Can a Driving Restriction Policy Improve Air Quality? Empirical Evidence from Chengdu

Abstract

Automotive exhaust emissions contribute significantly to air pollution in developing countries. However, the effectiveness of driving restriction policies (DRPs) is unclear, and most research on China emphasizes Beijing. This study used Chengdu, a typical large city in China, to examine the impact of a DRP on air quality. To alleviate potential endogeneity threats, we employed a regression discontinuity design to verify the policy’s effect. The results show that the DRP significantly reduced air pollution levels, effectively improving air quality in restricted areas. The heterogeneity analysis found that (1) the DRP effectively reduced pollution in newly added and original areas, while the air quality in adjacent areas deteriorated; and (2) the DRP significantly improved air quality during peak travel periods but had no significant impact in other periods. Our results indicate that the DRP is an effective tool for urban environmental governance but presents potential negative aspects. Therefore, restricted areas and periods should be carefully considered when designing similar policies. This study provides significant insights into the governance of automotive exhaust emissions pollution for large cities in developing countries.

1. Introduction

A healthy ecological environment is essential for human survival. Accordingly, air pollution poses a significant threat to human health and well-being. Long-term exposure to polluted air has been linked to various respiratory [1] and cardiovascular diseases and can reduce life expectancy [2]. Air pollution not only affects physical health but also influences mental health [3] and overall life satisfaction, with a more pronounced impact on vulnerable populations [4]. Furthermore, air pollution can disrupt labor supply [5,6] and economic productivity [7], as improved air quality has been shown to positively affect labor productivity [8,9] and housing values [10]. Consequently, policymakers globally are increasingly adopting precautionary measures to control air pollution and safeguard public health.

In China, vehicles are a significant source of air pollution [11]. They contributed approximately 55.59% of nitrogen oxides (NOx) and 34.5% of volatile organic compounds (VOCs) in 2020 [12]. China is currently addressing this issue by promoting energy conservation, emission reduction, and the development of renewable energy. However, the country’s end-use energy consumption structure remains heavily dependent on fossil fuels, particularly coal and petroleum. In 2020, coal accounted for 34.77% and petroleum for 20.15% of China’s total end-use energy consumption [13], reflecting an ongoing reliance on unclean energy sources that contribute to increasing environmental pollution. In this context, vehicle emissions are of particular concern, as motor vehicles primarily run on petroleum-based fuels, such as gasoline and diesel, which directly impact urban air quality. Over the past 40 years, traffic congestion and exhaust pollution from increased automotive use have seriously affected the public’s quality of life and health [1,14]. To address this issue, China has widely implemented vehicle license plate restrictions in Beijing and other major prefecture-level cities since the 2008 Beijing Olympics [15], and now has the most extensive implementation of such restrictions of all countries globally. For example, Chengdu implemented a motor vehicle license plate restriction policy in 2012 and expanded it in 2018.

In recent years, the effectiveness of driving restriction policies (DRPs) in improving air pollution and alleviating traffic congestion has become a focal point of scholarly debate [16,17]. Unfortunately, the research is inconclusive. For example, in a study on the impact of Beijing’s DRP on air pollution, some scholars concluded that the DRP significantly improved air quality in Beijing during the 2008 Olympic Games [18,19]. However, other scholars suggested that this DRP had no significant effect [20]. Moreover, there is a lack of consensus on whether DRPs in other cities can improve air quality. For example, Sun and Xu found that driving restrictions in provincial capital cities in China can change PM_2.5 [21], whereas Ye found that the DRP in Lanzhou exacerbated air pollution levels [22]. DRPs are not unique to China. Mexico also implemented a DRP to improve air quality; however, Davis found this DRP ineffective in improving air quality [16]. However, Carrillo et al. found that a DRP in Ecuador significantly reduced carbon monoxide (CO) emissions [23]. Additionally, the selection of different air quality indicators may be the reason for the inconsistent conclusions in the literature [24]. Therefore, this study adopts multiple indicators—including the Air Quality Index (AQI), PM_2.5, PM₁₀, SO₂, NO₂, and CO—instead of a single indicator for policy analysis to avoid the uncertainty of conclusions generated by the selection of single indicators.

Theoretically, DRPs have multidimensional effects on air pollution, potentially leading to varied outcomes [22]. A DRP study of Chengdu can obtain more representative conclusions than that of Beijing, which is why we selected the former. As a megacity, Beijing has a large population, large economic scale, more complete infrastructure, and more mature political environment. Therefore, the effectiveness of Beijing’s DRP might not generalize to other large cities. As an important city in western China, Chengdu offers a more representative case of large cities than Beijing regarding its socio-economic characteristics. The pilot program for Chengdu’s DRP provides a representative case, and its results could serve as a policy reference and guidance for similar cities in China and even the world. Therefore, this study examines the DRP implemented in Chengdu in 2018, utilizing hourly air observation data and a regression discontinuity design (RDD) to illuminate the impact of the DRP on the AQI and five other air pollutants (PM_2.5, PM₁₀, SO₂, NO₂, and CO). This study further addresses the spatial and temporal heterogeneity of policy effects. The results show that the DRP mainly improved air quality within restricted areas and during restricted times. However, the effect of the DRP was not significant during non-restricted hours. Unfortunately, the DRP even worsened the air quality in areas adjacent to the restricted zones. These findings enable us to evaluate the policy’s effectiveness and assess the scope of its impact.

This study extends the literature in several key areas. First, it is among the first to examine the DRP in a large inland Chinese city outside of Beijing, providing a new perspective on its effectiveness in cities with different socio-economic and infrastructural contexts. Second, it uses fine-grained, hourly air quality data and multidimensional evaluation indicators to improve the precision of the analysis. With a set of comprehensive indicators—AQI, PM_2.5, PM₁₀, SO₂, NO₂, and CO—the study allows cross-checking across pollutants, enhancing the reliability of the conclusions and reducing potential biases from relying on a single measure. Third, this study deepens understanding of the complex effects of DRPs on controlling vehicle pollution. It analyzes the overall effects of the policy and delves into its heterogeneous impacts across different zones and times, discussing potential negative externalities. This analysis provides empirical evidence for policymakers to formulate effective vehicle management policies.

The remainder of this paper proceeds as follows: Section 2 presents the policy background, Section 3 presents the data and identification strategies, Section 4 presents the regression results, Section 5 presents the heterogeneity analysis, and Section 6 presents the discussion and conclusions.

2. Policy Background

2.1. DRP Policy on Chengdu

Chengdu implemented a motor vehicle license plate restriction policy in April 2012. In subsequent years, the policy implementation area was adjusted twice. Following these adjustments, the region of restricted areas gradually expanded, whereas rules, such as restricted time and vehicle license plates, remained unchanged. Regarding restricted vehicles, Chengdu’s DRP applies to all “Chuan A” and non-local license plate vehicles. Under the “tail number” rule, restricted vehicles are divided into five groups based on the last Arabic digit of their license plates (including temporary license plates). Each group is prohibited from traveling through the restricted area during restricted hours on each working day. The specific restriction rules are as follows: vehicles with tail numbers “1” and “6” are restricted on Mondays. Vehicles with license plates ending in “2” and “7” are restricted on Tuesdays. Vehicles with license plates ending in “3” and “8” are restricted on Wednesdays. Vehicles with license plates ending in “4” and “9” are restricted on Thursdays. Vehicles with license plates ending in “5” and “0” are restricted on Fridays.

The restricted areas of Chengdu’s DRP have undergone a process of adjustment from line to surface and significant expansion. From 26 April to 30 September 2012, the first round of traffic restrictions in Chengdu was implemented. The initial restricted area included the entire Second Ring Road and seven main radial roads entering and exiting the city, as shown in the red section on the left picture of Figure 1. Since the DRP pilot, the Chengdu traffic management bureau has adjusted the restricted areas twice, on 8 October 2012 and 22 January 2018. The restricted area was adjusted for the first time on 1 October 2012 and expanded to include all roads between the Second and Third Ring Roads (including the Second and Third Ring Roads). The restricted area greatly expanded from linear to planar coverage, as shown in the red area in the middle of Figure 1. Travel restrictions were uniformly determined to be implemented between 7:30 and 20:00 from Monday to Friday after the first adjustment. On 22 January 2018, the restricted areas were subjected to a second adjustment. The restricted area was significantly expanded based on the second restricted area and further adjusted to include all roads within the Ring Expressway (excluding), as shown in the red area in Figure 1c. The most recent DRP adjustment on the map expanded the restricted area from the original “small circular ring” to a large “circular cake”.

2.2. Policy Motivation and Economic Environment

Chengdu was one of the most polluted cities in China before the first implementation of the DRP. From 2009 to 2010, the annual average PM_2.5 concentration in Chengdu reached 165 µg/m³ [25], which is more than five times the upper limit value recommended by the World Health Organization (35 µg/m³) and is generally higher than other Chinese cities. Around 2010, the air quality in Chengdu failed to meet China’s second-level standards, with the main sources of pollution being dust, automotive vehicle exhaust, and mixed coal smoke pollution. The primary pollutants were inhalable particulates, followed by NO₂, which has seen a yearly increase in the pollution load. Automotive vehicle exhaust emissions are the main sources of particulate and NO₂ pollution in the atmosphere.

To further illustrate the sources of Chengdu’s air pollution,

Table 1. The emissions of atmospheric pollutants in Chengdu, 2017.

		Total	Industrial Sources	Residential Sources	Mobile Sources
					Non-Road Mobile	Motor Vehicle
NOx	(104 tons)	13.54	2.35	0.52	4.38	6.28
	(100%)	100.00%	17.36%	3.84%	32.35%	46.38%
VOCs	(104 tons)	12.76	5.79	3.39	0.45	3.13
	(100%)	100.00%	45.38%	26.57%	3.53%	24.53%
PM	(104 tons)	6.32	4.16	1.85	0.24	0.07
	(100%)	100.00%	65.82%	29.27%	3.80%	1.11%

Note: the original data were sourced from the Second National Pollution Source Census Bulletin of Chengdu. Only major sources of atmospheric pollution are listed; emissions from centralized pollution control facilities are excluded.

presents emissions data for key pollutants in 2017. That year, Chengdu’s total NOx emissions reached 135,400 tons, with motor vehicles contributing the largest share at 46.38%, far surpassing industrial emissions (17.36%), which is the second-highest source. While direct PM emissions from vehicles are relatively low (1.11%), the contribution of vehicle emissions to PM pollution primarily arises from secondary aerosol formation. For instance, severe haze events are largely driven by the formation of secondary aerosols [26], with VOCs from photochemical reactions in the atmosphere serving as a major source of secondary organic aerosols in urban environments. As shown in Table 1, VOCs emissions in Chengdu totaled 127,600 tons in 2017, with approximately one-fourth originating from motor vehicles, underscoring their role in PM pollution.

A closer examination of the energy structure in Sichuan Province, with Chengdu as its capital, provides insights into the rationale for implementing DRP there. In certain aspects, Sichuan’s energy consumption structure resembles that of the national level; for instance, in 2017, coal accounted for 34.47% of Sichuan’s end-use energy consumption, a figure comparable to the national average of 34.77% in 2020, reflecting the province’s industrial reliance on coal. However, Sichuan demonstrates unique characteristics in its use of electricity and petroleum compared to the rest of China. The province’s electricity is predominantly generated from hydropower, resulting in lower pollution associated with electricity production. Meanwhile, petroleum products constitute a larger share of energy consumption in residential and transportation sectors, underscoring the significance of vehicle emissions in contributing to air pollution. Given this energy structure, DRPs are a particularly appropriate approach for reducing vehicle emissions and improving air quality.

Table 2. The end-use energy consumption structure of Sichuan Province in 2017.

	Coal	Petroleum	Natural Gas	Electricity and Other Energy
Total Final Consumption	34.47%	28.86%	17.01%	19.67%
Residential Consumption	4.16%	33.77%	33.38%	28.69%

Note: The original data were sourced from China Energy Statistical Yearbook 2017.

presents Sichuan’s end-use energy consumption structure in 2017, which further illustrates the policy’s context and relevance and provides a reference for other regions where vehicle emissions are a major contributor to pollution.

Amid expanding urbanization and economic development, the number of cars in Chengdu has grown rapidly over the past decade, from 1.94 million in 2011 to 4.52 million in 2017. This vehicle surge has been accompanied by a continuous deterioration in air quality. In response, the Chengdu traffic management bureau initiated a DRP in early 2012 and introduced stronger policy measures later that year to alleviate environmental pressure. At the same time, the Chengdu government has learned from the Beijing-Tianjin-Hebei region and other cities in China and implemented such measures as the “650 Project for Air Pollution Prevention and Control” and the “Ten Measures for Controlling Haze”. Chengdu further enhanced its DRP in 2018 to address air pollution issues more effectively. We aim to study whether Chengdu’s DRP has significantly improved air quality in this process.

2.3. Common Environmental Problems amid Global Urbanization

The implementation of the DRP in Chengdu mirrors the common challenges encountered by many rapidly urbanizing cities worldwide. As a fast-developing city, Chengdu faces environmental challenges that are common to many developing country cities, such as rapid urbanization, high population density, and a surge in the number of automobile vehicles [27]. For example, New Delhi, India, is often rated as having the most severe air pollution in the world [28], primarily owing to automobile vehicle emissions and industrial pollution [29]. Similarly, cities in developing countries, such as Pakistan [30], Indonesia [31], and Mexico [32], face environmental problems stemming from rapid urbanization.

The DRP pilot is an attempt to solve local urban problems and an important case study to explore the impact of urban traffic management on air quality globally. Specifically, regarding economic growth and development potential, Chengdu is parallel to most large cities in China. Over the past decade, Chengdu’s annual GDP has been approximately half that of Beijing, and in line with that of Wuhan, Chongqing, Suzhou, and Hangzhou. From the perspective of cultural and infrastructural construction, Chengdu’s approach closely aligns with that of China’s new first-tier cities, such as Xi’an, Chongqing, and Qingdao. Thus, Chengdu’s DRP is a more representative case study and could serve as a suitable reference for other large cities.

3. Data and Methodology

3.1. Data

The empirical study uses three datasets. The first comprises air quality indicators, such as the AQI, collected from the China National Environmental Monitoring Centre. The second dataset includes weather indicators from the National Oceanic and Atmospheric Administration of the United States. The third dataset consists of weekend and statutory holiday data collected from holiday arrangement notices issued by the General Office of the State Council of China.

Air quality indicators, like the AQI, include hourly data from all eight nationally controlled monitoring stations in Chengdu. The eight monitoring stations are located in or at Dashixi Road, Junping Street, Jinquan River, Sanwayao, Shahepu, Shilidian, Longquanyi District Government, and Lingyan Temple. Chengdu’s national air quality monitoring network was reconfigured in 2017. The Longquanyi District Government Station was brought online on 28 October 2017, replacing the previous national monitoring station at Liangjiaxiang. As the Liangjiaxiang Station was decommissioned on the same date, before the latest DRP took effect, its air quality data were excluded from this study’s empirical analysis. As

Table 3. The classification of Chengdu air monitoring stations.

Station	Newly Added Restricted Zones	Original Restricted Zones	Outside the Restricted Zones	Clean Control Point
Jinquan River	Yes
Shilidian	Yes
Sanwayao	Yes
Junping Street	Yes
Dashixi Road	Yes
Shahepu		Yes
Longquanyi District Government			Yes (adjacent to the restricted zones)
Lingyan Temple			Yes	Yes

shows, according to the changes in the DRP, these stations are classified into three categories: (1) stations located within newly added restricted areas under the new policy, among which three stations (Jinquan River, Shilidian, and Sanwayao) are located between the Third Ring Road and the Ring Expressway, and two stations (Junping Street and Dashixi Road) are located within the Second Ring Road; (2) one station (Shahepu) is located within the original restricted area between the Second and Third Ring Roads; and (3) two stations (Longquanyi District Government and Lingyan Temple) are located outside the restricted area, both situated outside the city.

In this study, Lingyanshi Station, one of two stations located outside the restriction area, serves as the clean control point; its data are used only in robustness checks. Conversely, data from the Longquanyi District Government Station and six additional stations are incorporated into all regression analyses. Although located outside the restricted zone, the Longquanyi District Government Station is situated in one of Chengdu’s central urban areas, close to the restricted area, approximately 9 km from the Ring Expressway. By contrast, Lingyan Temple Station is located in Dujiangyan, a county-level city under Chengdu’s administration, far from the restricted area and approximately 44 km from the belt expressway in a straight line. This study includes a comparative analysis of overall air quality and its variations before and after the policy’s implementation at each station. The detailed results are presented in Supplementary Material, Table S1. The findings indicate that air quality fluctuations at Lingyan Temple Station differ significantly from those observed at other stations. Specifically, the AQI at Lingyan Temple Station shows no significant changes post-implementation, contrasting sharply with the significant variations observed at the other stations, significant at the 1% level.

Regarding observation selection, given that the latest DRP took effect on 22 January 2018, this study chooses a sample of 120 days before and after policy implementation, specifically, hourly data from 23 November 2017 to 22 March 2018, for the RDD analysis. Among the observations, data from 00:00 on 23 November 2017 to 23:00 on 21 January 2018 represent the 60 days before the new policy, and data from 00:00 on 22 January 2018 to 23:00 on 22 March 2018 represent the 60 days following the new policy. Following Ye’s approach [21], this study analyzes hourly data from 30 days before and after the implementation date at each monitoring station for the baseline regression. The robustness test adjusts the window width to include intervals of 15, 45, and 60 days before and after the policy implementation.

3.2. Methodology

3.2.1. Regression Discontinuity Design (RDD)

Following the literature, we adopt the RDD approach to identify the treatment effects of Chengdu’s DRP on air quality [33,34]. The identification formula is as follows:

$$y_{sht}=\alpha +\beta polic{y}_t+\gamma contro{l}_{ht}+f^s(t)+v_s+u_h+{\epsilon }_{sht}$$

(1)

where $y_{sht}$ represents a series of air quality variables collected hourly from each monitoring station, including AQI, PM_2.5, PM₁₀, SO₂, NO₂, and CO. $s$ is the station identifier. $h$ is hours, and $t$ is the date. Furthermore, $polic{y}_t$ is the variable for the latest DRP. When the date is on or after 22 January 2018, $polic{y}_t$ is set as 1, and otherwise, it is 0. $\beta$ is the coefficient of interest and captures the treatment effect of DRPs in air quality. The control variables include weather and holiday variables. $v_s$ is the fixed effect of each air monitoring station, $u_h$ is the hourly fixed effect, and ${\epsilon }_{sht}$ is an idiosyncratic error term.

$f^s(t)$ is a station-specific time trend function, represented as a polynomial function of time. Considering the possibility of differing time trends of the monitoring station before and after the policy implementation, and referring to the different trend function method mentioned in Angrist and Pischke [35], we set the function as

$$f^s(t)={\displaystyle {\sum }_{j=1}^k{\tau }_j^s}{\left(t-t_{122}\right)}^j+{\displaystyle {\sum }_{j=1}^k{\tau }_j^s}polic{y}_t{\left(t-t_{122}\right)}^j$$

(2)

where $t_{122}$ is the date on which the latest DRP was implemented, $t-t_{122}$ is the number of days between the date of observation and DRP implementation date, and $k$ is the order of the time trend. In the empirical analysis, we consider two different time trends, first- and second-order, representing different situations of linear and non-linear time trends, respectively. Our model allows for different time trends for each station before and after implementing the latest DRP, facilitated through the incorporation of interaction terms between time trends $t-t_{122}$ and policy dummy variables $polic{y}_t$ within the model.

3.2.2. Difference-in-Differences (DID) Model

Based on local treatment effects, we focus on the average treatment effects of the DRP policy. If a suitable control group is available, the DID model provides an effective method to estimate the policy effects of the DRP. However, unlike the RDD model, the DID model captures the average treatment effect of the DRP on air quality.

Including clean control point data allows us to define the treatment and control groups and conduct a DID test. The identification strategy is as follows:

$$y_{sht}=\alpha +{\beta }_{1}pos{t}_t+{\beta }_{2}pos{t}_t\ast trea{t}_s+{\beta }_{3}trea{t}_s+\gamma contro{l}_{ht}+v_s+u_h+{\epsilon }_{sht}$$

(3)

where $pos{t}_t$ is a dummy variable; when $t$ is the time after policy implementation, $pos{t}_t$ is 1, and otherwise, it is 0. $trea{t}_s$ is the treatment variable for the monitoring station category. If monitoring station $s$ is located within or near the restricted area, $trea{t}_s$ equals 1; if it is outside and far from the restricted area, $trea{t}_s$ is 0. When a monitoring station is in the main urban area of Chengdu in our study area, $trea{t}_s$ is 1. Specifically, $trea{t}_s$ equals 1 for the other seven stations, except for the clean control point. The control variables are consistent with the baseline settings.

3.2.3. Seasonal Trend Adjustment

Air pollution often exhibits seasonal fluctuations. To verify whether the impact of the DRP on air quality stems from the policy itself, rather than from seasonal variations, we apply seasonal trend adjustments to the air quality indicators in our robustness checks. Following established research practices [36,37,38], we regressed each air quality indicator—AQI, PM_2.5, PM₁₀, SO₂, NO₂, CO—on seasonal dummy variables and used the sum of the regression residuals and the original series mean as the deseasonalized data.

Specifically, for each air quality indicator, we use the following model to remove seasonal trends:

$$y_t={\eta }_0+{\eta }_{1}Sprin{g}_t+{\eta }_{2}Summe{r}_t+{\eta }_{3}Autum{n}_t+{\epsilon }_t$$

(4)

where $y_t$ represents the air quality indicator at time t, and $Sprin{g}_t$, $Summe{r}_t$ and $Autum{n}_t$ are dummy variables for the spring, summer, and autumn seasons, respectively, with winter as the base category. The residuals ${\epsilon }_t$ from this regression capture the seasonal fluctuations removed from the original air quality indicators.

To maintain the average level of the data, we construct the deseasonalized data by adding the mean of the original series back to the residuals, obtaining the following equation:

$$y_{deseasonlized,t}=\overline{y}+{\widehat{\epsilon }}_t$$

(5)

where $\overline{y}$ is the mean of the original air quality indicator series. This ensures that the deseasonalized data retain the overall mean level of the original series while removing seasonal effects.

These deseasonalized data are subsequently used in robustness checks to further assess the impact of the DRP on air quality, isolating the effects of the policy from seasonal fluctuations.

3.3. Variables

3.3.1. Dependent Variables

Air quality indicators, including AQI, PM_2.5, PM₁₀, SO₂, NO₂, and CO, are measured hourly at each monitoring station, providing detailed data on urban air pollution levels. These indicators are closely associated with vehicle emissions, a major contributor to urban pollution. For example, in 2017, motor vehicles emitted 6.412 million tons of NOx, accounting for 47.6% of China’s total NOx emissions [39]. Additionally, vehicle emissions contribute approximately 10–50% of urban PM pollution, serving as a primary source of PM in some cities [40]. Furthermore, over 80% of CO in large city atmospheres originates from vehicle emissions [41]. While SO₂ largely stems from coal combustion, vehicle emissions still account for approximately 1.3% of SO₂ levels in major cities [41]. Additionally, vehicle exhaust contains significant amounts of VOCs, which play a role in photochemical reactions with NOx under ultraviolet light, forming ozone (O₃) and secondary organic aerosols that contribute to PM2.5 formation and elevate ground-level O₃ concentrations. For the preceding reasons, we use AQI as a comprehensive evaluative metric for the concentrations of the following six pollutants: PM_2.5, PM₁₀, SO₂, NO₂, CO, and O₃, providing a broader reflection of the levels of air quality [42]. Previous studies commonly use multiple air quality indicators to assess pollution control policies [43,44]. Building on this literature, we include AQI, PM_2.5, PM₁₀, SO₂, NO₂, and CO in our analysis to comprehensively evaluate the impact of the DRP on air quality.

3.3.2. Variable of Interest

Following the implementation of the latest DRP, the value of the DRP variable is 1, and otherwise, it is 0. Specifically, from 22 January 2018, the value of this variable is 1; at all other times, it is 0.

3.3.3. Control Variables

This study’s control variables include weather-related factors and holiday-related factors. Given that weather variables influence pollutant dispersion and decomposition rates, we include three types of weather variables. These weather-related control variables are commonly used as standard controls in existing research on air pollution [45,46,47]. The first category includes ground weather variables, such as temperature, dew point, humidity, air pressure, and wind speed. The second category comprises weather conditions represented by three dummy variables: clear, foggy, and rainy. The third category involves the wind direction. Given the potential negative impacts of severe exhaust emissions from key enterprises, we include six wind directions based on the locations of Chengdu’s major pollution sources: northwest, west, southwest, east, northeast, and south. We also include a dummy variable to identify the wind direction from non-exhaust gas pollution sources, distinguishing them from calm conditions. Additionally, we control for two holiday-related variables—regular weekends and statutory holidays—which are associated with changes in air quality due to human and economic activities patterns.

Table 4. Descriptive statistics.

Variable	Definition/Unit	N	Mean	St. Dev	Min	Max
AQI	AQI index	9282	130.943	67.421	0	452
PM2.5	μg/m3	9282	95.628	53.86	0	293
PM10	μg/m3	9282	146.132	81.05	0	552
SO2	μg/m3	9282	10.197	5.819	0	161
NO2	μg/m3	9282	58.088	30.27	0	198
CO	μg/m3	9282	1.19	0.493	0	7.5
policy	DRP policy variable	9282	0.515	0.5	0	1
temperature	°C	9282	7.364	3.892	−3	18
dew point	°C	9282	2.157	3.174	−9	9
humidity	%	9282	0.726	0.188	0.19	1
pressure	hPa	9282	1021.963	6.063	1006	1036
Wind_spd	km/h	9282	5.818	4.7	0	32.4
weath_sun	sunny	9282	0.17	0.375	0	1
weath_fog	foggy	9282	0.336	0.472	0	1
weath_rain	rainy	9282	0.091	0.288	0	1
wind_nw	northwest wind	9282	0.035	0.185	0	1
wind_n	north wind	9282	0.01	0.099	0	1
wind_sw	southwest wind	9282	0.122	0.328	0	1
wind_e	east wind	9282	0.02	0.139	0	1
wind_ne	northeast wind	9282	0.146	0.353	0	1
wind_s	south wind	9282	0.033	0.179	0	1
wind_oth	Other wind directions	9282	0.495	0.5	0	1
weekend	regular weekends	9282	0.221	0.415	0	1
holiday	statutory holidays	9282	0.149	0.356	0	1

presents the descriptive statistics of the variables.

4. Results

4.1. Baseline Results

4.1.1. Trend Analysis of Air Quality Indicators Around DRP Implementation

To preliminarily assess the impact of Chengdu’s DRP on air quality, we present a linear time trend fitting plot of air quality before and after policy implementation in Figure 2. As the figure shows, after the implementation of the DRP, air quality indicators, such as the AQI, showed a significant downward trend. Additionally, we present non-linear time trend fitting graphs in Supplementary Material, Figure S1. As in Figure 2, a significant discontinuity at the cutoff point is evident in air quality indicators. These results show that the latest round of driving restrictions in Chengdu might have positively affected air quality. Consequently, in the subsequent sections, we examine the significance of the treatment effect of DRP.

4.1.2. Without Time Trend

To examine the impact of Chengdu’s DRP on various air quality indicators, we estimate Equation (1) using hour-by-station observations before and 30 days after the temporary DRP. To ensure the accuracy of the results, we control for weather and holiday variables but not for time trends.

Table 5. Aggregate effect of DRPs on air quality without time trend.

	(1)	(2)	(3)	(4)	(5)	(6)
	AQI	PM2.5	PM10	SO2	NO2	CO
policy	−7.698 **	−3.867	−20.154 ***	−1.788 *	−11.624 ***	−0.134 **
	(2.744)	(2.161)	(4.135)	(0.861)	(2.419)	(0.047)
temperature	4.435 ***	3.712 ***	0.717	−0.187	−3.282 ***	0.013 ***
	(0.961)	(0.785)	(1.172)	(0.130)	(0.808)	(0.003)
dew_point	2.682 ***	1.833 ***	6.845 ***	0.283 **	3.964 ***	0.007
	(0.578)	(0.451)	(0.915)	(0.088)	(0.681)	(0.004)
humidity	104.443 ***	98.603 ***	−21.666	−13.458 ***	−69.315 **	1.012 ***
	(18.342)	(15.865)	(24.191)	(3.008)	(19.212)	(0.099)
pressure	2.409 ***	1.102 ***	4.299 ***	−0.049	0.722 ***	0.003 **
	(0.226)	(0.124)	(0.249)	(0.033)	(0.074)	(0.001)
wind_spd	−3.013 ***	−2.268 ***	−3.682 ***	−0.160 ***	−1.411 ***	−0.013 ***
	(0.098)	(0.097)	(0.167)	(0.016)	(0.118)	(0.002)
weath_sun	−12.145 ***	−5.316 ***	−21.423 ***	−0.188	−0.537	0.034 **
	(1.085)	(0.776)	(1.568)	(0.245)	(0.558)	(0.010)
weath_fog	48.427 ***	42.140 ***	45.332 ***	2.676 ***	12.478 ***	0.258 ***
	(2.061)	(1.949)	(2.873)	(0.352)	(1.330)	(0.022)
weath_rain	−19.155 ***	−14.571 ***	−24.838 ***	−0.372	−6.443 ***	−0.203 ***
	(1.257)	(0.974)	(1.828)	(0.263)	(1.668)	(0.033)
wind_nw	19.266 ***	16.293 ***	12.952 ***	1.119 **	1.619	−0.012
	(1.678)	(1.380)	(2.516)	(0.356)	(2.187)	(0.032)
wind_n	21.549 ***	2.564	31.873 ***	−1.121 ***	−6.879 *	−0.071 **
	(3.101)	(1.657)	(3.140)	(0.250)	(3.428)	(0.024)
wind_sw	15.109 ***	11.642 ***	10.609 ***	0.584 **	5.190 ***	0.004
	(0.818)	(0.622)	(1.912)	(0.196)	(1.338)	(0.014)
wind_e	−1.845 **	−0.039	−5.708 ***	−0.275	−3.556 **	0.056 **
	(0.704)	(0.499)	(1.518)	(0.322)	(1.369)	(0.018)
wind_ne	8.577 ***	4.232 ***	6.285 ***	0.041	−0.300	0.032
	(1.379)	(0.433)	(1.689)	(0.331)	(1.691)	(0.028)
wind_s	15.213 ***	5.597 ***	17.253 ***	−0.240	3.529 ***	−0.054
	(1.492)	(1.016)	(1.486)	(0.239)	(0.900)	(0.032)
wind_oth	14.540 ***	11.114 ***	11.809 ***	0.638 **	2.868 *	0.024
	(0.678)	(0.446)	(1.047)	(0.178)	(1.266)	(0.022)
weekend	0.272	−0.411	−4.926 *	0.823 **	−1.005	−0.042 **
	(1.908)	(1.431)	(2.193)	(0.262)	(0.671)	(0.014)
holiday	8.936 *	−6.533 **	31.094 ***	−1.190	−14.776 ***	−0.050
	(3.752)	(2.379)	(3.786)	(0.635)	(0.780)	(0.059)
Obs.	9282	9282	9282	9282	9282	9282
R-squared	0.432	0.456	0.398	0.250	0.422	0.380

Note: the values in parentheses are robust standard errors; significance levels: *** p < 0.01, ** p < 0.05, * p < 0.1. The same applies to subsequent tables.

presents the baseline results.

Columns (1)–(6) of Table 5 show AQI, PM_2.5, PM₁₀, SO₂, NO₂, and CO treatment effects. The first column shows that the DRP in Chengdu significantly reduces the AQI, markedly improving air quality in restricted areas. The results of the policy effects on the other five pollutants are presented in Columns (2)–(6), and we arrive at the same conclusion. The treatment coefficients for PM₁₀, SO₂, NO₂, and CO are negative and statistically significant. For PM_2.5, although the policy coefficient is not statistically significant, it remains negative. These results indicate that the DRP in Chengdu has improved air quality.

The impact of weather variables on air quality is significant, yet the direction of influence varies. Temperature, dew point, humidity, and air pressure significantly increase the AQI and reduce air quality, primarily because of increased PM_2.5 and PM₁₀ levels. Wind speed significantly improves air quality and reduces the AQI and levels of five pollutants, including PM_2.5. Regarding weather category indicators, sunny and rainy days improve air quality; however, foggy weather suppresses air quality.

Regarding the wind direction variables, winds from most directions increase the AQI level, except for eastward winds, which significantly reduce it. Winds from other directions increase air pollution levels in Chengdu. This is because the wind direction data selected in this study are used to control the potential impact of waste gas pollution source enterprises and to prevent the omission of important control variables. It is evident that the air pollution in Chengdu is most severe for winds from the west and northwest directions. This result is easily understood, because many key pollution source enterprises are located in the western, northwest, and southwest regions, as discussed in Section 3.

4.1.3. With Time Trend

Although regression discontinuity estimators effectively avoid the endogeneity issues caused by time trends, the literature recommends using local linear and quadratic polynomial time trends in studies [48]. Controlling for the time trend effect can further mitigate endogeneity issues caused by policy factors that change over time. Therefore, we incorporate linear and quadratic polynomial time functions into Equation (1), and the results are displayed in Columns (1) and (2) of

Table 6. Aggregate effect of DRPs on air quality with time trend.

	Symmetric Time Trend		Asymmetric Time Trend
	(1)	(2)	(3)	(4)
	Linear	Quadratic	Linear	Quadratic
AQI	−10.082 **	−13.015 ***	−11.092 **	−79.684 ***
	(3.492)	(3.302)	(3.467)	(3.553)
PM2.5	−11.362 ***	−14.012 ***	−12.212 ***	−62.726 ***
	(2.500)	(2.309)	(2.466)	(2.970)
PM10	−15.662 ***	−18.448 ***	−16.899 ***	−86.512 ***
	(4.080)	(3.849)	(4.026)	(2.435)
SO2	−3.568 ***	−2.986 ***	−3.271 ***	−4.033 ***
	(0.730)	(0.628)	(0.676)	(0.861)
NO2	−9.754 ***	−7.671 ***	−8.576 ***	−28.793 ***
	(1.522)	(1.434)	(1.472)	(2.417)
CO	0.018	−0.007	0.013	−0.200 ***
	(0.055)	(0.056)	(0.055)	(0.051)

Note: only the results for the policy variable are shown for each regression; all other control variables are consistent with the baseline regression.

. Furthermore, given that the time trend before and after the DRP implementation may vary, we also control for the interaction between time trends and policies. Following Angrist and Pischke [35], we integrate Equation (2) into Equation (1) for our analysis. The regression results are shown in Columns (3) and (4) of Table 6. The asymmetric time trend findings are consistent with those of the symmetric time trend. Both analyses suggest that the DRP has a good improvement effect on other air pollution indicators, except for CO levels.

4.2. Robustness Test

We assess the results using various sample restrictions and identification models. Specifically, we (1) change the choice of the window width; (2) add a lagged term; and (3) use an alternative difference-in-differences (DID) model.

4.2.1. Alternative Time Window

The bandwidth choice when using an RDD is an unresolved issue. Generally, selecting a narrower bandwidth may result in fewer samples, whereas a longer bandwidth reduces the randomness of observations and diminishes the estimation accuracy of the RDD. Nevertheless, following the literature, we expand the time window from 15 to 60 days for robustness testing.

Table 7. Alternative time window.

	15-Day Window Width	45-Day Window Width	60-Day Window Width
AQI	−28.798 ***	−13.396 ***	−10.499 ***
	(2.113)	(2.275)	(1.388)
PM2.5	−23.889 ***	−10.516 ***	−9.004 ***
	(1.648)	(1.896)	(1.176)
PM10	−35.114 ***	−21.855 ***	−12.501 ***
	(3.421)	(3.241)	(2.098)
SO2	−2.745 ***	−2.193 **	−1.968 *
	(0.733)	(0.844)	(0.860)
NO2	−13.347 ***	−12.324 ***	−7.805 ***
	(1.757)	(1.975)	(1.578)
CO	−0.129 **	−0.101 **	−0.045
	(0.053)	(0.041)	(0.025)

Note: control variables are consistent with the baseline model, and the same applies to subsequent tables.

lists the regression results for the bandwidth adjustments. Columns (1)–(3) display the RDD regression results for window widths of 15, 45, and 60 days. As expected, the bandwidth adjustment results are consistent with the baseline findings. The DRP significantly improves all pollution indexes. Moreover, the policy effect diminishes as the bandwidth increases, suggesting that the impact lessens over time.

4.2.2. Add Lagged Term

Generally, socio-economic and environmental indicators that change over time may exhibit time-series correlations that may be influenced by the developmental characteristics of the previous period. Given these correlations, the current air quality is likely to be influenced by the air conditions of the last period. Therefore, we incorporate a lagged term for the air quality variables into the regression model. The results presented in

Table 8. Add lagged term.

	(1)	(2)	(3)	(4)	(5)	(6)
	AQI	PM2.5	PM10	SO2	NO2	CO
policy	−0.490 **	−0.407 **	−1.282 ***	−0.459	−2.488 ***	−0.034 ***
	(0.182)	(0.136)	(0.333)	(0.377)	(0.613)	(0.008)
LagAQI	0.930 ***
	(0.007)
LagPM2.5		0.933 ***
		(0.006)
LagPM10			0.928 ***
			(0.011)
LagSO2				0.753 ***
				(0.098)
LagNO2					0.801 ***
					(0.019)
LagCO						0.761 ***
						(0.036)
Control	Y	Y	Y	Y	Y	Y
Obs.	9282	9282	9282	9282	9282	9282
R-squared	0.930	0.938	0.922	0.675	0.813	0.753

show that after accounting for the impact of lagged terms, the DRP still significantly improves air quality, albeit with some fluctuation in the significance of SO₂. Consistent with conventional expectations, the regression findings in Table 8 demonstrate that air conditions from the previous period significantly impact current air quality, exhibiting a strong positive correlation between the two periods.

4.2.3. Difference-in-Differences Model

In this subsection, we use the DID model to capture the average treatment effect of the DRP on air quality.

Table 9. Difference-in-differences results.

	(1)	(2)	(3)	(4)	(5)	(6)
	AQI	PM2.5	PM10	SO2	NO2	CO
post × treat	−25.581 ***	−14.617 ***	−33.923 ***	2.221	−8.005 ***	−0.002
	(3.061)	(2.512)	(4.564)	(1.358)	(2.866)	(0.051)
post	16.480 ***	9.358 ***	13.470 ***	−5.296 ***	−4.148 ***	−0.142 ***
	(1.493)	(1.452)	(0.653)	(1.367)	(0.793)	(0.012)
treat	59.551 ***	45.620 ***	62.733 ***	−6.023 ***	31.288 ***	0.310 ***
	(4.260)	(3.380)	(6.905)	(1.174)	(5.092)	(0.072)
Control	Y	Y	Y	Y	Y	Y
Obs.	10,608	10,608	10,608	10,608	10,608	10,608
R-squared	0.408	0.431	0.370	0.373	0.395	0.346

presents the regression results from the DID model. The interaction term “post × treat” is our variable of interest in the DID model. The “post × treat” coefficient represents the treatment effect on various air pollution indicators under the DRP. As shown, Chengdu’s DRP significantly reduces air pollution indicators, including the AQI, PM_2.5, PM₁₀, and NO₂. The source analysis results of atmospheric PM in Chengdu by the Chengdu Academy of Environmental Sciences reveal that vehicles directly contribute more than 10% to PM₁₀ and PM_2.5 levels. NOx and hydrocarbons from automobile exhausts are other main sources of air pollution.

Consequently, the DRP significantly improves PM_2.5, PM₁₀, and NO₂. However, the primary source of SO₂ in China’s atmosphere is the combustion of sulfur-containing fuels, like coal, rather than automobile emissions. This likely explains why the policy has not significantly impacted SO₂ levels. The results in Table 9 demonstrate that the DRP enhances air quality considerably.

4.2.4. Seasonal Trend Adjustment

Air pollution often exhibits seasonal patterns. To verify that the impact of the DRP derives from the policy itself, rather than seasonal patterns, we apply seasonal trend adjustments to the data and conduct robustness tests over an extended period.

Figure 3 presents hourly variations in air quality from 1 January 2016 to 31 January 2020, covering a 2-year period on either side of the policy implementation. In Figure 3, the blue line illustrates the linear fit, while the green curve represents the local polynomial fit. The downward slope of the blue fitted line suggests an overall improvement in air quality after the DRP was implemented. By contrast, the green curve highlights distinct seasonal fluctuations in air quality across each year.

Using seasonally adjusted data (as explained in Section 3.2.3), and to maintain consistency with the baseline analysis, we applied a linear fit to illustrate air quality trends before and after implementing the policy. Figure 4 presents a comparison of air quality during corresponding periods in other years. To better illustrate the policy’s effect, we select three dates—22 January 2017; 22 January 2018; and 22 January 2019—and plot breakpoint graphs around each, covering 60 days before and after, using the AQI, PM_2.5, and PM₁₀ as illustrative examples. The results reveal a distinct improvement in air quality around 22 January 2018, the policy’s actual implementation date, with no comparable breakpoint effect evident around the same date in 2017 and 2019. This pronounced difference further corroborates the positive impact of the DRP on air quality.

Finally, we perform a regression on seasonally adjusted data to assess the DRP’s effect over extended time spans and various window settings.

Table 10. Regression results on seasonally adjusted data.

Window Width	30 Days	60 Days	90 Days	180 Days	365 Days
AQI	−34.350 ***	−35.118 ***	−24.821 ***	−14.636 ***	−8.991 ***
	(1.913)	(1.247)	(0.823)	(0.571)	(1.234)
PM2.5	−24.485 ***	−27.508 ***	−20.324 ***	−10.148 ***	−5.753 ***
	(1.422)	(1.007)	(0.883)	(0.628)	(1.167)
PM10	−42.644 ***	−37.909 ***	−22.306 ***	−10.759 ***	−6.644 **
	(2.424)	(1.545)	(1.141)	(0.720)	(1.989)
SO2	−3.148 ***	−3.427 ***	−2.776 **	−1.193	−1.867 **
	(0.833)	(0.788)	(0.821)	(0.976)	(0.751)
NO2	−0.306 ***	−0.254 ***	−0.198 ***	−0.101 ***	−0.048 **
	(0.032)	(0.019)	(0.024)	(0.017)	(0.019)
CO	−16.597 ***	−12.149 ***	−5.769 ***	−1.936	−4.392 **
	(1.908)	(1.139)	(1.013)	(1.084)	(1.196)

displays regression results across windows ranging from 30 to 365 days. The analysis reveals that across all window settings, the DRP exerts a significant negative impact on each pollution indicator, strongly supporting the conclusions of the baseline analysis. This suggests that, even after controlling for seasonal fluctuations, the DRP’s positive impact on air quality remains significant, further strengthening the robustness of the results.

5. Heterogeneity Analysis

Following the introduction of a new round of DRPs in Chengdu, diverse responses among individuals may lead to heterogeneous policy outcomes. Some residents switch to commute by subway or bus, potentially enhancing urban air quality. Conversely, others may bypass the DRP’s restrictions by detouring through nearby non-restricted areas or altering travel times. Therefore, the DRP might spatially and temporally redistribute traffic flow, which could negatively impact air quality in areas and at times not under restriction. Consequently, this study conducted a heterogeneity analysis to ascertain whether the policy’s impact differs across various locations and times.

5.1. Spatial Heterogeneity

Following the implementation of the new round of the DRP, the central urban area of Chengdu is divided into three zones: original restricted zones, newly added restricted zones, and non-restricted zones adjacent to the restricted zones. We categorize the sample data into three groups based on the monitoring station locations and conduct grouped regressions to explore the spatial heterogeneity effects of DRPs on air quality. This heterogeneity analysis assesses the impact of the DRP on air quality in both the new and original restricted zones and its potential negative effects on the non-restricted zones adjacent to the restricted zones.

Table 11. Spatial heterogeneity analysis.

	Newly Added Restricted Zones	Original Restricted Zones	Adjacent to the Restricted Zones
AQI	−10.350 **	−7.612 **	5.480 *
	(2.309)	(3.450)	(3.312)
PM2.5	−5.801 *	−3.437	5.373 **
	(2.141)	(2.609)	(2.489)
PM10	−24.623 ***	−17.581 ***	−0.387
	(3.310)	(4.425)	(4.057)
SO2	−1.230 ***	−6.726 ***	0.358 **
	(0.153)	(0.350)	(0.179)
NO2	−14.310 ***	−9.693 ***	−0.122
	(1.911)	(1.384)	(1.176)
CO	−0.173 **	−0.160 ***	0.088 ***
	(0.041)	(0.024)	(0.031)

presents the results of the spatial heterogeneity analysis. As expected, the results indicate that the DRP significantly improves the air quality in the newly added and original restricted zones, with a more pronounced improvement in the newly added zones. However, in non-restricted zones, the DRP promotes the diversion of vehicle travel, leading to spatially substitutive travel choices by residents. This spatial selection behavior results in negative externalities and reduces air quality in the non-restricted zones adjacent to the restricted zones.

5.2. Temporal Heterogeneity

The DRP improves air quality by reducing vehicle traffic volume and decreasing exhaust emissions. Traffic reduction alleviates congestion and reduces additional emissions. Consequently, we hypothesize that air quality improvements attributable to the DRP would be more pronounced during peak travel periods. However, to avoid the effects of DRPs, people may modify their travel times. Thus, following DRP implementation, we anticipate increased traffic volume during periods adjacent to the restricted times. Essentially, the DRP may temporally redirect traffic flow.

Consequently, residents adopt temporally substitutive travel patterns in response to DRPs. We conduct a detailed analysis of several key periods to verify these hypotheses. Given that commuting and work primarily occur during the day, we analyze the period from 07:00 to 21:00. The peak commuting periods during the restricted times generally range from 08:00 to 09:00 and 18:00 to 19:00. The off-peak periods during the restricted times are generally from 10:00 to 17:00 and from 20:00 onwards. Additionally, the non-restricted periods adjacent to the restricted times are 07:00 and 21:00, each 1 h before and after the restricted times, respectively. Therefore, we conduct a temporal heterogeneity analysis based on three time periods—peak, off-peak, and non-restricted periods adjacent to restricted times—to determine whether residents engage in temporally substitutive travel.

The regression results presented in

Table 12. Temporal heterogeneity analysis.

	Restricted Periods	Restricted Periods	Non-Restricted Periods
	Peak Periods	Off-Peak Periods	Adjacent to the Restricted Times
	08:00–09:00 and 18:00–19:00	10:00–17:00 and 20:00	07:00 and 21:00
AQI	−11.185 ***	−1.576	6.777
	(2.134)	(2.545)	(4.212)
PM2.5	−11.605 ***	−4.188 **	3.432
	(1.423)	(2.086)	(3.336)
PM10	−13.502 ***	1.769	4.650
	(2.417)	(3.047)	(4.376)
SO2	−2.209	−2.152 ***	−1.814 **
	(1.325)	(0.238)	(0.658)
NO2	−7.066 ***	−6.698 ***	−12.665 **
	(1.398)	(1.215)	(4.508)
CO	−0.193 ***	−0.156 ***	−0.145 *
	(0.045)	(0.020)	(0.061)

Note: the regression is conducted solely on samples within restricted areas, excluding weekends and holidays. The model and control variables are consistent with those used in the baseline regression.

show that the DRP’s effect on improving air quality persisted throughout the restricted period, with more substantial improvement during peak periods than off-peak periods. However, during periods adjacent to the restricted times, the DRP did not significantly improve air quality. Regression analyses of AQI, PM_2.5, and PM₁₀ indicate the DRP may even negatively affect air quality during these periods adjacent to restrictions. Although insignificant, this effect supports the hypothesis that residents opt for alternative travel times.

6. Discussion and Conclusions

This study aimed to assess the effectiveness of DRPs in improving air quality, focusing on Chengdu as a case study representative of large cities in China. In today’s rapidly advancing transportation industry, motor vehicle pollution has become a major factor affecting urban air quality. Many cities in China have implemented DRPs to mitigate motor vehicle pollution and traffic congestion. However, it is essential to assess their effectiveness.

Using hourly data and RDD, this study examined the impact of Chengdu’s DRP on air quality. The results demonstrate that the DRP significantly improved air quality in Chengdu, markedly reducing the AQI levels and pollution concentrations related to motor vehicle emissions, including PM_2.5, PM₁₀, SO₂, NO₂, and CO. Additionally, as residents may adjust their travel times and routes to circumvent the DRP, we further analyzed the heterogeneous effects of the DRP on air quality across different zones and periods. The results indicate that although the DRP generally improved air quality in Chengdu, its effects varied significantly across different local zones and periods. The DRP mainly improved air quality within restricted areas and during restricted times. However, the DRP failed to improve and even worsened the air quality in non-restricted areas adjacent to the restricted zones and during non-restricted times. These results suggest that although DRPs are generally beneficial for improving urban air quality, their effectiveness is limited. The findings indirectly suggest that DRPs negatively affect residents’ travel convenience, leading them to circumvent the policy by altering their travel routes and times.

While this study provides valuable insights into the effects of DRPs, it has some limitations owing to Chengdu’s unique characteristics. First, Chengdu’s basin-like geography may affect air circulation and pollutant dispersion, potentially influencing air quality independently of DRP effects. Although we incorporated weather conditions (e.g., sunny, foggy, rainy, and wind speed) as indirect controls to account for geographical characteristics, it remains challenging to fully capture all geographic influences. Second, examining the residential component of end-use energy consumption revealed that Chengdu’s residents primarily rely on petroleum, followed by natural gas and electricity (predominantly hydropower-based). This structure results in relatively low reliance on coal in the residential sector, making vehicle emissions a more prominent contributor to urban air pollution. This pattern contrasts with cities heavily reliant on coal, particularly those in colder climates where coal heating significantly impacts pollution levels [49]. This distinction is also closely related to Chengdu’s subtropical monsoon climate, which has mild winters and minimal heating requirements. Given that this climate zone is characteristic of many densely populated regions worldwide, the findings from Chengdu’s DRP may be broadly applicable to similar urban environments where vehicular emissions dominate pollution sources. However, future research should incorporate a broader range of city types with diverse geographical, structural, and energy characteristics to enhance the generalizability of the findings and provide insights into how these variables interact with DRP effects. Moreover, as highlighted by China’s New Energy Vehicle Industry Development Plan (2021–2035), the transition toward cleaner fuels and electric vehicles offers a potential avenue for improving urban air quality. Although fuel-powered vehicles will likely dominate for the foreseeable future, exploring these additional factors alongside DRPs could provide a more comprehensive understanding of urban air pollution. Multi-city analysis across varied urban contexts would further elucidate the nuanced impacts of DRPs. It can guide policymakers in adapting strategies to fit different regional conditions.

The conclusions of this study inform policy recommendations for countries seeking to manage motor vehicle emission pollution. DRPs, which are crucial measures for urban environmental governance, have undeniable positive effects. They alleviate traffic congestion, reduce motor vehicle exhaust emissions, and significantly improve air quality. However, attention must also be paid to the potential negative externalities of DRPs. It is necessary to consider the potential impact of air quality deterioration in non-restricted areas adjacent to restricted zones and to balance these factors in policy formulation. Moreover, in cities with energy structures dominated by coal-based power, DRPs alone may not be sufficient to address air pollution; rather, complementary policies targeting industrial and residential emissions may be necessary.

Additionally, when implementing policies, multiple factors should be considered to ensure environmental improvements while meeting residents’ transportation needs. To consider social welfare and policy effectiveness comprehensively, governments should ensure that DRP implementation is coordinated with related policies. Governments should upgrade and invest in the public transportation system, enhancing its convenience, comfort, and coverage to make it the preferred choice for residents. Simultaneously, continuous optimization of the refined management of DRPs is necessary to control the policy intensity, ensuring that it reduces traffic emissions while minimally affecting residents’ daily travel needs. Based on the characteristics of the city and using big data analysis, differentiated and detailed DRPs should be formulated, implementing restrictions by region, period, and vehicle type to balance traffic flow and air quality.