The effects of PM_2.5 concentrations on traffic violations and accident severity in Guangdong China

Air pollution - especially fine particulate matter (PM_2.5) - poses a major global health burden, causing an estimated 4.2 million premature deaths in 2019¹ and ranking as the second leading risk factor for mortality worldwide². A large body of evidence links PM_2.5 exposure to cardiovascular disease, respiratory illness, and cancer^3,4, with sources ranging from natural events such as wildfires to anthropogenic emissions from industrial activity and vehicle exhaust^5,6.

However, the societal costs of PM_2.5 extend well beyond physical health impacts. A growing body of research documents wide-ranging behavioral and economic externalities - from increased criminal activity⁷ to reduced labor participation⁸. Among these, the traffic safety externalities are particularly critical, given annually that road accidents remain a global crisis, claiming approximately 1.19 million lives annually⁹. However, the empirical evidence regarding air pollution’s impact on traffic safety remains inconclusive. While some studies demonstrate a positive association between elevated PM_2.5 levels and increased fatal accidents¹⁰, others, such as Shr, et al.¹¹, suggest that pollution-induced visibility reduction may actually deter risky driving behaviors in specific contexts.

Two principal mechanisms may link PM_2.5 exposure to traffic risk. The first involves physiological effects, where exposure to fine particulates impairs cognitive function and emotional regulation, potentially leading to slower reaction times, reduced attention, and a greater propensity for risk-taking or error^12,13. The second involves environmental effects, as PM_2.5 scatters and absorbs light, which can significantly reduce visibility - sometimes to below 100 m - and thereby increase the likelihood of collisions^14,15.

To investigate the impact of air pollution on road safety - including both pre-accident behaviors (traffic violations) and post-accident outcomes (accident severity) - we match traffic accident records from the Road Traffic Accident Database of the Ministry of Public Security of China with daily PM_2.5 concentrations at the county level. Our analysis focuses on all reported traffic accidents in Guangdong Province from 2006 to 2014. The dataset includes 186,641 records of driver violations extracted from 118,240 accidents, along with detailed measures of accident severity. However, identifying the causal effect of air pollution on traffic violations and accident severity poses substantial challenges, as pollution and traffic incidents are both influenced by structural, temporal, and weather-related factors¹⁶. Moreover, endogeneity arises because vehicles themselves emit air pollutants, potentially biasing the estimated effects. To address these challenges, we adopt three strategies. First, we include city and time fixed effects in our models to account for spatial heterogeneity and temporal trends. Second, we use temperature inversions as an instrumental variable to isolate exogenous variation in PM_2.5 concentrations. Third, we apply a random forest model to perform weather normalization of PM_2.5, helping to eliminate bias stemming from complex correlations between pollution and meteorological conditions.

In this paper, we contribute to the literature by examining whether local PM_2.5 concentrations affect traffic violations and accident severity. Our results show that higher PM_2.5 levels are associated with an increase in both the severity of traffic violations and accident outcomes. However, these effects exhibit substantial heterogeneity across driver characteristics and geographic areas. Male drivers and those with lower educational attainment appear more susceptible to the influence of PM_2.5, in terms of both traffic violations and accident severity. Moreover, as drivers age, the impact of PM_2.5 shifts from influencing traffic violations to affecting accident severity. In economically developed regions, PM_2.5 concentrations primarily affect traffic violations, while in less developed areas, they mainly impact accident severity.Furthermore, although data limitations prevent us from directly incorporating drivers’ physiological or psychological states into our model, we explore indirect evidence to inform future research. Regarding the potential visual mechanism, we find that PM_2.5 tends to have stronger effects under conditions of clear visibility - namely during daytime and in good weather - when drivers may perceive conditions as safe. Regarding the psychological mechanism, we draw on Chapman, et al.¹⁷, who categorize traffic violations into those driven by overconfidence and those driven by inexperience. We find that increases in PM_2.5 levels are associated only with greater severity in overconfidence-related violations and accidents, suggesting that air pollution may exacerbate risk-taking behavior rather than impairing driving skills.

The remainder of this paper is structured as follows. Section Research design introduces our research design, including data sources and model specification. Section Results presents the main empirical results and robustness checks. Section Discussion discusses heterogeneity in the effects and explores potential underlying mechanisms. Section Conclusion concludes with policy implications and future research directions.

Data sources

The traffic accident data in this article comes from the Road Traffic Accident Database of the Ministry of Public Security of China from 2006 to 2014, which includes accident reports from Guangdong province¹⁸. This database is officially and legally recognized in China as the most comprehensive, extensive, and reliable data source for traffic safety research. It contains detailed information for each accident, including the characteristics of the parties involved, the degree of injury, vehicle features, road conditions, the time and environment of the accident, as well as 11 sections of information such as the level of road traffic accidents, accident patterns, accident causes, the degree of vehicle damage, responsibility types of the parties involved, the degree of injury of the parties involved, the body parts injured, travel purposes, vehicle driving status, types of drivers, and the insurance status of the parties involved.

Guangdong province has the highest number of road traffic accidents among China’s 31 provinces, making its traffic safety issue the most serious and prominent in the country. It is the only province in China with a total resident population exceeding 100 million and also has the largest floating population. Therefore, studying traffic accidents in Guangdong provides a representative and typical case¹⁸. The analysis results and policy implications from this study on Guangdong’s traffic accidents can serve as a model for promoting innovative national traffic management efforts. Despite focusing on data from Guangdong, provided by the Ministry of Public Security, this study is broadly applicable to understanding the impact of air pollution on traffic violations and severity in China and other developing countries. Moreover, unlike existing traffic safety literature that relies mainly on observational data and limited sample surveys, this study uses comprehensive data from China’s official traffic management departments. This data spans a wide geographical and temporal range and includes rich accident-related indicators, making it the most complete and authoritative source in Chinese traffic safety research. Additionally, this study offers an intuitive and reliable theoretical and analytical framework that can inspire traffic safety policy formulation in other developing countries¹⁹. It is noteworthy that Dongguan and Zhongshan were not included in our study due to their unique administrative status. Unlike other cities, both were originally counties but were later upgraded to city-level status. Their geographic size and population fall between those of typical cities and counties. To avoid potential bias arising from this distinct classification, observations from these two cities were excluded from the sample.

The air pollution data in this paper comes from satellite inversion data provided by NASA. The original data, with hourly resolution, is averaged and gridded on a daily basis. Finally, the vector data of China’s provinces, autonomous regions, municipalities, and counties is used to clip and summarize the daily PM_2.5 measurements of Guangdong province at the county level from 2006 to 2014.

Variable measure

Dependent variables(1) To analyze the impact of air pollution on traffic violations, i.e., the causal factors leading to traffic accidents, this article uses the penalty points for traffic violations as the dependent variable. In China, each driver is initially given 12 penalty points after passing the driving test and obtaining a driver’s license. Depending on the severity of the violation, the driver may be deducted 0, 1, 2, 3, 6, or 12 points following a traffic violation. (2) To analyze the impact of air pollution on the consequences of accidents, this article uses the severity of accidents as the second dependent variable, with specific classification criteria based on China’s transportation management departments as shown in Table 1. Table 2 (1)-(2) displays the statistical descriptions of these two explained variables. The average penalty points for traffic violations in the data used in this article are 2.019, while the average severity of accidents is 1.554.

Independent variable This paper mainly uses the daily PM_2.5 concentration to represent air pollution, as it is widely recognized as main component that aggravates air pollution²⁰. Figure 1 illustrates the spatial distribution of daily PM_2.5 concentration, penalty points, and accident severity across counties in Guangdong Province. Figure 1 (a) shows that areas with high daily PM_2.5 concentrations are primarily concentrated in the industrially developed Pearl River Delta region (e.g., Guangzhou and Foshan) and its surrounding areas. Comparing this with Fig. 1 (b), we observe that counties with higher average daily PM_2.5 concentrations tend to have higher average penalty points. However, in some counties in eastern and western Guangdong, low PM_2.5 concentrations coexist with high penalty points. Figure 1 (c) highlights differences in the distribution of average accident severity compared to average daily PM_2.5 concentration and average penalty points. Zhang et al. (2013) also found that accident severity results from the combined effects of multiple risk factors, including traffic violations, individual factors, vehicle factors, road factors and environment factors. Therefore, simultaneously examining the impact of daily PM_2.5 concentration on penalty points and accident severity provides a more comprehensive understanding of the traffic-related costs of air pollution. Figure 2 presents the trends in the average of daily PM_2.5 concentrations, driver penalty points, and accident severity from 2006 to 2014. The distribution of PM_2.5 concentrations is right-skewed (Mean = 33.860; SD = 15.485; Skewness = 1.032). To normalize the data, we take the natural logarithm of daily PM_2.5 concentrations (Mean = 3.419; SD = 0.458; Skewness = − 0.108).

Distribution of PM_2.5 Concentration, Traffic Violations and Severity in Guangdong Province. The figure was created using the software QGIS Desktop, version 3.40.2.

Full size image

The Trend of PM_2.5, Point and Severity. The figure was created using the software Stata, version 18.

Full size image

Instrumental variable To address the endogeneity issue caused by mutual causality, this study uses inverse temperature as an instrumental variable. In terms of correlation, inverse temperature is highly correlated with air pollution. In terms of exogeneity, inverse temperature does not cause traffic violations or more severe accident losses.

Control variables: According to previous research, traffic violations and accident severity are influenced by four factors: individuals, vehicles, roads, and the environment¹⁸. Firstly, individual factors. Driver’s gender, age, and driving experience are potential factors affecting traffic accidents and accident losses. In addition, drivers’ socioeconomic conditions also influence their risk preferences and behaviors, while household registration and occupation can serve as proxy variables^18,19. Secondly, vehicle factors. In this article, three types of vehicles are included: automobiles, trucks, and motorcycles. Different types of vehicles have different content in driving tests, and trucks are often heavier than passenger cars, resulting in more severe accident losses²¹. Motorcycles are often subject to serious injuries due to their poor safety performance. Previous studies have shown that the nature of vehicle use is strongly correlated with driver behavior and the resulting accident losses. Commercial vehicles are more likely to cause serious or fatal injuries than private vehicles¹⁸. Thirdly, road factors. Road infrastructure investment can greatly promote economic growth in developing countries, but it is accompanied by more traffic accidents. Protective road facilities are one of the important policy tools for governments to reduce traffic accidents. In this article, roads are classified according to type and grade: highways, ordinary roads, and urban roads. Among them, ordinary roads include first-class roads, second-class and lower-grade roads, while urban roads include urban expressways, urban ordinary roads, and urban other roads. This article also controls the impact of whether there are isolation facilities on the road, the road surface conditions (dry or wet), and whether there are traffic signal facilities. Fourthly, environmental factors. In addition to air pollution, other environmental factors may also affect traffic violations and accident losses. Therefore, this article also controls the lighting conditions and weather at the accident site. Finally, this article also controls the fixed effects of year, month, city, and week.

Order logit model

Our study aims to estimate the effect of air pollution on traffic violations and severity. Although both dependent variables have specific numerical values, they are strictly ordered categorical values. While the penalty points for traffic violations are numerical, each penalty point actually represents a category of violations with similar levels of danger. Additionally, a violation with a penalty score of 2 cannot be considered twice as dangerous as a violation with a penalty score of 1, and the same is true for the severity of accidents. Therefore, this study employs an Ordered Logit model to estimate the impact of PM_2.5 on traffic accidents.

In the Ordered Logit model, the relationship between air pollution and the explained variable can be interpreted as follows: for every increase in PM_2.5 concentration by \(\:q\)%, the likelihood of more severe traffic violations or accident severity increases by \(\:q\beta\:\)%²². The estimated equation is as follows (McCullagh, 1980):

In the above equations, \(\:Y\) represents the explained variable, which refers to the penalty points for traffic violations or the severity of accidents, and \(\:j\) is the categorical labels for different levels of penalty points or severity. Independent variables \(\:{x}_{i}\) included the surface PM_2.5 mass concentration and the control variables. \(\:{\beta\:}_{j0}\) represents the constant term of the regression, and \(\:{\beta\:}_{ji}\)represents the regression coefficient. By using Eq. (1), the cumulative density function can be estimated. After a natural logarithm transformation, Eq. (1) can be converted into Eq. (2), which is easier to calculate and understand. Finally, Eq. (3) is used to calculate the corresponding probability for each level.

It is worth noting that there are two main endogenous issues in the above analysis that need to be addressed. Firstly, there is a reverse causality between air pollution and traffic accidents. On the one hand, air pollution can cause traffic accidents through visual and psychological factors; on the other hand, traffic accidents can also lead to congestion and other problems that will increase air pollution. Secondly, air pollution and traffic accidents have a tendency to change together with the season, which may lead to a spurious correlation rather than a causal relationship between the two. Therefore, this article uses instrumental variable methods and fixed effects models to address the endogenous issues in the model^16,23,24. Specifically, we can expand formula (2) as follows:

.

Equation (4) represents the first-stage regression equation, where \(\:\text{I}\text{V}\) is inverse temperature. Equation (5) represents the second-stage regression equation. \(\:{x}_{i}\) represents the control variables, including individual factors, vehicle factors, road factors and environment factors (as show in Table 2); \(\:\omega\:\:\)represents the fixed effects, including year, month, day of the week, city, etc. \(\:{\beta\:}_{1}\) is the coefficient of daily PM_2.5 concentrations.

Descriptive and correlation analysis

The final sample used in this study consists of 186,641 records of driver violations extracted from 118,240 road traffic accidents that occurred in Guangdong Province, China, between 2006 and 2014. Descriptive statistics are presented in Table 2. Among these drivers, 68.6% were aged between 26 and 45, 93.6% were male, and 19.2% came from rural areas. These drivers had diverse occupational backgrounds.

Furthermore, Pearson correlation coefficients were calculated for all variables in this study to gain a preliminary understanding of their relationships, as shown in Fig. 3. First, multicollinearity among explanatory variables was examined using Pearson correlation coefficients, following the general rule that variables with high correlations (r > 0.7) should be excluded from the model²⁵. In this study, all explanatory variables had correlation coefficients below 0.7, suggesting that multicollinearity is unlikely to be a concern. Second, the small Pearson correlation coefficients between Penalty points, Severity, and PM_2.5 do not necessarily imply that PM_2.5 has no significant effect on Penalty points and Severity, as Pearson correlation only captures linear relationships between two variables. Omitting key control variables could lead to estimation bias. Therefore, this study further employs an Ordered Logit Model for analysis.

Correlation Heatmap of Variables. The figure was created using the software Stata, version 18.

Full size image

Baseline result

Table 3 shows the impact of air pollution on traffic violations and accident outcomes. Columns (1)-(4) present the results of the ordered logit model, with columns (1)-(2) excluding control variables. The results indicate that there is no significant correlation between air pollution and traffic violations, but there is a significant positive correlation between air pollution and accident casualties. An increase of 1% in PM_2.5 concentration increases the probability of more severe traffic accidents by approximately 3.68%. Columns (3)-(4) present the results after adding control variables, showing that an increase in PM_2.5 concentration leads to an increase in the probability of more severe accidents but no significant effect on traffic violations.

To address potential endogeneity issues, columns (5)-(7) present the results after controlling for fixed effects and using instrumental variables. Column (5) presents the first-stage regression, i.e., the estimation of formula (4), which shows a significant positive correlation between inversions and PM_2.5 concentration. Columns (6)-(7) present the impact of air pollution on traffic violations and accident outcomes after excluding endogeneity, indicating that an increase in PM_2.5 concentration simultaneously leads to an increase in the probability of high-penalty violations and more severe accidents. An increase of 1% in PM_2.5 concentration leads to a 1.2661 times increase in the probability of high-penalty violations and a 1.5607 times increase in the probability of more severe accidents, with significance levels of 5% and 1% respectively. To more rigorously assess potential multicollinearity issues, we computed variance inflation factors (VIFs) for all independent and control variables. As demonstrated in Appendix Table A1, all VIF values fall well below the conventional threshold of 5, with the majority clustering around 1–3. This provides strong evidence that multicollinearity does not pose a substantive threat to our estimation results. Our findings align with Sager¹⁶, indicating that elevated PM_2.5 levels adversely affect traffic safety. However, this study does not consider another crucial aspect of road safety: the severity of traffic accidents. To our knowledge, the literature on the impact of air pollution on traffic accident severity is sparse.

Robustness analysis

Other air pollution

This paper investigates the impact of various air pollutants on traffic violations and accident consequences, providing insights beyond PM_2.5 and presenting results in Table 4 Columns (1) and (2) examine PM₁₀’s effect, revealing a 2.0304 times increase in high-penalty violations and a 3.8045 times increase in more severe accidents per 1% increase in concentration. Similarly, Columns (3) and (4) assess SO₂’s impact, showing a 1.0209 times increase in high-penalty violations and a 1.0398 times increase in more severe accidents. Columns (5) and (6) delve into SO₄’s influence, indicating a 1.2461 times increase in high-penalty violations and a 1.5145 times increase in more severe accidents with a 1% concentration rise. Despite differing coefficients compared to PM_2.5, all pollutants exhibit significant positive impacts on traffic violations and accident consequences, emphasizing the broader role of air pollution in shaping driving behavior and road safety.

Weather normalization

Across all spatial and temporal scales, weather influences concentrations of atmospheric pollutants and in turn ambient air quality²⁶. In order to exclude the influence of the correlation between weather and pollutants on the results, a random forest (RF)-based weather normalization approach was used to decouple the effects of meteorology from observed air pollutant concentrations in each city²⁷. This method accounted for meteorological variables such as temperature, relative humidity, wind speed, rainfall, atmospheric pressure, and solar radiation intensity, as well as temporal variables like Unix time, Gregorian date (day of the year), and day of the week. The estimation results using normalized PM_2.5 concentrations are presented in Table 5, aligning generally with the baseline results. They reveal that a 1% increase in PM_2.5 concentration corresponded to a 1.4435 times increase in the probability of high-penalty violations and a 1.9987 times increase in the probability of more severe accidents. All independent variables in this model also exhibit VIF values below 5 (see Appendix Table A2), confirming that multicollinearity is not a substantive concern in our estimation.

Heterogeneity analysis: driver characteristics

This paper investigates the impact of air pollution on traffic violations and accident consequences across different genders, age groups, and education levels. The following sections provide detailed analyses and results, highlighting how varying driver characteristics influence the relationship between PM_2.5 concentrations and driving behavior.

Firstly, this paper estimates the impact of air pollution on drivers of different genders, as shown in Table 6. The analysis reveals that air pollution affects male and female drivers differently regarding traffic violations and accident consequences. Columns (1) and (2) indicate that a 1% increase in PM_2.5 concentration raises the probability of high-penalty violations by male drivers by 1.2323 times and the probability of more severe accidents by 1.5276 times, which is slightly smaller than the total effect observed in the basic results. Columns (3) and (4) show that a 1% increase in PM_2.5 concentration increases the probability of high-penalty violations by female drivers by 1.9842 times but has no significant impact on accident consequences. Notably, while the coefficient for female drivers is larger, its significance is poorer, suggesting that the impact of air pollution is primarily on male drivers, with some female drivers possibly more affected.

Previous studies by Shr, et al.¹¹ and Sager¹⁶ do not find evidence of heterogeneous effects of air pollution based on gender. Our findings provide a different perspective, suggesting gender-specific sensitivities to air pollution. This discrepancy may be because women are generally more sensitive to environmental pollution²⁸ and have a higher level of risk perception regarding air pollution²⁹. Additionally, female drivers exhibit higher levels of impulse control, less speeding, and fewer risky driving behaviors³⁰ under high PM_2.5 levels. Male drivers, however, are more frequently associated with all risk factors for traffic accidents³¹, especially in dark conditions³². These findings align with our results, highlighting gender-specific impacts of air pollution.

Secondly, this paper also estimates the impact of air pollution on drivers of different age groups, as shown in Table 7. The results indicate that the effect of air pollution on driving behavior varies significantly with age. Columns (1) and (2) show that a 1% increase in PM_2.5 concentration raises the probability of high-penalty violations by young drivers (under 25 years old) by 1.7564 times but does not significantly affect accident consequences. Columns (3) and (4) reveal that for young adult drivers (25–44 years old), a 1% increase in PM_2.5 concentration raises the probability of high-penalty violations by 1.2121 times and more severe accidents by 1.5013 times. Columns (5) and (6) demonstrate that for middle-aged and elderly drivers (over 45 years old), a 1% increase in PM_2.5 concentration raises the probability of more severe accidents by 2.0908 times.

These age-related differences suggest that air pollution impacts driving behavior differently across age groups. Air pollution can evoke negative emotions such as anxiety and anger^33,34, and may be further associated with an increased risk of aggressive behavior³⁵. Young drivers, may be impacted by air pollution through this psychological influence. As drivers age, their experience and cognitive abilities improve, reducing the impact of air pollution on traffic violations. However, air pollution may impair reaction times, making it harder for elder drivers to respond promptly to sudden incidents, leading to more severe accidents. This heterogeneity is also explored in Shr, et al.¹¹, suggesting that the effect of air pollution on physiological functions may be relatively stronger among elderly drivers. The middle-aged drivers suffer both physiological and psychological harm, reflected in increased violations and accident severity.

Finally, this paper examines the impact of air pollution on drivers with different education levels, as shown in Table 8. The findings indicate that the impact of air pollution on traffic violations and accident severity varies with the driver’s education level. Columns (1) through (4) show that air pollution has no significant impact on the traffic violations or accident consequences of drivers with higher education or high school education. In contrast, Columns (5)-(6) reveal that for drivers with junior high school education or below, a 1% increase in PM_2.5 concentration leads to a 1.2530 times increase in the probability of high-penalty violations and a 1.5495 times increase in the probability of more severe accidents.

These results suggest that drivers with lower education levels bear a greater burden of the accident costs related to air pollution. Several factors could explain this. Firstly, individuals with higher education levels generally have a broader knowledge base and better comprehension skills, allowing them to accurately understand traffic warning signals. Secondly, they possess higher risk awareness and decision-making abilities, enabling them to assess the threat of haze on driving safety more accurately and take appropriate measures. Lastly, they place greater emphasis on health awareness and social responsibility, making them more proactive in protecting their own and others’ health³⁶. Factor, et al.³⁷ also link incorrect driving behavior to educational background, supporting our findings.

Heterogeneity analysis: economic and social development of City

This study utilizes data from 19 cities in Guangdong Province, China - one of the nation’s most economically dynamic regions with pronounced intra-provincial disparities. According to Xinhua News Agency (2020), China’s official state media, the Pearl River Delta (PRD) region’s nine cities contribute over 80% of the province’s GDP. As Fig. 1 illustrates, both PM_2.5 concentrations and traffic penalty points are significantly higher in PRD cities. To examine this spatial heterogeneity, we conduct separate regression analyses for PRD and non-PRD cities, as shown in Table 9. Columns (1)-(2) reveal that, in PRD cities, a 1% increase in PM_2.5 concentration leads to a 1.7624 times increase in the probability of high-penalty violations, but has no significant impact on the accident consequences. However, Columns (3)-(4) indicate that, in non-PRD cities, a 1% increase in PM_2.5 concentration leads to a 1.7624 times increase in the probability of more severe accidents.

Regional differences in the impact of PM_2.5 concentration on traffic violations and accident severity between the PRD and non-PRD cities may be attributed to factors such as traffic enforcement intensity, infrastructure quality, and driver behavior. In PRD cities, strict traffic enforcement increases detection of minor violations. Elevated PM_2.5 levels may impair cognitive functions, leading to more such violations. However, robust infrastructure and effective emergency responses may mitigate resulting accident severity³⁸. In non-PRD cities, less rigorous traffic enforcement and monitoring result in fewer recorded minor violations. Elevated PM_2.5 levels may impair cognitive functions, leading to more severe accidents due to delayed reactions and poor decision-making¹⁰. Additionally, less developed infrastructure and emergency response systems in these areas can exacerbate accident severity.

Potential visual mechanisms

This paper investigates the impact of air pollution on traffic violations and accident consequences under varying visibility, weather, lighting conditions, and for different types of drivers. The following sections provide detailed analyses and results, showing how these factors influence the relationship between PM_2.5 concentrations and driving behavior.

Firstly, this paper estimates the impact of air pollution on traffic violations and accident consequences under different visibility conditions, as shown in Table 10. The analysis reveals that air pollution has varying effects depending on visibility. Columns (1) and (2) present data for when visibility is between 50 and 100 m, showing no significant impact of air pollution on either traffic violations or accident consequences. However, Columns (3) through (6) demonstrate that when visibility is between 100 and 200 m and above 200 m, an increase in PM_2.5 concentration significantly raises the probability of high-penalty violations and more severe accidents. These findings indicate that PM_2.5 primarily impacts driving behavior when visibility is greater than 100 m.Our research confirms that air pollution significantly impacts traffic violations and accident consequences when visibility exceeds 100 m (p < 0.05). Poor visibility impairs drivers’ visual search, distance control, and motion perception, thereby increasing the likelihood of crashes and violations^14,15. These results suggest that PM_2.5 influences risky driving behaviors through visual channels.

Secondly, this paper estimates the impact of air pollution on traffic violations and accident consequences under different weather conditions, as shown in Table 11. The findings highlight that weather plays a crucial role in how air pollution affects driving behavior. Columns (1) and (2) show that on sunny days, an increase in PM_2.5 concentration significantly increases the probability of high-penalty violations and more severe accidents. In contrast, Columns (3) through (6) indicate that on cloudy and rainy days, air pollution impacts traffic violations only on cloudy days and accident consequences only on rainy days.These results suggest that the impact of air pollution on driving behavior is more pronounced on sunny days compared to cloudy or rainy days. One potential reason is that the reduction in visibility due to PM_2.5 is more noticeable during clear weather when visibility is naturally higher.

Thirdly, this paper examines the impact of air pollution on traffic violations and accident consequences under different lighting conditions, as shown in Table 12. The analysis reveals differences in how air pollution affects driving behavior during daytime and nighttime. Columns (1) and (2) show that during the day, a 1% increase in PM_2.5 concentration leads to a 1.2749 times increase in high-penalty violations and a 1.6464 times increase in more severe accidents. Columns (3) and (4) indicate that at night, a 1% increase in PM_2.5 concentration results in a 1.2730 times increase in high-penalty violations and a 1.5052 times increase in more severe accidents. The impact is slightly greater during the day, with stronger statistical significance.Our findings indicate that air pollution has a more significant impact on traffic violations and accident severity during the daytime compared to nighttime. This could be because the reduction in visibility due to PM_2.5 is more apparent during the day when natural light is abundant.

Finally, this paper estimates the impact of air pollution on traffic violations and accident consequences among motorcycle and vehicle drivers, as shown in Table 13. The results demonstrate that the type of vehicle significantly influences how air pollution affects driving behavior. Columns (1) and (2) reveal that for motorcycle drivers, a 1% increase in PM_2.5 concentration leads to a 2.3886 times increase in the probability of more severe accidents, with no significant impact on traffic violations. Columns (3) and (4) show that for car drivers, a 1% increase in PM_2.5 concentration results in a 1.3889 times increase in high-penalty violations and a 1.3446 times increase in more severe accidents. Our results suggest that air pollution significantly increases traffic violations among car drivers but not motorcycle drivers, while it exacerbates the severity of accidents more for motorcyclists. Possible reasons include greater exposure of two-wheel drivers to air pollution, leading to more physical and psychological discomfort^39,40. Gao, et al.⁴¹ support this by finding that early traffic warnings may better protect two-wheel drivers from the effects of smog. Shr, et al.¹¹ note that air pollution affects non-enclosed vehicle drivers slightly more negatively than enclosed vehicle drivers, suggesting visual rather than respiratory impacts. However, they overlook the higher severity of accidents among motorcyclists, indicating both physical and psychological impact mechanisms.

Potential psychological mechanisms

This paper explores the impact of air pollution on overconfidence, inexperienced behavior, and their accident consequences, drawing from previous studies by Chapman, et al.¹⁷. The analysis distinguishes between overconfidence, linked to deteriorated cognitive performance and lack of driving skills, and inexperienced behavior, resulting from deliberate risk-taking or breaking traffic rules^42,43. The definition of inexperienced and overconfident violations is showed in Table 14. The results in Table 15 elucidate these distinctions, showing that air pollution significantly impacts overconfidence but not inexperienced behavior.

The findings indicate that air pollution does not significantly affect inexperienced behavior or its accident consequences. Columns (1) and (2) in Table 15 show that increased PM_2.5 concentration does not correlate with higher instances of inexperienced driving behavior or related accidents. This suggests that other factors may play a more significant role in influencing inexperienced drivers’ actions and the outcomes of their driving.In contrast, air pollution does not correlate with higher instances of inexperienced driving behavior or related accidents. Columns (3) and (4) in Table 15 reveal that a 1% increase in PM_2.5 concentration leads to a 1.5107 times increase in the probability of high-penalty violations and a 1.7260 times increase in the probability of more severe accidents. This can be attributed to the adverse effects of air pollution on mental health, which can evoke negative emotions such as anxiety, restlessness, and anger^33,34,44. These emotions may lead to “road rage” and riskier driving behaviors, predominantly affecting overconfident drivers.

This study investigates the correlation between air pollution and road safety, utilizing data from diverse sources, including atmospheric climate data from NASA and the Road Traffic Accident Database of the Ministry of Public Security of China. Our findings suggest a positive and likely causal relationship between air pollution and traffic safety. Specifically, higher PM_2.5 concentrations are associated with an increased likelihood of high-penalty violations and more severe accidents. However, these effects vary significantly across driver characteristics and geographic regions. We find that male drivers and those with lower levels of education are more susceptible to the impact of PM_2.5, experiencing greater severity in both traffic violations and accidents. Additionally, the influence of PM_2.5 shifts with age, moving from an impact on traffic violations to a stronger effect on accident severity as drivers get older. The effects of PM_2.5 also differ across regions, with economically developed areas primarily seeing its influence on traffic violations, while less developed regions experience a greater impact on accident severity. Despite limitations in our data that prevent direct incorporation of drivers’ physiological or psychological factors, we provide indirect evidence to guide future research. Specifically, our results suggest that PM_2.5 has a more pronounced effect under conditions of clear visibility, such as during daylight and in good weather, when drivers might perceive the environment as safe. Furthermore, we find that PM_2.5 levels are associated primarily with increased severity in traffic violations related to overconfidence, rather than those linked to inexperience. This suggests that air pollution exacerbates risk-taking behavior, rather than directly impairing driving skills.

Our research enriches the understanding of the multifaceted impacts of air pollution on road safety and provides practical insights for transportation departments aiming to mitigate traffic accidents. However, our study has limitations. Data on traffic accidents are summarized daily, lacking hourly variations. This hinders the identification of differences between rush hours and non-rush hours. Neglecting to account for traffic volumes may also introduce endogeneity, potentially overestimating the effect of air pollution on accidents^45,46. Moreover, our focus on daily pollution variations limits our ability to draw conclusions about the long-term effects or the immediate versus cumulative impacts of pollution exposure. Future research should address these limitations and explore the long-term effects of residing permanently in polluted areas.