 2008 as the year China officially launched its HSR system. We first define whether a city has an HSR line based on whether the city has an HSR station. Second, because some cities have more than one HSR line, we use the opening time of the city’s first line. Third, many cities opened HSR lines after June or even in December, making HSR’s treatment effect difficult to determine effectively that year. Therefore, we take June 30 as the cut-off date and define the opening of HSR before that date as that year’s operation; after that, the operation year lags by one year.

4.2.3. Mediation Variables

Transportation substitution (TS). HSR breaks residents’ dependence on highways and civil aviation for travel, creating a substitution effect. Following the method of reference [32], TS is calculated by using the logarithm of the passenger traffic on highways and in airplanes.

Economic Agglomeration (EA). Urban economic activities depend more on non-agricultural sectors, such as secondary and tertiary industries. We adopt the approach presented by reference [52], which uses employment density to quantify the aggregation level of economic activity. EA is obtained by dividing the urban secondary and tertiary employment by the city’s built-up area.

Industrial structuring (IS). Increasing the tertiary sector’s share can inhibit emissions because the tertiary sector consumes fewer fossil fuels and electricity than other sectors [45]. Therefore, the IS variable is obtained by dividing tertiary sector value added by secondary sector value added.

Technological innovation (TI). Following reference [29], the city innovation capacity index represents technological innovation. These data are taken from the China Cities and Industries Innovation Power Report 2017.

4.2.4. Control Variables

Economic development (PG) is defined as a city’s actual gross domestic product (GDP) capitation while adding its squared term. Population density (PD) is obtained by dividing the year-end population of a city by its land area. Science and education level (SE) is the share of budgeted science and education expenditures in the budgeted fiscal expenditures. Environmental regulation (ER) is a city’s general industrial solid waste comprehensive utilization rate. Infrastructure construction (IC) is quantified as the logarithm of the road area per capita. Foreign direct investment (FDI) refers to the ratio of actual foreign investment utilized to the city’s GDP.

Table 1. Results of descriptive statistics.

Variable	Unit	Obs.	Mean	Std. Dev.	Min	Max
CE	megaton	3962	2.874	0.807	0.425	4.71
HSR	-	3962	0.170	0.376	0.000	1.00
TS	104 person	3962	10.369	2.842	0.020	19.50
EA	104 person/km2	3962	0.767	1.222	0.020	16.17
IS	-	3962	0.833	0.414	0.090	4.17
TI	-	3962	5.096	15.92	0.000	114.78
PG	104 yuan/person	3962	2.290	1.826	0.010	9.87
PD	104 person/km2	3962	0.423	0.324	0.005	2.638
SE	%	3962	19.747	4.571	1.990	30.87
ER	-	3962	78.025	24.139	0.000	115.00
IC	person/m2	3962	2.509	0.480	−0.942	3.50
FI	%	3962	0.021	0.022	0.000	0.11

presents the results of descriptive statistics.

4.3. Empirical Model

Our basic model is a time-varying DID regression model based on differences in HSR headways across cities. The city dummy variable (denoted as “du”) takes 1 for HSR cities and 0 for other cities. Likewise, the year dummy variable (marked as “dt”) is assigned a value of 1 during HSR operation and 0 otherwise. Based on the above, we set HSR = du $\times$ dt, where HSR is a dummy variable for HSR operation. Its estimated coefficient describes the difference between the influence of HSR on the treatment and control groups. We also performed a Hausman test on the model, and the p-value was 0.0000. Thus, we chose fixed effects to identify the influence of HSR on CO₂ emissions. The model is specified as follows:

$${CE}_{it}=\alpha +\beta {HSR}_{it}+\gamma X_{it}+{\epsilon }_i+{\mu }_t+{\eta }_{it}$$

(1)

where i represents city; t expresses year; ${CE}_{it}$ denotes CO₂ emissions; ${HSR}_{it}$ denotes HSR opening; $X_{it}$ indicates the set of control variables; ${\epsilon }_i$ and ${\mu }_t$ denote individual and time-fixed effects, respectively; and ${\eta }_{it}$ denotes the error term.

5. Empirical Analysis

5.1. Parallel Trend Test

Meeting the parallel trend assumption is crucial to estimating DID accurately. Drawing on reference [53], we include interaction terms in the model to test whether the parallel trend is valid and examine how HSR treatment effects evolve. The model settings are as follows:

$${CE}_{it}=\alpha +{\sum }_{k\ge -5, k\ne -1}^5{\beta }_k{HSR}_{it}^k+\gamma X_{it}+{\epsilon }_i+{\mu }_t+{\eta }_{it}$$

(2)

where ${HSR}_{it}^k$ denotes the “event” of HSR operation. The assignment of ${HSR}_{it}^k$ is as follows: use $s_i$ to denote the specific year when the city HSR opens; if t − $s_i$ ≤ −5, then define ${HSR}_{it}^{-5}=1$; otherwise, ${HSR}_{it}^{-5}=0$. If t − $s_i$ = k, then define ${HSR}_{it}^{5+}=1$; otherwise, ${HSR}_{it}^{5+}=0$ (k ∈ [−5, 5] and k ≠ −1). If t − $s_i$ ≥ 5, then ${HSR}_{it}^{5+}=1$ is defined; otherwise, ${HSR}_{it}^{5+}$ = 0. Regarding the setting of the first five periods before and after, because of the small size of the sample of the earliest cities with HSR (the earliest city with HSR is 2008, and thus, there are eight periods before opening), the first eight periods are categorized as the first five periods. The baseline year used is the year before HSR opening and the dummy variable k = −1 is removed from the equation. The other variables in Equation (2) are the same as in the basic model (1).

In Figure 3, the timeline is represented on the x-axis, showing the years leading up to and following the start of HSR operations. For instance, −2 represents the second year before the HSR operation and 2 represents the second year after the HSR operation. CO₂ emissions in the control and treatment groups have consistent temporal trends without HSR operation, indicating that the DID method in our study complies with the common trend hypothesis. We also find that the HSR’s carbon-mitigation effects increase over time.

5.2. Benchmark Regression

In the benchmark regression analysis, model (1) is used to test the impact of HSR on CO₂ emissions. The regression results are shown in columns (1) and (2) of

Table 2. Regression results.

	(1)	(2)	(3)	(4)	(5)
HSR	−0.0411 ***	−0.0452 ***	−0.0362 ***	−0.0408 ***	−0.0588 ***
	(0.0056)	(0.0055)	(0.0055)	(0.0054)	(0.0217)
PG		0.0228 ***		0.0244 ***	0.0096 ***
		(0.0048)		(0.0049)	(0.0020)
PG2		−0.0017 ***		−0.0020 ***	−0.0002 ***
		(0.0005)		(0.0005)	(0.0001)
PD		0.3374 ***		0.3730 ***	0.2798 ***
		(0.0492)		(0.0500)	(0.0446)
SE		−0.0018 ***		−0.0019 ***	−0.0018 ***
		(0.0006)		(0.0006)	(0.0006)
ER		−0.0003 **		−0.0003 ***	−0.0003 ***
		(0.0001)		(0.0001)	(0.0001)
IC		−0.0050		−0.0064 *	−0.0046
		(0.0039)		(0.0039)	(0.0035)
FI		−0.2863 **		−0.4909 ***	0.0003 ***
		(0.1431)		(0.1398)	(0.0000)
Constant	2.8807 ***	0.9898 ***	2.8799 ***	0.7895 ***	2.0082 ***
	(0.0018)	(0.2814)	(0.0018)	(0.2867)	(0.2897)
City FE	YES	YES	YES	YES	YES
Year FE	YES	YES	YES	YES	YES
N	3962	3962	3881	3881	3962
Adj-R2	0.9864	0.9867	0.9871	0.9876	0.9884
Kleibergen–Paap rk Wald F statistic					288.048

Note: Standard errors in parentheses; * p < 0.1, ** p < 0.05, *** p < 0.01. To facilitate comparison and robustness analysis, we present the results of the benchmark regression alongside those of the robustness test and endogeneity test. Columns (3) and (4) display the regression outcomes obtained by using the PSM-DID method, while column (5) shows the regression results of the instrumental variables.

. Column (1) is the result without control variables, and column (2) is the result after adding control variables. The coefficient estimates of HSR are significantly pessimistic at the 1% level regardless of whether control variables are added. The results in column (2) show that by controlling other factors constant, HSR cities decrease CO₂ emissions by 4.52% on average relative to non-HSR cities. Additionally, the coefficient estimates for PG and its quadratic term show positive and negative correlations, respectively, indicating that China’s economic growth and CO₂ emissions have an inverted U-shaped correlation.

5.3. Robustness Test

5.3.1. PSM-DID Method

Although the parallel trend test above has justified the DID method, it may be biased due to the non-random nature of establishing HSR cities. Therefore, we employ the DID-propensity score matching (PSM-DID) method for further testing. During the propensity score matching process, we employ a probit model to estimate the “propensity values” of the treatment and control groups. The matching method is caliper nearest neighbor matching (k = 3). The model is designed as follows:

$$p_i\left(X\right)=Prob\left(HSR=1|X_i\right)=f\left[g\left(X_i\right)\right]$$

(3)

where $p_i\left(X\right)$ is the propensity value for city i having HSR; $f$ is a logit function; $X_i$ is a set of matching vectors, which we denote by selected control variables; and $g$ denotes a linear function.

Based on the general principles of using the PSM-DID method, we perform a balancing test.

Table 3. Results of propensity score matching balance test.

Variable	Unmatched	Mean		Bias	t-Test	p > |t|
	Matched	Treated	Control	(%)	Value
PG	U	2.5830	2.1076	27.6	5.89	0.000
	M	2.5830	2.6035	−1.2	−0.17	0.868
PD	U	6.0098	5.7064	39.4	7.56	0.000
	M	6.0098	6.0129	−0.4	−0.07	0.947
SE	U	20.2870	19.6830	13.9	2.70	0.007
	M	20.2870	20.3070	−0.4	−0.07	0.946
ER	U	85.0640	76.8880	35.7	6.96	0.000
	M	85.0640	83.9610	4.8	0.83	0.409
IC	U	2.4549	2.5113	−12.0	−1.85	0.016
	M	2.4549	2.4373	3.7	0.51	0.573
FI	U	0.0195	0.0209	−7.4	−0.46	0.183
	M	0.0195	0.0193	0.8	0.18	0.893

reports the differences in city attributes with and without HSR before and after propensity score matching. The absolute standard deviations of all matching vectors after matching are less than 5%, indicating that the matching vectors and methods selected for the sample are reasonable.

We estimate model (1) again using the data after propensity score matching. The PSM-DID analysis outcomes are presented in columns (3) and (4). The estimated coefficients of HSR are significantly negative regardless of whether control variables are added, which is consistent with the results of the regressions using DID method, further validating the core conclusion of this paper that HSR significantly decreases CO₂ emissions.

5.3.2. Placebo Test

Other exogenous factors, unrelated to the inauguration of HSR, are likely to impact CO₂ emissions during the study period. We conduct a placebo test using separate repeatable tests to determine whether external effects exist. Drawing on the reference [54], we perform a placebo test by randomly selecting the HSR opening time. To reduce the disturbances of small probability events, we conducted 500 independent replicate samples of the sample data and regressed them following model (1). Figure 4 illustrates the kernel density distribution of the estimated coefficients and their corresponding p-values. The estimated coefficients have a distribution close to zero, and most regression results are insignificant. The true estimated coefficient is a significant outlier in the independent repetition experiments, indicating that the benchmark regression result is a small probability event in the placebo test. Therefore, our results being due to unobservable factors can be excluded.

5.4. Endogeneity Test

This paper employs the instrumental variables approach to address possible endogeneity issues. Referring to reference [55], using the minimum geographical open cost principle, we construct minimum spanning trees using appropriate geographical information, such as slope, hydrology, and undulation. The logic of corresponding processing is as follows. First, geographical development costs are a crucial basis for influencing the direction of the HSR network and meeting the relevance requirement. Second, topographic and geomorphic features that determine the cost of geographical development are established objective facts and meet the homogeneity requirement. Because geographical information remains constant, we use the minimum spanning tree multiplied by the year dummy variable as the instrumental variable.

We introduce the instrumental variable into model (1) and use a Two-Stage Least Square approach to discuss the robustness of the baseline regression results. Table 2, column (5), displays the outcomes of the second stage estimation. The regression results of instrumental variables are consistent with the benchmark regression. The Kleibergen–Paap rk Wald F-value is 288.048, much larger than the empirical value of the proposed correlation instrument 10, indicating the validity of instrumental variables. This result proves that the benchmark results are robust and statistically not endogenous.

5.5. Heterogeneity Analysis

5.5.1. Regional Heterogeneity

Due to economic conditions and population density, China’s HSR network coverage is highest in the east, second in the center, and lowest in the west. Thus, we divide the sample into three regions (variable symbols are East, Mid, and West). As shown in columns (1), (2), and (3) of

Table 4. Heterogeneity test of HSR on CO₂ emissions.

	(1)	(2)	(3)	(4)	(5)
HSR × East	−0.0636 ***
	(0.0073)
HSR × Mid		−0.0067
		(0.0065)
HSR × West			−0.0047
			(0.0142)
HSR × Large				−0.0466 ***
				(0.0056)
HSR × Small					−0.0028
					(0.0184)
Control	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES
Time FE	YES	YES	YES	YES	YES
N	3962	3951	3951	3962	3962
Adj-R2	0.9868	0.9865	0.9865	0.9867	0.9865

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

only the HSR × East coefficient is prominently negative, implying that the carbon-mitigation effect of HSR is merely prominent in the eastern regions. The finding is due to the late HSR construction in Midwest China, and its emission reduction effect has not been fully revealed. Moreover, some central and western cities are sparsely populated and far from developed regions. Thus, even if HSR is opened, its substitution effect on other transportation modes is relatively small.

5.5.2. Scale Heterogeneity

Considering that different city sizes may affect HSR’s carbon-mitigation effects, we classify all cities into two categories following the Notice of the State Council on the Adjustment of City Size Standards released in 2014. We use the population size of 1 million as the dividing line and construct relevant dummy variables (variable symbols are Large and Small). Results from columns (4) and (5) show that only the HSR × Large coefficient is prominently negative, suggesting that the carbon-mitigation effects of HSR are only pronounced in larger cities, which may be due to two reasons. First, cities with large population sizes tend to gather more highly skilled talents, which can more efficiently utilize the technological innovation effect of HSR opening, thus effectively exerting the emission reduction effect [32]. Second, small and medium-sized cities are limited by their development. Their acceptance and integration of HSR are much less than that of large cities in the short term, making it difficult to reveal the emission reduction effect.

6. Mediation Analysis

6.1. Mediation Model

Our theoretical analysis argues that HSR may affect CO₂ emissions through four pathways. Therefore, we validate the mediation pathways in this section. Using a three-path mediation model, we first identify and test the potential mediation pathways. Following the “mediating effect” model designed by [56], this paper establishes the following mediation model:

$${CE}_{it}={\alpha }_1+{\beta }_{1}HSR+{\gamma }_1{X}_{it}+{\epsilon }_i+{\mu }_t+{\eta }_{it}$$

(4)

$${Med}_{it}={\alpha }_2+{\beta }_{2}HSR+{\gamma }_2{X}_{it}+{\epsilon }_i+{\mu }_t+{\eta }_{it}$$

(5)

$${CE}_{it}={\alpha }_3+{\beta }_{3}HSR+{\theta Med}_{it}+{\gamma }_3{X}_{it}+{\epsilon }_i+{\mu }_t+{\eta }_{it}$$

(6)

where ${Med}_{it}$ denotes the possible mediating variables, including TS, EA, IS, and TI. The remaining variables remain unchanged from model (1). The results of the mediation effects are shown in

Table 5. Mediation analysis.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
	CE	TS	CE	EA	CE	IS	CE	TI	CE
HSR	−0.0452 ***	−0.1367 **	−0.0444 ***	0.0400 **	−0.0438 ***	0.0234 **	−0.0442 ***	9.6932 ***	−0.0244 ***
	(0.0055)	(0.0643)	(0.0055)	(0.0201)	(0.0055)	(0.0113)	(0.0055)	(0.6596)	(0.0053)
TS			0.0063 ***
			(0.0021)
EA					−0.0353 ***
					(0.0057)
IS							−0.0436 ***
							(0.0115)
TI									−0.0022 ***
									(0.0002)
Control	YES	YES	YES	YES	YES	YES	YES	YES	YES
City FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
Time FE	YES	YES	YES	YES	YES	YES	YES	YES	YES
N	3962	3962	3962	3962	3962	3962	3962	3962	3962
Adj-R2	0.9867	0.8767	0.9868	0.9196	0.9870	0.8244	0.9868	0.6747	0.9873

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

.

The HSR coefficient in column (1) is prominently negative, indicating that HSR operation notably reduces highway and airline traffic. According to column (3) in Table 5, the TS coefficient is prominently negative, indicating that the increase in total passenger traffic by highways and airlines significantly increases CO₂ emissions. These results suggest that the replacement of traditional transport modes by HSR can decrease CO₂ emissions. Therefore, transportation substitution is one of the mechanisms by which HSR mitigates CO₂ emissions.

The HSR coefficients in columns (4), (6), and (8) are prominently positive, showing that HSR operation stimulates economic aggregation, industrial structure adjustments, and technology innovation. The estimated coefficients for EA, IS, and TI in columns (5), (7), and (9) consistently demonstrate a negative trend, indicating that these economic factors could hinder CO₂ emissions. The results suggest that the mediation effects of economic agglomeration, industrial structuring, and technological innovation are all present.

6.2. Bootstrap Method

We employ the bootstrap method to verify the significance level of the mediating effect and measure the effect sizes of the four transmission pathways. The bootstrap 1000 repetitions sampling method is adopted, and a 95% confidence interval is set.

Table 6. Test results of the bootstrap method.

Variable	Indirect	Effect	Z Value	p > |Z|	Normal-Based [95% Conf. Interval]
	Direct				Lower	Upper
TS	I	0.1400	1.97	0.049	0.0005	0.2795
	D	14.4993	14.30	0.000	12.5124	16.4862
EA	I	0.0933	8.65	0.000	0.7218	0.1145
	D	0.4686	16.41	0.000	0.4127	0.5246
IS	I	0.0141	2.79	0.005	0.0042	0.0240
	D	0.5584	19.04	0.000	0.5001	0.6159
TI	I	0.2843	16.12	0.000	0.2497	0.3189
	D	0.2776	9.25	0.000	0.2188	0.3364

Note: I denotes indirect/mediated effect results; D denotes direct effect results; (I/(I + D)) denotes the proportion of mediation effects.

shows that the 95% confidence intervals for TS, EA, IS, and TI do not contain the number zero, demonstrating that the mediating effects of these pathways are all significant. The mediation effects of TS, EA, IS, and TI account for 0.95%, 16.60%, 2.46%, and 50.60% of HSR’s total carbon-mitigation effects, respectively. The results indicate that HSR inhibits CO₂ emissions primarily by affecting economic factors, particularly technological innovation.

7. Conclusions and Policy Implications

This study employs a quasi-natural experimental design to examine the carbon-mitigation effects of HSR and its underlying mechanism. We utilize the DID method to assess its carbon-mitigation effects and analyze how these effects vary in different cities and regions. Further, we construct a three-path mediation model and use a bootstrap approach to verify the underlying mechanisms. The primary conclusions are as follows.

First, HSR operations have prominent carbon-mitigation effects, and the effects tend to increase over time. Compared to non-HSR cities, CO₂ emissions in HSR cities decreased by an average of 4.52%. Second, HSR’s carbon-mitigation effects are heterogeneous and more potent in eastern and large cities. Third, HSR achieves the carbon-mitigation effects through four mediation pathways: transportation substitution, economic accumulation, industrial structuring, and technological innovation, with the technological innovation pathway contributing the most, accounting for 50.60% of the total carbon-mitigation effects.

Based on the above conclusions, this paper provides some practical insights for policy making in HSR planning and carbon governance.

First, as a public good provided by the government, the main objective of HSR is to maximize social welfare. Although HSR construction and operating costs are costly, its welfare effects involve various aspects, such as the economy, the environment, and health. This paper provides evidence of the additional benefits of HSR from an environmental welfare perspective. Based on the core findings, we argue that HSR network density should increase continuously to enhance HSR’s environmental welfare effects.

Second, the construction of HSR should be promoted according to local conditions, and the HSR network should be reasonably expanded. China’s HSR network is distributed mainly in the east-central region. Thus, to achieve a balanced construction of HSR, the government should expand regional development and promote the development of HSR to the West. At the same time, the government should pay attention to the planning of HSR projects in large cities; in the short term, however, small and medium-sized cities should not mindlessly pursue the expansion of HSR construction.

Third, giving full play to the technological innovation effect of HSR, CO₂ emissions can be reduced by accelerating the dissemination of knowledge and technology. The government should actively break down the existing institutional and technological barriers between regions; promote the exchange and spillover of environmental protection measures, management experience, and emission reduction technologies from various regions; and continuously enhance the positive spatial spillover of carbon productivity.

Finally, HSR’s role in promoting industrial restructuring and economic clustering should be fully utilized to realize high-quality economic development and maximize its carbon-mitigation effects. HSR cities should fully use their location advantages, build industrial chains along lines, and lead industrial development with the HSR economy. Additionally, local governments should optimize urban infrastructure and business environments to promote resource intensification and economic clusters.

Despite the valuable insights presented in this paper, it has certain limitations that require attention in future studies. First, our evaluation of the carbon-mitigation effect of HSR systems focuses on the operational phase, excluding the emissions generated during the construction phase. The extraction and transport of raw materials for HSR and the construction process could increase CO₂ emissions. However, emission quantification at this stage is complicated, and the research caliber is often inconsistent [57,58], leading to difficulty in obtaining CO₂ emission data. Therefore, future studies could establish a framework for analyzing CO₂ emission quantification in the whole life cycle of HSR projects to investigate the overall impact of CO₂ emissions comprehensively. Second, we used the administrative division criterion of HSR stations to identify which cities they are located in, and this classification result may somewhat bias the results. Because some HSR stations in China are set at the border, non-HSR cities close to HSR stations in other cities might also gain from HSR. Limitations in the study design may underestimate the actual impacts of the HSR. Future studies need to consider the specific locations of HSR stations further.