The Environmental Benefits of New Energy Vehicle Promotion and Their Mediation Pathways: Evidence from Chengdu in China

Abstract

New energy vehicle promotion (NEVP) is of great significance for the green and low carbon development of urban transportation. Based on the panel data of new energy vehicle sales, carbon emissions, and air quality in Chengdu, China, from 2014 to 2024, this paper employs multiple linear regression, distributed lag and multiple mediation pathway models to empirically examine the environmental benefits of NEVP. A heterogeneity analysis is also conducted by integrating the distribution of charging stations across urban circles. The results show that: (1) In the multiple mediation pathway model, the total effect of NEVP includes direct effect and indirect effect. Based on the total effect, the total carbon emission from the effect of NEVP is reduced by about 3.95% of the total carbon emissions, and 40% of carbon emission within the transportation sector in Chengdu. NEVP in Chengdu has a significant direct emission reduction effect, accounting for about 39.80% of the total effect, with the annual average carbon emissions being reduced by about 432,800 tons, accounting for about 1.57% of the total carbon emissions in Chengdu. In terms of indirect effects, NEVP significantly reduces carbon emissions through three pathways: industrial structure upgrading (1.02%), green consumption transformation (1.12%), and technological innovation (0.25%). However, the benefits of NEVP on improving urban air quality are limited. (2) The lag effect analysis shows that the environmental benefits of NEVP exhibit distinct characteristics of time lag and long-term persistence. (3) The environmental benefits show significant sub-circle heterogeneity. As carbon emissions decrease, the air quality of the central urban zone (the first circle) and the suburbs (the second circle) improves significantly, while the impact on the outer suburbs (the third circle) is not significant. There is an imbalance in the layout of charging piles in Chengdu. This research offers empirical evidence and policy insights for the green and low carbon development of urban transportation.

1. Introduction

New energy vehicle promotion (NEVP) is of great significance for the green and low carbon development of urban transportation. The environmental benefits of new energy vehicles are manifested not only in instantaneous pollution emission reduction but also in long-term cumulative environmental benefits. For instance, life cycle carbon emissions (LCEs) have been continuously decreasing, the air quality in cities has gradually improved [1], and the environmental benefits brought about by the transformation of the energy structure have also increased [2]. A growing body of research indicates that NEVP significantly reduces direct carbon emissions in the transportation sector, delivering substantial direct environmental benefits. González et al. found that the adoption of small battery electric vehicles (BEVs) directly reduces CO₂ emissions and abatement costs [3]. Awan et al. examined the impact of new energy vehicles’ technological innovation on CO₂ emissions in the transportation sector [4]. Holland et al. investigated the effects of new energy vehicle promotion on air pollution control across 32 states of the United States, concluding that its direct environmental benefits are somewhat overestimated [5]. Ferrero et al. discovered that significant reductions in air pollution levels occur only when new energy vehicles directly replace fuel-powered vehicles [6]. Marmiroli et al. found that using small new energy trucks for urban freight transport can effectively reduce greenhouse gas emissions [7]. Teixeira and Sodré highlighted notable advantages of new energy vehicle promotion in carbon emission reduction [8]. Chengdu is an important economic hub in western China, the capital of Sichuan Province, and one of the first national-level new energy vehicle demonstration cities. Therefore, NEVP policies and practices implemented in Chengdu have significant typicality and reference value. The main contributions of this paper include three aspects:

(1)

The existing literature mainly empirically studies NEVP based on cross-country data or cross-province data. Based on city panel data, this paper explores the micro-transmission pathways of the environmental benefits of NEVP at the city scale by the multiple mediation model.

(2)

Previous studies regard cities as homogeneous spaces, ignoring the spatial heterogeneity of urban circles. Based on the data of charging piles, this paper studies the heterogeneity of the environmental benefits of NEVP among different urban circles, and finds that it has the characteristics of a decreasing circle.

(3)

Most existing studies focus on the WTW emissions as the outcome variable. This paper introduces the distributed lag model to test whether there is a time lag, and its long-term sustainability in the environmental benefits of the NEVP. We find that the environmental benefits of NEVP exhibit distinct characteristics of time lag and long-term persistence. Among them, carbon emission shows a significant decline in the current period, 1-period lag and 2-period lag, and long-term effect. The air quality shows a significant improvement in the 3-period lag and long-term effect.

The rest of this paper is as follows: Section 2 is the literature review. Section 3 is the current situation analysis. Section 4 presents the research hypotheses, model construction, data sources, and variable descriptions. Section 5 analyzes the results of the empirical model. Finally, based on the results of the data analysis, the conclusions and policy recommendations are provided in Section 6.

2. Literature Review

2.1. The Cross-Country Evidence of NEVP’s Environmental Benefits

Empirical studies based on samples from different countries have shown that NEVP has significant environmental benefits. Jochem et al. conducted a forecast and assessment of carbon dioxide emissions from new energy vehicles in Germany in 2030 and found that NEVP can effectively reduce carbon emissions in the transportation sector [9]. Trost et al. studied future changes in the vehicle energy structure in Germany and found that NEVP can mitigate the greenhouse effect in the long run [10]. Lane et al. used data from California to study the energy supply chain and charging scenarios of new energy vehicles, and concluded that new energy vehicles replacing fuel-powered vehicles yield certain direct environmental benefits before achieving a zero-emission new energy supply chain [11]. Kobashi et al. analyzed the energy system of Kyoto, Japan, and found that actively utilizing new energy vehicles for power storage would reduce carbon dioxide emissions by 60–74% [12].

2.2. The Implementation Pathways About Carbon Reduction Benefits of NEVP

Previous research has generally established that new energy vehicles hold significant advantages in direct carbon emissions compared to fuel-powered vehicles. NEVP yields certain direct environmental benefits, but there is some controversy over its implementation pathways. Early research mainly focused on well-to-wheel (WTW) emissions. Ke et al. analyzed the WTW emissions of new energy vehicles based on municipal-level data from Beijing and found that new energy vehicles did not exhibit significant advantages in terms of WTW emissions [13]. Girardi et al. conducted a WTW carbon emission assessment of new energy and fuel-powered vehicles based on Italian data and found that new energy vehicles possess greater potential for lower carbon emissions throughout their life cycle [14].

Relying solely on “well-to-wheel (WTW) emissions” to evaluate the environmental benefits of NEVP has certain limitations, as it fails to consider the substitution effect of new energy vehicles for fuel-powered vehicles and the benefits associated with streamlining carbon emission abatement processes. At present, research on the pathways toward realizing the environmental benefits of NEVP has taken into account the issues of road traffic congestion and charging station congestion. Ju et al. argue that current NEVP policies may exacerbate traffic congestion in Beijing [15]. Quddus et al. found that constructing large-scale charging stations and increasing battery storage capacity can mitigate road congestion and the “charging congestion” issue caused by NEVP [16]. Wu et al. contend that the current NEVP will lead to further deterioration of traffic congestion and power load problems, while a targeted charging subsidy scheme could partially alleviate these issues [17]. In addition, NEVP can facilitate the optimization of energy structures and promote industrial upgrading. Richardson and Dixon pointed out that NEVP can significantly upgrade the energy consumption structure and promote industrial restructuring, thereby reducing carbon emissions to achieve green development. In the process of advancing NEVP, consumers’ awareness of, and demand for, new energy vehicles has continued to grow, and the transformation of their green consumption concept has also become one of the important pathways to realizing environmental benefits [18,19]. Sheldon and Dua noted that market mechanisms can render new energy vehicle subsidy policies more targeted and enhance consumers’ willingness to purchase new energy vehicles [20]. Beak et al. demonstrated that NEVP can significantly boost consumers’ willingness toward green consumption, increase new energy vehicle sales, and thus generate environmental benefits related to energy conservation and emission reduction [21].

2.3. Regional Heterogeneity of NEVP’s Environmental Benefits

There are significant regional differences in the results of empirical research on the impact of NEVP on environmental benefits, and existing studies have failed to reach a consistent conclusion on the pathways through which such differences arise. Zhang et al. believed that differences in energy consumption structure are the main source of regional differences [22]. Zhang et al. compared the impact of differences in policy instruments across Germany, France, Japan, China, Norway and other countries on the environmental benefits of NEVP [23], and they found that differences in NEVP policies and their implementation are the main factors causing differences in environmental benefits. Tang et al. argued that the electric power structure, policy instruments and geographical factors are important factors affecting regional differences in environmental benefits [24].

2.4. Literature Summary

From a data perspective, existing studies include the analysis of time series data and panel data from a single region [2,3,4]. From the perspective of methodology, they also adopt single policy text analysis and mixed qualitative–quantitative analysis [10,11,23]. However, there are still some limitations:

(1)

There is a paucity of research on the specific pathways through which the environmental benefits of NEVP are realized, and the main research findings have focused on the environmental benefits of “direct emission reductions”.

(2)

There is a lack of research on samples from specific cities and their sub-circles, as research on the impact of NEVP on environmental benefits has mostly relied on cross-country or cross-provincial data.

3. Current Situation Analysis

3.1. The Sales Situation of New Energy Vehicles in Chengdu

Chengdu’s new energy vehicle market maintains a robust growth momentum. According to the 2025 Red Star Ranking of Chengdu’s auto market, 12 of the top 20 best-selling models are new energy vehicles, accounting for 60%, five of which are produced by BYD. Tesla, Xiaomi, Li Auto and other brands also showed strong market performance. By 2025, the ownership of new energy vehicles in Chengdu had exceeded 930,000, ranking among the top-tier cities in China, and has become one of the first batch of pilot cities nationwide for the full electrification of public sector vehicles. The spatial distribution of the four major new energy vehicle trading clusters in Chengdu, namely Airport Road, Sansheng Township, Yangxi Line, and Longtan Temple, is shown in Figure 1.

The Airport Road business district is located adjacent to Shuangliu International Airport and benefits from convenient transportation. It is characterized by high-end brand showcase 4S stores, including premium Tesla and NIO flagship stores. The Sansheng Township business district is situated near the Eastern New Area and benefits from substantial preferential policies to attract investment. As a result, it is primarily dominated by cost-effective brands such as BYD, GAC Aion, and Wuling Automobile. The Yangxi Line business district, a traditional automobile business district in Chengdu, forms a mixed sales model of “new energy and fuel vehicles”, relying on mature commercial facilities such as Jinniu Kaide. The Longtan Temple business district, located close to the Chenghua District industrial base, has benefited from the “Industrial Cluster Ecosystem Building and Chain Strengthening Projects” policy, attracting 4S dealerships of renowned manufacturers such as FAW-Volkswagen and Volvo. The district primarily sells Volkswagen ID series and Volvo new energy models. The details of the four business districts are shown in

Table 1. The four major new energy vehicles commercial districts in Chengdu.

Business District	Main Features	Representative Brands	Development Trend
Airport Road	High-end, internationalization	Tesla NIO BMW i-series	Focus on luxury new energy vehicles, equipped with supercharging stations, attracting high net worth customers
Sansheng Township	Policy pilot zone, emerging market	BYD GAC Aion Shenye Auto	Government subsidies favoring the creation of a “new energy vehicles mall” integrating culture and tourism experiences
Yangxi Line	Traditional transformation, hybrid sales	Xiaomi XPeng Ganghong	Fuel-powered vehicles dealers accelerating new energy vehicles transformation, offering one stop “oil to electric” services
Longtansi	Industrial synergy, supply chain advantage	Volkswagen ID series Volvo new energy vehicles	Leveraging local automobile manufacturing to develop a “front store–back factory” model, reducing logistics costs

.

3.2. Layout and Policy Support for the New Energy Vehicles Industry in Chengdu

3.2.1. Overall Layout of the New Energy Vehicles Industry Chain in Chengdu

Chengdu has achieved remarkable outcomes in the layout of the new energy vehicles industry chain and has built a complete industrial ecosystem covering upstream materials, midstream components, downstream vehicle manufacturing and aftermarket services. In the field of vehicle manufacturing, the industrial park hosts 28 vehicle enterprises, including FAW-Volkswagen, FAW-Toyota, Geely, Volvo, Dongfeng Peugeot-Citroën Automobiles (DPCA) and China National Heavy Duty Truck Group (CNHTC). In the key component sector, over a thousand enterprises are gathered such as Bosch, Johnson Controls, CATL, Zhongchuang Xinhang, and Dongfang Electric.

In terms of spatial layout, Chengdu has formed a “one core and multi-point” industrial development pattern with Longquanyi District as the core high-tech zone, and Qingbaijiang District, Pidu District, Xinjin District and other districts (cities) and counties as key hubs. As a traditional automotive industry cluster, Longquanyi District is accelerating its transformation toward new energy vehicles. The high-tech zone, leveraging its strengths in the electronic information industry, focuses on the development of intelligent connectivity and automotive electronics. Qingbaijiang District has demonstrated prominent strengths in hydrogen energy equipment manufacturing. Pidu District and Xinjin District have made strategic deployments in power battery and special purpose vehicle sectors. This differentiated and synergized spatial layout effectively avoids homogeneous competition within Chengdu and significantly enhances overall industrial efficiency.

3.2.2. Layout of Charging Piles in Chengdu

By the end of 2024, Chengdu had built over 170,000 charging piles, and the charging pile to vehicle ratio had been optimized to 3.9:169. At the same time, after the Chengdu Power Grid Rongyao Project improved the clean power supply capacity, the city’s share of clean energy consumption exceeded 60%, better than the national average, and further strengthened the carbon emission reduction benefits of new energy vehicles. The layout of charging piles in Chengdu is shown in

Table 2. The layout of charging infrastructure in Chengdu.

Administrative District	Charging Units (Count)	Range of Charging Piles (Units)	Total Power Range (kW)	Regional Characteristics
Chenghua District	11	3466–4350	207,960–261,000	Industrial transformation zone; CH-5 unit exceeds 1000 piles
Eastern New Area	14	2666–3345	159,960–200,700	Emerging area with uneven distribution
Dujiangyan City	6	832–1046	49,920–62,760	Tourist destination with prominent unit power capacity
High-Tech South Zone	5	2068–2603	124,080–156,180	Business center area with high power demand
High-Tech West Zone	2	1262–1582	75,720–94,920	Industrial park area with centralized large stations
Jianyang City	7	1092–1371	65,520–82,260	Suburban county/city with relatively concentrated distribution
Jinniu District	13	3768–4728	226,080–283,680	Transportation hub area
Jintang County	5	1902–2384	114,120–142,080	Ecological area with dispersed layout
Jinjiang District	11	2022–2538	121,320–152,280	Old urban area with mixed distribution
Longquanyi District	9	4865–6104	291,900–366,240	Automobile industry base; largest unit scale
Pidu District	6	2671–3353	160,260–201,180	University cluster area
Qingbaijiang District	5	2161–2712	129,660–162,720	Logistics center area
Qingyang District	12	2319–2910	139,140–174,600	Balanced distribution
Qionglai City	3	1304–1637	78,240–98,220	Industrial county/city
Shuangliu District	8	4033–5060	241,980–303,600	Airport radiation area
Tianfu New Area	9	2329–2920	139,740–175,200	Core development area with high-density coverage and large single-station power
Wuhou District	11	2716–3408	162,960–204,480	Commercial and residential mixed area
Xinjin District	8	1307–1640	78,420–98,400	Low-density suburban area

. The distribution of charging piles varies significantly in different regions due to their own characteristics: Chenghua District, a strong traditional industrial area, has over 1000 CH-5 units. Dujiangyan City, a tourist destination, prioritizes well-equipped charging infrastructure and features outstanding unit power. The southern high-tech zone, a business center, features high power demand. The western high-tech zone, an industrial park, adopts a centralized large-scale charging station layout. Jinniu District, a transportation hub, has a large number of charging facilities. Longquanyi District, a new energy vehicles industry base, boasts a leading scale of charging piles. Shuangliu District, an airport radiation zone, has a large number of high-powered charging facilities. Tianfu New Area, a core development zone, achieves high density coverage and features a high single-station power rating. The other suburban districts and counties exhibit a low density distribution pattern.

3.2.3. Support Policies for NEVP in Chengdu

Chengdu has introduced multi-dimensional policy support measures for NEVP (see

Table 3. Support policies for NEVP in Chengdu.

Policy Category	Specific Measures	Policy Source
Purchase Incentives	Provide production and sales rewards for new energy vehicles models, with a maximum reward of 50 million CNY per model	Implementation Opinions on Promoting the Development of the New Energy Vehicles Industry in Chengdu
	Continuing exemption from vehicle purchase tax (total exemption of 5.49 billion CNY in 2024)
	District-/County-level supporting subsidies (e.g., up to 5000 CNY “instant discount” in Qingyang District)
Convenience of Use	New energy vehicles are exempt from tail number traffic restrictions
	Public parking lots are free for up to 2 h, with half-rate charges after that
	Relaxed access permits for new energy vehicle freight vehicles (30 day electronic pass)
Industrial Synergy	Target of 50% local component matching rate by 2025
	Establishment of large local automotive groups (through mergers and acquisitions)
	Joint construction of the “hydrogen corridor” and “electric corridor” between Chengdu and Chongqing
Infrastructure	Build 160,000 charging piles and 3000 battery-swapping stations by 2025 Subsidies for public (or dedicated) charging stations based on star ratings (maximum 200,000 CNY per site) 100% of new residential buildings must reserve installation conditions for charging facilities	Special Plan for Electric Vehicle Charging and Battery-Swapping Infrastructure in Chengdu (2023–2025)
Public Sector Electrification	Newly added or renewed buses, taxis, and sanitation vehicles shall in principle be fully electrified	Chengdu Clean Energy Substitution and Energy Consumption Structure Adjustment Implementation Plan (2021–2025)
	100% electrification ratio for government vehicles (except for special purpose vehicles)
	100% electrification ratio for concrete mixer trucks and cold chain vehicles within the Third Ring Road
Technological Innovation Support	Additional deduction of R&D expenses (e.g., Teld Power enjoyed a deduction of 1.3 million CNY)	Detailed Implementation Rules for the Construction and Operation of Chengdu’s Electric Vehicle Charging and Battery-Swapping Infrastructure
	Encourage battery-swapping business models (maximum subsidy of 1 million CNY per station)	Notice on the Publication of the First Batch of Large Scale V2G Pilot Projects
	Pilot vehicle-to-grid (V2G) interactive technology (selected as one of the first national projects)
Used Car and Replacement Policy	Encourage “trade in” purchases (maximum subsidy of 15,000 CNY + partial purchase tax exemption)	Sichuan Province Passenger Vehicle Replacement Subsidy Scheme for Individual Consumers
	Promote the replacement of National IV and below fuel vehicles with new energy vehicles

).

(1)

For vehicle purchase incentives, the production and sales of new energy vehicles are eligible for rewards, with a maximum of 50 million CNY for a single model, and the exemption from vehicle purchase tax has been extended, according to the Implementation Opinions of Chengdu on Promoting the Development of the New Energy Vehicles Industry. In 2024, the total exemption amount reached 5.49 billion CNY, and some districts (counties), such as Qingyang District, also offer a maximum 5000 CNY “instant payment subsidy”.

(2)

For ease of use, new energy vehicles enjoy preferential policies such as exemption from license plate number restrictions, free parking within 2 h in public parking lots, a 50% discount on parking fees for overtime stays, relaxed traffic access rights for new energy freight vehicles, and the availability of a 30 day electronic access permit.

(3)

For industrial coordination policies, Chengdu is planned to achieve a 50% local supporting rate by 2025, establish large local automobile groups through mergers and acquisitions, and actively work towards jointly building the “hydrogen corridor” and “electricity corridor” between Chengdu and Chongqing.

(4)

For charging pile construction, it is to be promoted in accordance with the “Special Planning for Electric Vehicle Charging and Battery Swapping Infrastructure in Chengdu (2023–2025)”. By 2025, 160,000 charging piles and 3000 battery swapping stations are to be built. Public (dedicated) charging piles will be subsidized according to their star ratings, with a maximum subsidy of 200,000 CNY per station. New residential buildings are required to reserve 100% of the installation conditions for charging piles.

(5)

For electrification in the public sector, all newly added or renewed public transport vehicles, taxis, and sanitation vehicles shall be electrified in principle, in accordance with the Work Plan of Chengdu for Implementing Clean Energy Substitution and Accelerating Energy Consumption Structure Adjustment (2021–2025). The electrification rate of official vehicles (except special purpose vehicles) shall reach 100%, and the electrification rate of concrete transport vehicles and cold chain vehicles within the Third Ring Road shall also reach 100%.

(6)

For technological innovation support, enterprises qualify for R&D expense super deduction, single battery swap stations are eligible for a maximum subsidy of 1 million CNY, and Chengdu has been selected as one of the first batch of national vehicle-to-grid (V2G) technology pilot projects.

(7)

For second-hand vehicles and vehicle replacement, the city encourages “trade in” programs with subsidies of up to 15,000 CNY and exemption from partial vehicle purchase tax, and promotes the phaseout and replacement of fuel-powered vehicles complying with the National IV Vehicles Emission Standard (hereinafter referred to as “National IV”) or lower with new energy vehicles.

4. Methodology

4.1. Hypotheses

4.1.1. The Direct Environmental Benefits of NEVP

Drawing on existing studies [3,11], NEVP can produce direct environmental benefits through energy conservation and emission reduction. Specifically, the substitution of fuel-powered vehicles with new energy vehicles can significantly reduce greenhouse gas emissions during the usage phase and reduce the emissions of pollutants such as particulate matter. If the electricity comes from clean energy, the emissions across the entire life cycle will be further reduced. Therefore, the following research hypothesis is proposed:

H1: 

NEVP generates direct environmental benefits, which is the direct effect.

4.1.2. The Indirect Environmental Benefits of NEVP

Referring to existing studies [1,13,15,20,23,24], NEVP can generate indirect environmental benefits through multiple pathways.

(1)

The Pathway of Industrial Structure Upgrading

NEVP has promoted the increase in proportion of the tertiary industry and the green transformation of industry by stimulating the development of high-end manufacturing industries (such as power batteries and lightweight materials) and modern service industries (such as charging operation services and smart energy management). The construction of charging piles is characterized by being technology intensive, low polluting, and high-end manufacturing-oriented, which can promote the application of renewable energy, reduce dependence on traditional high-energy-consuming industries, and the increase in the proportion of supporting tertiary industries can significantly reduce energy consumption per unit of GDP, thereby indirectly reducing overall carbon emissions and realizing the greening of the economic structure. Therefore, the following research hypothesis is proposed:

H2: 

NEVP generates indirect environmental benefits through industrial structure upgrading, which is one of the indirect effects.

(2)

The Pathway of Green Consumption Transformation

The popularity of new energy vehicles has transformed residents’ transportation consumption patterns, shifting from fuel expenses to charging costs. If the power source is clean, carbon emissions across the entire life cycle will be significantly reduced. At the same time, the promotion of shared mobility and electric public transport reduces reliance on private cars and optimizes the urban traffic structure. Policy subsidies and environmental protection campaigns further enhance green consumption awareness, foster a positive cycle of market demand, industrial upgrading and low-carbon travel, and indirectly advance social low-carbon transformation. Therefore, the following research hypothesis is proposed:

H3: 

NEVP generates indirect environmental benefits through the pathway of green consumption transformation, which is one of the indirect effects.

(3)

The Pathway of Technological Innovation Driving Development

Policy incentives (such as subsidies, carbon quotas, etc.) will encourage enterprises to increase R&D investment, drive technological breakthroughs in areas such as battery technologies (e.g., solid-state batteries), fast charging technology, and vehicle-to-grid (V2G) technology, as well as improve energy utilization efficiency, thereby contributing to energy conservation and emission reduction. Technological progress significantly reduces the resource consumption of new energy vehicles across the entire life cycle and optimizes the grid’s green power utilization rate through intelligent energy management. At the same time, the collaborative innovation of digital and intelligent technologies (e.g., Internet of Vehicles (IoV)) and green technologies has continuously amplified the indirect emission reduction effect of new energy vehicles. Therefore, the following research hypothesis is proposed:

H4: 

NEVP generates indirect environmental benefits through the pathway of technological innovation driving development, which is one of the indirect effects.

In summary, the mediation pathways through which NEVP influences environmental benefits in Chengdu are illustrated in Figure 2.

4.2. Modeling

4.2.1. Baseline Regression Model

A multiple linear regression model is employed to conduct baseline regression and preliminarily examine the basic relationship between new energy vehicles sales and carbon emissions in Chengdu. The multiple linear regression model is specified as follows:

$$\ln Y_t ={\beta }_{0 }+{\beta }_1 \ln E{S}_{t }+\gamma CV{s}_t +{ϵ}_t,$$

(1)

where lnY_t represents the logarithmically transformed annual carbon dioxide emissions or air quality index (AQI) of Chengdu, lnES_t represents the logarithmically transformed annual sales of new energy vehicles in Chengdu, CVs stands for the control variables, and $ϵ$_t is the error term.

4.2.2. Lagged Variable Models

Considering the delayed impact of new energy vehicle popularization on air quality, to further capture the time lag of NEVP’s impact on environmental benefits in Chengdu, we assume that the air quality improvement resulting from new energy vehicles sales growth takes 1–2 years to materialize, where the lag periods are set to 1 and 2 years, respectively. The distributed lag model is as follows.

$$\ln Y_t =\alpha +{\displaystyle \sum _{k=0}^n{\theta }_k \ln E{S}_{t-k}} +\gamma CV{s}_t +{ϵ}_t,$$

(2)

where n = 1, 2; the remaining variables are the same as those in the previous text and will not be elaborated here.

4.2.3. Multiple Mediation Model

The commonly used methods for testing multiple mediation pathways include the causal steps approach, the Sobel test, the bootstrap method, and the MCMC method. Among them, the Sobel test requires the assumption of normality, which is often difficult to satisfy in practice. In contrast, the bootstrap method relaxes the assumptions on sample distribution [25,26,27]. To avoid confidence interval bias that may arise in the nonparametric percentile bootstrap procedure, this study employs the bias-corrected and bootstrap confidence interval method for mediation testing, with 5000 bootstrap resamples and a 95% confidence level. The causal steps approach, known for its wide applicability and high reliability, is first adopted to examine multiple mediation effects. Subsequently, to mitigate potential endogeneity concerns, the two-step approach is applied to further test the mediation pathways. Based on the research hypotheses, a transmission mechanism model incorporating multiple mediation pathways is constructed in this paper as follows:

$$\ln Y_{t }={\mathrm{a}}_1+{\mathrm{b}}_1\ln E{S}_t+{\mathrm{r}}_{1}CV{s}_t+e_{1t},$$

(3)

$$T{R}_t={\mathrm{a}}_2+{\mathrm{b}}_2\ln E{S}_t+{\mathrm{r}}_{2}CV{s}_t+e_{2t},$$

(4)

$$lnA{C}_t={\mathrm{a}}_3+{\mathrm{b}}_3\ln E{S}_t+{\mathrm{r}}_{3}CV{s}_t+e_{3t},$$

(5)

$$\ln Z{L}_t={\mathrm{a}}_4+{\mathrm{b}}_4\ln E{S}_t+{\mathrm{r}}_{4}CV{s}_t+e_{4t},$$

(6)

$$\ln Y_{t }={\mathrm{a}}_5+{\mathrm{b}}_5\ln E{S}_t+{\mathrm{d}}_{11}T{R}_t+{\mathrm{d}}_{12}\ln A{C}_t+{\mathrm{d}}_{13}\ln Z{L}_t+{\mathrm{r}}_{5}CV{s}_t+e_{5t},$$

(7)

where TR_t, lnAC_t, and lnZL_t respectively represent three mediation variables, namely the proportion of added value of the tertiary industry, per capita green consumption expenditure of residents (logarithm), and the number of authorized invention patent applications related to new energy vehicles (logarithm). The remaining variables are the same as those in the previous text and will not be elaborated here.

4.3. Variable Selection and Data Sources

4.3.1. Variable Selection

(1)

Dependent Variables

Following previous studies [3,11], this paper selected two core dependent variables to measure the environmental benefits of NEVP.

Firstly, this paper adopts the annual carbon dioxide emissions in Chengdu’s transportation sector (unit: 10,000 tons) as the carbon emission index (CE). Relevant data are obtained from the Greenhouse Gas Emission Inventory and the Low Carbon City Development Report released by the Chengdu Municipal Ecology and Environment Bureau. This indicator directly reflects the contribution of new energy vehicles to transportation carbon emission reduction.

Secondly, the Air Quality Index (AQI) of Chengdu is adopted as the air quality indicator, and the relevant data is sourced from the public annual reports of the air quality monitoring stations of the Chengdu Municipal Ecology and Environment Bureau.

(2)

Core Explanatory Variable

Existing studies mostly use sales volume to measure NEVP [1,2,23]. This paper also selects Chengdu’s new energy vehicles sales volume to examine the impact of the NEVP on environmental benefits in Chengdu. However, it is difficult to obtain new energy vehicles sales data at the sub-circle level in Chengdu. Therefore, the number of charging piles in each urban circle is selected to reflect the actual usage status of new energy vehicles, so as to further explore the sub-circle level heterogeneous impact of NEVP on environmental benefits.

(3)

Control Variables

To mitigate confounding effects, and based on prior research [1,2,23], this paper incorporates the following control variables: sales of conventional fuel-powered vehicles (FS), GDP growth rate (GS), and secondary industry value added ratio (SR). These variables are introduced to exclude the interference of non-core factors on the research results.

(4)

Mediation Variables

According to the research hypothesis, the following three variables are selected as mediation variables (ME): tertiary industry value added ratio (TR), residents’ per capita green consumption expenditure (AC), and the number of authorized invention patents related to new energy (ZL). These three mediations correspond to distinct indirect pathways: industrial structure upgrading pathway (ME1), green consumption transformation pathway (ME2), and technological innovation pathway (ME3).

4.3.2. Data and Variables

This paper uses monthly data from 2014 to 2024 on Chengdu’s new energy vehicles sales, the number of charging piles across different urban circles, air quality, and carbon emissions as the main sample.

(1)

New Energy Vehicles Sales

Data on Chengdu’s new energy vehicles sales are obtained from field surveys, the Energy Conservation and New Energy Vehicles Yearbook (2012–2024), and relevant research reports issued between 2014 and 2024 by Ping An Securities and Southwest Securities.

(2)

Charging Pile Data

Since Chengdu has not released monthly new energy vehicles charging pile stock data, and considering the long construction period of charging infrastructure such as charging piles and power supply grids, this paper uses the annual charging pile stock data (lnCR) of each urban circle in Chengdu to analyze regional heterogeneity across circles, and relevant data are obtained from the Baidu charging pile map database.

(3)

Air Quality and Carbon Emission Data

Relevant data on air quality and carbon emissions come from the CSMAR database, Chengdu Statistical Yearbook and relevant research reports from Ping An Securities and Southwest Securities Company from 2014 to 2024. The CO₂ data used in the baseline regression and mediation effect regression refer to CO₂ emission data, excluding CH₄, NO and other greenhouse gas emission data. The CO₂ equivalent (CO₂e) data in the robustness test is from the CEADs database.

(4)

Mediation and Control Variables

Data for the mediation variables and other control variables are sourced from the China Statistical Yearbook, the Sichuan Statistical Yearbook, the Chengdu Statistical Yearbook, and other publications. Given that indicators such as GDP growth rate, the proportion of value added of the secondary and tertiary industries, and the number of authorized invention patent applications are only released on a quarterly basis, the average distribution method is adopted for monthly disaggregation, and some missing data are supplemented via interpolation. To mitigate the impact of extreme values, continuous variables are winsorized at the 1st and 99th percentiles. Meanwhile, to avoid the influence of dimensional differences and other confounding factors on data stationarity, numerical variables are logarithmically transformed.

5. Results and Analysis

5.1. Sample Descriptive Statistics

The sample descriptive statistics results of all variables are presented in

Table 4. Descriptive statistics of variables.

Variable Name	Variable Type	Variable Definition	Sample Size	Mean	Standard Deviation (SD)	Minimum	Maximum
lnES	Explanatory Variable	Log of New Energy Vehicles Sales	132	7.19	2.251	2.303	10.772
lnCR	Explanatory Variable	Log of Charging Pile Stock by Circle Layer	18	10.123	1.199	6.59	11.714
AQI	Dependent Variable	Air Quality Index	132	85.515	25.476	43.9	222.39
lnCE	Dependent Variable	Log of Carbon Dioxide Emissions	132	3.651	0.358	2.553	4.297
lnFS	Control Variable	Log of Fuel-Powered Vehicles Sales	132	6.157	0.214	5.817	6.507
GS	Control Variable	GDP Growth Rate	132	0.068	0.019	0.028	0.089
SR	Control Variable	Proportion of Secondary Industry Added Value	132	0.279	0.03	0.244	0.349
TR	Mediation Variable	Proportion of Tertiary Industry Added Value	132	0.72	0.029	0.657	0.755
lnAC	Mediation Variable	Log of Per Capita Green Consumption Expenditure	132	10.652	0.137	10.39	10.81
lnZL	Mediation Variable	Log of Authorized Patents for New Energy Vehicles-Related Technological Inventions	132	10.177	0.389	9.575	10.814

. In the dependent variables, the mean of AQI is 85.515 (SD = 25.476), and the mean of lnCE is 3.651 (SD = 0.358). The core explanatory variables, lnES and lnCR, have mean values of 7.19 (SD = 2.251) and 10.123 (SD = 1.199). For the control variables, the mean of lnFS is 6.157 (SD = 0.214), the mean of GS is 0.068 (SD = 0.019), and the mean of SR is 0.279 (SD = 0.030). In the mediation variables, the mean of TR is 0.720 (SD = 0.029), the mean of lnAC is 10.652 (SD = 0.137), and the mean of lnZL is 10.177 (SD = 0.389). The sample contains no outliers and all variables are within normal ranges.

5.2. Analysis of Baseline Regression Results

The baseline regression results are presented in

Table 5. Baseline regression results.

Variables	lnCE	AQI
lnES	−0.276 ***	−10.814 *
	(0.0692)	(6.279)
lnFS	2.839 ***	19.637
	(0.866)	(86.779)
GS	2.815 *	−151.752
	(1.551)	(75.875)
SR	−7.769	−219.637
	(6.147)	(285.782)
Constant	18.516 **	764.792 *
	(7.051)	(648.070)
Mean VIF	3.090	5.961
White Test	16.73 **	32.44 **
ADF Test	−3.874 ***	−8.102 ***

Note: ***, ** and * respectively indicate significance at the 1%, 5%, and 10% levels. The standard deviation is in parentheses.

. In the regression results of the dependent variable lnCE, the regression coefficient of lnES is −0.276, which is significant at the 1% level, and Chengdu’s carbon emissions will reduce about 0.276% for every 1% increase in sales of new energy vehicles. The emission from the effect of NEVP is reduced by about 1,087,500 tons per year, accounting for about 4% of the city’s total carbon emissions.

In the regression results of the dependent variable AQI, the regression coefficient of lnES is −10.814, which is significant at the 10% level. This indicates a significant negative correlation between lnES and the two dependent variables. NEVP in Chengdu brings about significant environmental benefits. Therefore, the research hypothesis H1 is valid. In terms of model diagnosis, the mean variance inflation factor of the lnCE model is 3.090, and the AQI model is 5.961, both below the critical value of 10, indicating that the model does not have serious multicollinearity problems. The results of the heteroscedasticity test show that the White statistic of the lnCE model is 16.73 (significant at the 10% level) and the White statistic of the AQI model is 32.44 (significant at the 5% level), indicating a certain degree of heteroscedasticity. Therefore, robust standard errors are used to correct the estimation results, and the corrected results are reported in Table 5. In terms of stationarity testing, the ADF unit root test results show that the ADF statistics of the key variables are significant at the 1% level, indicating that the sequence is stationary and there is no unit root problem. The baseline regression results are reliable.

5.3. Distributed Effects Model Analysis

A distributed lag model (Equation (2)) is constructed to analyze the distributed effects of NEVP on carbon emissions and air quality in Chengdu. The corresponding regression results are presented in

Table 6. Distributed lag model results.

Variables	Current Period (β0)	1-Period Lag (β1)	2-Period Lag (β2)	3-Period Lag (β3)	Long-Term Effects (∑β)
lnCE	−0.205 ***	−0.0393 *	−0.0686 **	0.0177	−0.1189 **
	(0.029)	(0.021)	(0.019)	(0.164)	(0.017)
AQI	−0.941	2.969	0.225	−5.586 **	−3.332 **
	(2.496)	(2.496)	(2.694)	(2.672)	(2.466)
Controls	Yes Yes
Time trend

Note: ***, ** and * respectively indicate significance at the 1%, 5%, and 10% levels. The standard deviation is in parentheses.

.

In the lnCE model, the β₀ (−0.205, 1% significant), β₁ (−0.0393, 10% significant), β₂ (−0.0686, 5% significant), and ∑β (−0.1189, 5% significant) are all significant, indicating that the impact of NEVP on carbon emissions in the current, 1-period lag, 2-period lag and long-term are significant. In the AQI model, only the β₃ (−5.586, 5% significant) and the ∑β (−3.332, 5% significant) are significant, while the other coefficients are not significant. The carbon emission reduction effect of NEVP has a significant negative impact in the current period, 1-period lag and 2-period lag, after which the effect diminishes. In the long run, the cumulative emission reduction effect is significant, indicating that the NEVP has sustainable carbon reduction benefits. The air quality improvement effect of the NEVP has a notable time lag. The effect is not statistically significant from the current period, 1-period lag and 2-period lag. However, the effect becomes significantly negative at the 3-period lag and in the long-term effect. The long-term environmental benefits of NEVP are significantly greater than its short-term benefits.

5.4. Multiple Mediation Model Analysis

The two-step method is employed to test the mediation pathways to mitigate potential endogeneity issues. The regression results of the mediation models are presented in

Table 7. Mediation model regression results.

Variables			Industrial Structure Upgrading Pathway		Green Consumption Transformation Pathway		Technological Innovation Pathway
			lnCE	AQI	lnCE	AQI	lnCE	AQI
Explanation Variable		lnES	−0.071 ***	−2.593	−0.078 ***	−1.056	−0.017 **	−4.948 **
			(0.020)	(1.854)	(0.021)	(1.940)	(0.009)	(2.044)
Mediation Variable		TR	−3.597 **	−164.321	- -	- -	- -	- -
			(1.571)	(145.176)
		lnAC	- -	- -	−0.598 *	−62.368 *	- -	- -
					(0.351)	(31.835)
		lnZL	- -	- -	- -	- -	−0.603 ***	3.418
							(0.119)	(11.827)
Bootstrap test			Industrial Structure Upgrading Pathway		Green Consumption Transformation Pathway		Technological Innovation Pathway
Mediation Pathways	Upper limit		88.990	23.092	87.795	24.789	118.423	21.400
	Lower limit		41.412	1.635	38.082	2.954	51.806	1.913
Mediation Effect			Partial	Not significant	Partial	Not significant	Partial	Not significant
Controls			Yes

Note: ***, ** and * respectively indicate significance at the 1%, 5%, and 10% levels. The standard deviation is in parentheses.

. The results of the three mediation pathways indicate that the environmental benefits of NEVP are primarily achieved through industrial structure upgrading, green consumption transformation, and technological innovation mechanisms.

The results of the multiple mediation pathway model and Bootstrap test show that there are significant differences in the indirect effects of the three pathways. The results of the mediation effect test, with carbon emissions (lnCE) as the dependent variable, show that all three pathways play a significant partial mediation role. In the pathway of industrial structure upgrading, the coefficient of lnES is −0.071 and significant at the 1% level. In the multiple mediation pathway model, the total effect of NEVP includes direct effect and indirect effect. NEVP in Chengdu has a significant direct emission reduction effect, accounting for about 39.80% of the total effect, and the annual average carbon emissions are reduced by about 432,800 tons, accounting for about 1.57% of the total carbon emissions in Chengdu.

After incorporating the mediation variable TR, the coefficient decreases compared to the baseline regression. The coefficient of TR is −3.597 and significant at the 5% level. The bootstrap test shows that this pathway has a significant indirect effect on lnCE, indicating that NEVP can reduce carbon emissions by increasing the proportion of the tertiary industry. Based on the TR path, the annual average carbon emissions are reduced by 25.71% of the total effect, and the annual average carbon emissions are reduced by about 279,500 tons, accounting for about 1.02% of the total carbon emissions in Chengdu.

Following the pathway of green consumption transformation, the coefficient of lnES is −0.078 and significant at the 1% level. After incorporating lnAC, the coefficient decreases compared to the baseline regression results. The coefficient of lnAC is −0.598 and significant at the 10% level. The bootstrap test confirms the existence of a significant indirect effect, indicating that NEVP can generate carbon emission reduction benefits by promoting residents’ green consumption expenditures. Based on the lnAC path, the annual average carbon emissions are reduced by 28.29% of the total effect, and the annual average carbon emissions are reduced by about 307,800 tons, accounting for about 1.12% of the total carbon emissions in Chengdu.

Regarding the technological innovation pathway, the coefficient of lnES is −0.017 and significant at the 5% level, while the coefficient of lnZL is −0.603 and significant at the 1% level. The bootstrap test results confirm the existence of a significant indirect effect, indicating that NEVP can reduce carbon emissions by increasing the number of relevant technology patent authorizations. Based on the lnZL path, the annual average carbon emissions are reduced by 6.20% of the total effect, and the annual average carbon emissions are reduced by about 67,400 tons, accounting for about 0.25% of the total carbon emissions in Chengdu.

Based on the direct effect and indirect effect, the total carbon emission from the effect of NEVP is reduced by about 3.95% of the total carbon emissions, and 40% of carbon emission within the transportation sector in Chengdu.

All three mediation pathways can effectively explain the carbon reduction effect of NEVP, therefore the research hypotheses H2, H3, and H4 regarding carbon emissions are confirmed.

The results of the mediation effect test with AQI as the dependent variable showed that the mediation effects of the three pathways are not significant. In the pathway of industrial structure upgrading, the coefficient of lnES is −2.593 but not significant, and the coefficient of TR is −164.321 but also not significant. In the pathway of green consumption transformation, although the coefficient of lnAC is −62.368 and significant at the 10% level, the coefficient of lnES is only −1.056 and not significant. In the pathway driven by technological innovation, the coefficient of lnES is −4.948 and significant at the 5% level, while the coefficient of lnZL is 3.418 and not significant. Although the three mediation pathways mentioned above pass the bootstrap test, there are cases where their main effects or mediation variables are not significant. Therefore, none of the existing three mediation pathways can effectively explain the impact of NEVP on air quality. In practice, the AQI is primarily determined by the emission of pollutants such as PM2.5, PM10, SO₂, CO, NO₂ and O₃ within the region. According to the Chengdu Action Plan for Optimizing the Energy Structure and Promoting Green and Low Carbon Urban Development, the electricity used by Chengdu’s new energy vehicles is mainly sourced from hydropower and other clean energy forms. Among the aforementioned pollutants, only SO₂, CO, and NO₂ are directly related to the transportation sector, while most others originate from non-transportation sources. Consequently, NEVP in Chengdu has a limited impact on overall air quality improvement.

5.5. Endogeneity Discussion and Robustness Test

5.5.1. Endogeneity Discussion

A higher level of NEVP can generate environmental benefits of energy conservation and emission reduction, thereby facilitating local green development. Meanwhile, these environmental benefits enable these regions to offer policy, capital, technology, and market support for NEVP in terms of industrial growth in science and technology, industry, policy, and market environments. Therefore, there may be a reverse causal endogenous relationship between new energy vehicle market promotion and energy conservation and emission reduction. This paper incorporates data from an additional five cities, namely Beijing, Shanghai, Guangzhou, Shenzhen and Chongqing, expands the sample data to panel data, and uses the instrumental variable method for the endogenous test. This paper introduces the average sales volume of other cities, the winter season variable and the dual-credit policy impact in 2018 as exogenous IVs. The instrumental variable fixed effect model (IV-FE) is used for two-stage least squares regression. The validity of the IVs meet the two key restrictions of relevance and exclusion, as follows.

(1)

The sales of new energy vehicles in other cities influence the sales in the local city through competition and learning mechanisms, which satisfies the relevance restriction. The sales in other cities do not directly affect the local air quality, which satisfies the exclusion restriction.

(2)

In winter, new energy vehicle manufacturers offer substantial discounts to achieve record annual sales, thereby influencing new energy vehicle sales, which satisfies the relevance restriction. Winter is an exogenous natural phenomenon. Its impact on air quality is controlled by month fixed effects, which ensures that the exclusion restriction is satisfied.

(3)

The dual-credit policy in 2018 directly promotes enterprises to increase new energy vehicle sales, which satisfies the relevance restriction. The policy implementation is determined by the central government and has no direct relationship with the air quality of individual cities, only indirectly affecting air quality through its influence on sales, which satisfies the exclusion restriction.

To further confirm the validity of the IVs, we conduct an identification test and a weak instrumental variable test. The results show that the IVs selected in this paper have strong explanatory power and there is not a weak instrumental variable problem. At the same time, we supplement the causal identification strategy diagram, which visually presents the causal path between the IVs, endogenous explanatory variables, and the dependent variable. The strategy of endogeneity test is shown in Figure 3, and the results of the endogeneity test are presented in

Table 8. Endogeneity test results.

Variables	(1) AQI	(2) lnCE	(3) AQI	(4) lnCE	(5) AQI	(6) lnCE
	FE	FE	IV-FE	IV-FE	IV-FE(Full)	IV-FE(Full)
lnES	−1.813 ***	−0.037 ***	−2.732 ***	−0.060 ***	−2.825 ***	−0.063 ***
	(0.312)	(0.008)	(0.419)	(0.006)	(0.414)	(0.006)
L1.AQI	0.455 ***	-	0.422 ***	-	0.419 ***	-
	(0.037)	-	(0.033)	-	(0.032)	-
L1.lnCE	-	0.662 ***	-	0.568 ***	-	0.556 ***
	-	(0.062)	-	(0.032)	-	(0.032)
Time Fixed Effect	Yes	Yes	Yes	Yes	Yes	Yes
Individual Fixed Effect	Yes	Yes	Yes	Yes	Yes	Yes
Controls	Yes	Yes	Yes	Yes	Yes	Yes
R2 (within)	0.360	0.795	0.355	0.788	0.354	0.786
Hausman Test	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001	p < 0.001
Cragg–Donald F	-	-	110.86 ***	11.03 ***
LM	-	-	273.423 ***	1191.225 ***	297.438 ***	665.606 ***
IV	-	-	Average Sales Policy Impact	Average Sales Policy Impact	Full IV	Full IV

Note: *** respectively indicate significance at the 1% levels. The standard deviation is in parentheses.

.

The results of the Hausman test show that the p-values for all models are less than 0.001, leading to a strong rejection of the null hypothesis that the coefficients from the FE and IV-FE models are not systematically different at the 1% significance level. This suggests that endogeneity is a concern in the models and the use of the instrumental variable method is necessary and reasonable. In terms of instrumental variable validity testing, the LM statistic of the underidentification test is significant at the 1% level (the AQI model is 273.423, the lnCE model is 1191.225, the Full model is 297.438 and 665.606), rejecting the null hypothesis of “underidentification instrumental variables”. The Cragg–Donald F statistic for weak instrumental variable testing is 110.86 in the AQI model and 11.03 in the lnCE model, both of which are above the critical value of 10. The results suggest that the weak instruments problem is not a concern and the selected instrumental variables have strong explanatory power. The two-stage regression results show that the coefficient direction and significance of the core explanatory variable lnES remained consistent with the results of the baseline model. The direction of the coefficients of the main explanatory variables in the above regression results is consistent with Table 5, Table 6 and Table 7, and the absolute values of the coefficients increase, indicating that the FE model underestimates the real environmental benefits of promoting new energy vehicles due to endogeneity issues, while the instrumental variable method effectively controls endogeneity issues, making the estimation results more reliable.

5.5.2. Robustness Test

In order to ensure the reliability of the baseline regression results, this paper uses the variable replacement method, sample period adjustment method, and panel data model comparison method for robustness testing. The specific strategies are as follows:

(1)

We replace the carbon emissions (lnCE) with the green development level (GD), the industrial structure upgrading with the overall industrial structure upgrading, the green consumption level with the digital economy level, the new energy vehicle invention patent authorization with the new energy transportation infrastructure development level, and the carbon emissions and air quality index with the specific pollutant indicators (PM_2.5, PM₁₀, SO₂, CO₂e) to test whether the regression results can support the conclusions drawn from baseline regression.

(2)

We test whether the core conclusions are sensitive to sample selection by changing the sample cycle, randomly excluding months (excluding 36 months of sample data in 2016, 2019 and 2020), and excluding the special period of the COVID-19 epidemic.

(3)

We compare the estimation results of fixed effects (FEs) and random effects (REs) models based on panel data from 2014 to 2024 (including six cities of Beijing, Shanghai, Guangzhou, Shenzhen, Chongqing, and Chengdu) to test whether the results are affected by data frequency and model specification.

The results of the replacement variable method test are shown in

Table 9. Results of using alternative variable measures.

Variables	Replacement of the Explanatory and Mediation Variables
	GD	GD	GD
lnES	0.007 ***	0.009 ***	0.017 ***
	(0.002)	(0.002)	(0.001)
TR_new	0.150 ***	-	-
	(0.017)	-	-
lnAC_new	-	0.284 ***	-
	-	(0.049)	-
lnZL_new	-	-	0.028 ***
	-	-	(0.007)
Controls	Yes	Yes	Yes

Note: *** respectively indicate significance at the 1% levels. The standard deviation is in parentheses.

and

Table 10. Results of using panel data and changing variables.

Variables	Replacement of the Explanatory Variable
	lnPM2.5	lnPM10	lnSO2	lnCO2e
lnES	−0.007 ***	−0.009 ***	−0.017 ***	−0.244 ***
	(0.002)	(0.002)	(0.001)	(0.001)
Controls	Yes	Yes	Yes	Yes

Note: *** respectively indicate significance at the 1% levels. The standard deviation is in parentheses.

. The coefficient direction and significance of the relevant variables have no significant differences from Table 5, Table 6 and Table 7, indicating that the empirical results of this study have high robustness.

The test results of the sample period adjustment method are shown in

Table 11. Results of changing the sample period and randomly excluding months.

Methods	Change Sample Period (lnCE)	Randomly Eliminate Months (lnCE)	Eliminate the COVID-19 Period (lnCE)
lnES	−0.031 ***	−0.026 ***	−0.111 ***
	(0.002)	(0.001)	(0.016)
Controls	Yes	Yes	Yes

Note: *** respectively indicate significance at the 1% levels. The standard deviation is in parentheses.

. After changing the sample period, the coefficient of lnES is −0.031 and significant at the 1% level. The coefficient after randomly excluding months is −0.026 and significant at the 1% level. The coefficient after excluding the special period of the COVID-19 epidemic is −0.111 and significant at the 1% level. The above results are consistent with the baseline regression, indicating that the empirical results of this paper do not depend on the selection of specific sample periods; the emission reduction has a more obvious effect after excluding the impact of the COVID-19 epidemic.

The comparison results of panel data models are shown in

Table 12. Results of FE and RE models based on annual data.

Variables	AQI		lnCE
	FE	RE	FE	RE
lnES	−5.627 ***	−5.610 ***	−0.133 ***	−0.132 ***
	(0.696)	(0.690)	(0.014)	(0.014)
Constant	134.123 ***	133.937 ***	2.666 ***	2.660 ***
	(7.494)	(9.834)	(0.148)	(0.163)
Controls	Yes	Yes	Yes	Yes
Number of cities	6	6	6	6
R2 (within)	0.526	0.526	0.614	0.614
F-test	F = 65.35 ***	χ2 = 66.12 ***	F = 93.81 ***	χ2 = 95.41 ***
Hausman test	0.151 (p = 0.701)		0.160 (p = 0.688)

Note: *** respectively indicate significance at the 1% levels. The standard deviation is in parentheses.

. When annual panel data are employed, the explanatory variable regression coefficients of FE and RE in the AQI model are −5.627 and −5.610 and both are significant at the 1% level. In the lnCE model, the explanatory variable regression coefficients of FE and RE are −0.133 and −0.132 and both are significant at the 1% level. The coefficient estimates from the two models are highly consistent, and the p-values of the Hausman test are 0.701 and 0.688. The results do not reject the null hypothesis, indicating that the results are not affected by the model specification form. In summary, the negative impact of NEVP on carbon emissions and air quality passes all robustness checks, validating the results of baseline regression.

The robustness test in this article has some limitations. Based on the availability of life cycle assessment data, the empirical research in this paper mainly focuses on the environmental benefits of new energy vehicles in the use phase without the life cycle assessment data. Specifically, the emission factors in the production process of new energy vehicles, the emission factors in the power production process, and the environmental factors in the scrapping and recycling stage of new energy vehicles are not included in the scope of this paper. This may lead to the underestimation of the carbon emission reduction effect of NEVP in the empirical study. Therefore, the empirical result of this paper is the estimation of the environmental benefits generated by NEVP, rather than a life cycle assessment of NEVP’s benefits.

The indirect carbon emission level of new energy vehicles in Chengdu is significantly lower than that of other cities dominated by thermal power, because Chengdu mainly uses clean electricity such as hydropower. Therefore, the lack of the life cycle assessment data has a relatively small impact on the empirical results of this paper. Future research can combine the heterogeneity data of urban power structures with the life cycle emission coefficients to calculate the life cycle emissions of new energy vehicles, and further analyze the environmental benefits of NEVP.

5.6. Heterogeneity Analysis of Sub-Circles

Chengdu is administratively divided into three concentric zones: the first circle (central urban), the second circle (suburbs), and the third circle (outer suburbs), as illustrated in

Table 13. Chengdu’s urban circles.

Urban Circles	Districts/Counties
The First Circle (Central Urban)	Jinjiang District, Qingyang District, Jinniu District, Wuhou District, Chenghua District, High-Tech Zone, Tianfu New Area (Directly Managed by Chengdu)
The Second Circle (Suburbs)	Longquanyi District, Qingbaijiang District, Xindu District, Wenjiang District, Shuangliu District, Pidu District
The Third Circle (Outer suburbs)	Jianyang City, Dujiangyan City, Pengzhou City, Qionglai City, Chongzhou City, Jintang County, Dayi County, Pujiang County, Xinjin District

.

The results of the circle-specific regression are presented in

Table 14. Regression results for Chengdu’s urban circles.

Variables	The First Circle		The Second Circle		The Third Circle
	lnCEaverage	AQIaverage	lnCEaverage	AQIaverage	lnCEaverage	AQIaverage
lnCR	−0.312 ***	−12.217 ***	−0.258 ***	−10.965 **	−0.021	1.266
	(−0.031)	(3.361)	(−0.030)	(2.733)	(0.020)	(1.680)
Controls	Yes		Yes		Yes

Note: *** and ** respectively indicate significance at the 1% and 5% levels. The standard deviation is in parentheses.

. The impact of NEVP on environmental benefits in Chengdu exhibits significant spatial heterogeneity across urban circles. The increase in charging pile stock (lnCR) significantly reduces carbon emissions (lnCE) and improves air quality (AQI) in both the first circle (central urban) and the second circle (suburbs), with the strongest environmental benefits observed in the first urban circle, followed by the second. However, the environmental benefits in the third circle (outer suburbs) are statistically insignificant, which may be attributed to the insufficient charging infrastructure or the low new energy vehicle penetration rate in Chengdu’s outer suburban regions. This result indicates that NEVP in Chengdu has a significant “decreasing circle” impact on environmental benefits, with significant effects in the central urban area and suburbs, but no significant environmental benefits in the outer suburbs.

Since the annual charging pile stock is used as the explanatory variable, and the annual mean of lnCE and the annual mean of AQI are used as the dependent variables, the frequency mismatch between the annual and monthly variables may cause the following three biases:

(1)

The time aggregation bias. As the annual stock data of charging piles will smooth the monthly stock data, if the incremental distribution of charging piles in each month is uneven, using the annual average value may lead to biased regression results.

(2)

The measurement error bias. The annual stock data cannot accurately observe the actual stock data of charging piles in each month. The actual number of charging piles may show significant seasonal or monthly changes due to weather, travel behavior and other factors, and these changes cannot be reflected in the annual stock data.

(3)

Reverse causality bias.

From the perspective of the bias direction, since the construction of charging piles usually follows a positive growth trend, although the time aggregation bias leads to the annual charging pile stock data smoothing the monthly stock data, it will not systematically reverse the direction of the variable relationship, so the estimated effect direction is still reliable. From the perspective of the bias degree, the construction of charging piles in Chengdu is a planned and gradual accumulation process, and its monthly growth rate is relatively stable. Therefore, we believe that the bias degree is moderate and will not fundamentally change the results. In the robustness test, we re-estimate the model after aggregating the monthly data of the result variables into the annual data. The two results are consistent, indicating that the frequency mismatch will not affect the empirical results of this paper.

6. Conclusions and Suggestions

Based on the data of NEVP, carbon emissions, and air quality indicators in Chengdu (both citywide and across urban circles), this paper empirically examines the impact pathways and spatial heterogeneity of NEVP on environmental benefits. The results show that NEVP exerts a significant impact on regional environmental benefits. The environmental benefits of NEVP exhibit distinct characteristics of time lag and long-term persistence. Specifically, in terms of direct effects, NEVP significantly reduces carbon emissions, but its impact on air quality improvement has not reached a statistically significant level. Further mechanism analysis indicates that NEVP generates environmental benefits mainly through indirect pathways such as industrial structure upgrading, green consumption transformation, and technological innovation driving development. These transmission mechanisms jointly promote a significant reduction in carbon emissions, but their overall effect on air quality improvement remains relatively limited. From a dynamic perspective, the environmental benefits of NEVP show significant time heterogeneity. The short-term lag effect is not significant, whereas the long-term effect is substantially strong. This suggests that the realization of the environmental benefits of NEVP depends on the continuous promotion process of technology diffusion and industrial structure adjustment. In addition, the impact of NEVP on environmental benefits displays distinct spatial heterogeneity across different urban circles. NEVP in the first circle (central urban) and the second circle (suburbs) significantly reduces carbon emissions and improves air quality, while the corresponding effects in the third circle (outer suburbs) are not statistically significant. In general, NEVP in Chengdu has a significant “decreasing circle” impact on environmental benefits.

This paper introduces the data of Beijing, Shanghai, Guangzhou, Shenzhen and Chongqing to build a multi-city panel data regression model and tests the robustness and generalizability of the baseline regression results. The robustness test results show that the regression coefficient direction and significance level of the panel data regression model are consistent with the baseline regression results. This shows that the empirical results are not only highly robust but also have a certain degree of universality, and are not accidental phenomena in a specific urban context. Therefore, this paper first makes the baseline regression through the data of Chengdu, and then tests the robustness and generalizability of the baseline regression results through the multi-city panel data. The two constitute the primary findings of the study in different dimensions.

This paper proposes the following policy implications:

(1)

New energy vehicle trade in subsidy schemes should be further optimized, and policy support for the three electric systems enterprises (power batteries, motors, and electronic control systems) should be enhanced. The environmental benefits of energy conservation and emission reduction can be sustained and strengthened by synergizing the pathways of industrial structure upgrading, green consumption transformation, and technological innovation driving development.

(2)

Enterprises related to new energy vehicles should be encouraged to intensify investment in technological R&D, and a multi-year environmental benefits evaluation system should be established, in which long-term carbon emission reductions are incorporated as a core assessment indicator, thereby guiding enterprises toward sustained green innovation.

(3)

Differentiated policies for charging infrastructure construction should be advanced in light of local conditions. In the first circle (central urban), the spatial layout and density of charging networks should be optimized. In the second circle (suburbs) and the third circle (outer suburbs), the construction of charging infrastructure should be accelerated to increase new energy vehicle penetration. Meanwhile, differentiated traffic rights policies should be promoted, such as stricter driving and parking restrictions for fuel-powered vehicles in the first circle, and toll exemptions on ring expressways for new energy freight vehicles in the third circle, so as to further amplify the environmental benefits of NEVP.

Since this paper only studies the environmental benefits of NEVP and does not consider the social equity dimension, future research should further study the social equity dimension of NEVP based on household financial micro data, GIS data and industrial employment structure data.

(1)

The NEV purchase subsidies and price preferential policies are more conducive to low- and middle-income groups, to a certain extent.

(2)

The spatial layout of charging infrastructure may affect the use and convenience of different income groups.

(3)

NEVP may have a structural employment impact on workers utilizing fuel-powered vehicles.