Abstract
Young drivers are among the most vulnerable groups in traffic accidents, showing a significant association with risky driving behaviors and increased susceptibility to injury and fatality. However, existing research largely focuses on developed countries and often neglects the distinct conditions in rapidly motorizing contexts like China, such as limited driving experience, poor infrastructure, and weaker traffic law enforcement. To address this gap, this study analyzes 78,668 administrative traffic accident records involving young drivers from China’s Ministry of Public Security between 2006 and 2014, accounting for temporal shifts in safety technology and infrastructure. The research investigates the risk factors linked to at fault and casualty severity among young drivers, compares behavioral risk patterns between young and non-young drivers, and explores how young drivers behave across different violation types. Results show that in China, young drivers are about twice as likely as older drivers to be at fault in accidents and have a higher probability of being involved in crashes with casualties, especially in severe cases. Their inexperience is associated with heightened risks under complex conditions like nighttime driving. Notably, an Expressway Paradox is identified, where expressways are associated with lower at fault risk but significantly higher casualty severity. Key violations—unlicensed driving, drunk driving, speeding, aggressive driving, and disobeying traffic signals—are significantly correlated with young drivers' elevated at fault rates. However, these illegal behaviors do not necessarily correspond to greater casualty severity. The study reveals that different violations show heterogeneous associations with accident outcomes and emphasizes the need to distinguish between at fault risk and casualty risk. These findings provide important insights for policymakers in developing countries to design targeted interventions, such as enhanced driver education, stricter licensing processes, and context-specific enforcement strategies, to improve traffic safety outcomes for young drivers.
Keywords: Young driver, At fault, Casualty severity, Risk factor, Logistic regression
Introduction
Road traffic accidents cause approximately 1.3 million deaths globally each year, with the World Health Organization identifying young people as the most vulnerable group among road users [1]. Regarding at fault, young drivers with limited experience are at a significantly higher risk of being involved in accidents [2, 3]. Research indicates that young drivers are disproportionately involved in accidents linked to risky behaviors, such as distracted driving, speeding, and fatigued driving—factors that increase their likelihood of engaging in dangerous driving practices [4–8]. However, recent studies emphasize that modern risk factors, particularly smartphone-based digital distractions and the proliferation of e-bikes in Asian contexts, have introduced new complexities to road safety [9–12]. From a casualty severity perspective, young people also experience some of the highest rates of injury and fatality in traffic accidents. Road traffic accidents are the leading cause of death for individuals aged 15 to 29 worldwide [13, 14]. In developed countries, such as the United States, teenage drivers exhibit significantly higher mortality rates compared to more experienced drivers. For example, in Florida, drivers under the age of 18 account for two-thirds of fatal accidents involving 15-year-olds [3]. Overall, traffic accidents remain the primary cause of death among American teenagers [15, 16]. Given these alarming statistics, it is both urgent and essential to investigate the risk factors associated with young drivers, particularly their engagement in illegal driving behaviors, to support efforts aimed at mitigating traffic-related fatalities in alignment with the United Nations’ 2030 Agenda for Sustainable Development.
It is noteworthy that 93% of traffic accident casualties occur in developing countries, with the vast majority of these deaths and serious injuries being preventable [1]. Despite some progress in improving road safety, traffic-related fatalities and injuries remain a significant public health and development challenge in many developing nations [13]. Compared to developed countries, developing nations are witnessing a dramatic surge in car ownership, accompanied by a sharp increase in the number of young drivers. However, inadequate road infrastructure, insufficient legal regulations, and outdated driver training systems are associated with a rising number of traffic accidents involving young drivers [17]. In China, for instance, the number of novice drivers continues to grow, which coincides with a corresponding increase in road traffic accidents and casualties among young drivers [18]. Drivers aged 16 to 25, with less than three years of driving experience, account for over 40% of total accidents [19].
While the global burden of road traffic injuries is concentrated in developing countries, empirical research leveraging large-scale administrative data in these contexts remains scarce. Recent investigations leveraging large-scale administrative datasets have begun to reshape our understanding of traffic enforcement and behavioral patterns in rapidly motorizing regions [12]. As the world’s largest developing country, China faces road safety challenges characteristic of many developing countries, such as rapid motorization and a surging population of young drivers. The complexity of China’s traffic environment provides a critical empirical basis for understanding risk patterns throughout developing countries. First, the data in this study are extracted from the official administrative Road Traffic Accident Database of the Ministry of Public Security of China [20]. Compared to the limited samples in existing literature, this study utilizes the most comprehensive official administrative source available. This national-level depth ensures robust findings and provides a reliable foundation for extrapolating insights to other developing countries [20]. Second, within this framework, Guangdong Province serves as a representative case study. Featuring a resident population exceeding 100 million and the nation’s largest migrant population, Guangdong’s traffic safety issues are typical of challenges found across China and other developing countries [18]. Analyzing 21 cities within Guangdong thus provides a generalizable basis for studying young drivers' risk factors.Finally, Guangdong’s socioeconomic characteristics as a reform "testing ground" mirror conditions in other developing countries experiencing rapid transitions. Thus, this study offers essential implications for formulating effective road safety policies throughout developing countries [21].
In this study, we analyze 78,668 traffic accident reports involving young drivers from the Ministry of Public Security of China’s database (2006–2014). The research examines risk factors associated with at fault and casualty severity, considering personal, vehicular, road, and environmental characteristics to explore the behavioral patterns linked to risky driving behaviors. This study contributes to the literature in several ways. First, addressing the scarcity of research in developing countries, we investigate how factors such as higher legal driving age, limited experience, and underdeveloped infrastructure influence accident risk. Second, we bridge a critical gap by establishing a dual-risk framework that distinguishes between the likelihood of at fault and the severity of casualties. This framework addresses the limitations of traditional models that often conflate behavioral risks with physical injury outcomes [12], allowing for the identification of counter-intuitive scenarios—such as the 'Expressway Paradox'—where certain conditions show associations with lower at fault rates but significantly higher casualty severity. This methodological distinction is consistent with modern empirical findings that emphasize the heterogeneity of risk factors across different accident stages. Third, we conduct a comparative analysis between young and non-young drivers to identify distinct risk patterns. Finally, we evaluate the role of specific illegal behaviors—such as unlicensed driving and speeding—and how they show associations with both the probability of at fault and resulting casualty severity.
The rest of this article is organized as follows. In Sect. "Literature review", the literature on the traffic safety of young drivers is reviewed, and improvement methods are proposed to address the existing problems. Sect. "Data and models" describes the data and presents descriptive statistics. The results of empirical analysis are presented in Sect. "Results". Sect. "Discussion" discusses the risk factors for young drivers in accidents involving different violations of the law.
Literature review
Young drivers are among the most vulnerable groups of road users, accounting for a disproportionately high share of traffic accidents [2]. A synthetic evaluation of existing literature suggests that the risks associated with this group should be analyzed through a dual-risk framework: the risk of at fault and the risk of sustaining a casualty severity. On the one hand, limited driving experience, high risk propensity, and overconfidence contribute to more hazardous driving behaviors and traffic violations [14]. On the other hand, factors such as vehicle safety conditions and heightened susceptibility to adverse environmental factors result in higher mortality and casualty rates in accidents [12, 22, 23]. Distinguishing between these two dimensions is crucial because the mechanisms underlying traffic violations often differ from those determining casualty severity. While many studies identify young drivers as a high-risk group, they frequently overlook that certain illegal behaviors or personal characteristics do not necessarily correspond to a higher likelihood of serious or fatal outcomes [18]. For instance, while inexperience increases the probability of collision [24], the severity of the injury is more closely linked to physical factors like seat belt use and impact speed [22]. Therefore, a dual-risk approach is essential to avoid a partial assessment of the overall risk profile and to support the formulation of targeted safety policies in developing countries [18, 21].
The risk of at fault among young drivers is primarily associated with their limited experience and inherent risk-taking tendencies. Research indicates that young drivers face a higher accident risk in the initial period after obtaining their licenses [16], as their lack of experience often leads to more frequent hazardous driving behaviors [14]. However, Chapman et al. [25] observed a sharp peak in crash rates immediately following unsupervised licensure, suggesting a "licensing spike" that highlights the critical transition to independent driving. This phenomenon is often linked to the “Experience Paradox”, where initial skill acquisition leads to inflated overconfidence. As novice drivers master basic vehicle control, their subjective risk perception may decrease faster than their objective skill improves, leading to miscalibration and increased risk-taking [25]. Beyond experience, bad driving habits, such as the use of electronic devices while driving, can significantly increase young drivers’ risk of at fault [9, 12, 26, 27]. These behaviors are often influenced by social and contextual factors, including parental influence and peer pressure [2], which shape the risk attitudes of young drivers. For instance, young drivers are more likely to have traffic violations when driving their own vehicles than when driving others’ vehicles, possibly because parents who allow their children to drive family vehicles have a higher risk preference [2]. External environments also serve as critical inducements for at fault in developing countries. While adverse conditions like larger road gradients increase driving difficulty [26], drier roads may paradoxically increase the risk of at fault for young drivers due to relaxed vigilance while driving [2]. Furthermore, the timing of driving plays a decisive role,the accident risk is higher on weekends and at night [2, 26, 28], which can be explained by the fact that these periods represent “recreational driving” that poses a higher risk for teenagers [29]. Collectively, these personal and environmental factors constitute the primary drivers of the at fault risk profile in developing countries [18].
While at fault risk is primarily driven by behavioral choices, casualty severity in accidents involving young drivers is largely determined by the physical safety environment and the force of impact. In terms of vehicular factors, young drivers face a higher risk of serious casualties due to factors such as bad vehicle safety status, inadequate safety features [2], older vehicle age, and a lack of insurance [30]. Research shows significant differences in casualty severity among different types and features of vehicles [2, 31, 32]. For instance, young drivers have a higher probability of serious casualties when driving small vehicles compared to large passenger or cargo vehicles [30]. Furthermore, the lack of safety habits, such as not wearing seat belts, is a critical factor leading to higher casualty risk among young drivers [3, 33]. Road characteristics and environmental conditions also play a decisive role in the physical outcomes of a crash. Young drivers have a higher probability of accidents and casualties on rural roads than on urban roads, which can be attributed to factors such as gravel roads, higher speeds, and intersections lacking traffic signals [26]. Additionally, road infrastructure such as steeper road slopes increases the potential for severe outcomes [26]. Environmental factors further exacerbate these physical risks; for example, insufficient lighting at night and poor visibility significantly increase the accident risk for young drivers [28]. In developing countries, these risks are often compounded by substandard vehicle conditions and underdeveloped transportation infrastructure, which heighten the susceptibility of young drivers to fatal injuries when a collision occurs [22, 23]. The situation is further complicated by the proliferation of electric bikes and diverse micro-mobility modes [12].
A critical dimension of young driver risk is the intersection between illegal behaviors and accident outcomes, although the relationship between the two requires further empirical clarification. Research indicates that illegal behaviors such as drunk driving, speeding, aggressive driving, and distracted driving show a significant association with both an increased probability of collision and casualty severity [24]. For example, studies in New Zealand found that factors such as drinking and the lack of a seat belt are among the strongest predictors of serious casualties, with drinking increasing the likelihood of fatal injuries by 85.9% [34]. Similarly, speeding is consistently identified as a major factor contributing to both the occurrence of accidents and the severity of casualties among young drivers [12, 24]. However, the correlation between different traffic violations and their corresponding risks is complex and sometimes non-linear. While illegal behaviors generally heighten risk, they do not always correspond to greater casualty severity in a uniform manner. For instance, certain high-risk situations, such as late-night driving or driving under the influence, are often accompanied by a lower likelihood of seat belt use, which further compounds the risk of injury [35, 36]. Conversely, recent evidence reveals a more structured heterogeneity. For example, the “Expressway Paradox” demonstrates that while expressways may show associations with a lower risk of at fault due to a more controlled environment, they are correlated with significantly higher casualty severity when accidents occur [12, 37]. These findings underscore the importance of distinguishing between the factors that associated with triggering at fault and those correlated with the casualty severity of the resulting outcome within developing countries.
To evaluate the risk factors of traffic accidents involving young drivers, scholars have employed a variety of methodologies, ranging from early questionnaire-based surveys [14, 17] to scenario simulation experiments [2, 9]. While these methods, alongside techniques like structural equation modeling [17] and Poisson regression [2, 38], have provided foundational insights, they often face issues regarding sample representativeness and the generalizability of results. More recently, the Naturalistic Teenage Driving Study (NTDS) has emerged as a high-fidelity tool for recording field driving behaviors [16]. However, due to the inherent limitations of small sample sizes and sampling bias in surveys and NTDS, it remains difficult to generalize research findings to broader geographical areas. To address these limitations, recent research has increasingly utilized large-scale traffic accident datasets from specific regions or nations [12, 30, 32, 34]. While such efforts have resolved the sample size issue to some extent, existing studies remain heavily concentrated on developed nations. Consequently, it remains to be verified whether findings from these contexts are applicable to the unique socioeconomic and infrastructural realities of rapidly motorizing contexts like China [12]. Given the distinctive challenges of rapid motorization and evolving road environments, there is a critical need to leverage comprehensive, national-level administrative data within a developing country context. Such an approach not only ensures statistical robustness but also provides the empirical rigor necessary to formulate evidence-based policies tailored to the majority of developing countries.
Data and models
Data
The data in this study consist of 78,668 traffic accident reports involving young drivers across 21 cities in Guangdong Province from 2006 to 2014, extracted from the official administrative Road Traffic Accident Database of the Ministry of Public Security of China [21]. Guangdong serves as a critical case study for several reasons: it consistently reports among the highest frequencies of road traffic accidents in China, maintains a permanent resident population exceeding 100 million, and hosts the nation’s largest migrant population. These demographic and socioeconomic complexities make Guangdong’s traffic safety challenges particularly prominent and characteristic of the broader issues faced during rapid motorization [18].While the dataset is concentrated within a single province, the findings offer essential reference value and insights into the risk factors associated with young drivers in China [20]. Compared with the limited samples often collected through localized observation points or questionnaires, this study utilizes official administrative data with an extensive temporal range and a rich set of accident-related indicators, allowing for controls over rapid technological and infrastructural shifts during the study period. Consequently, this research provides one of the most comprehensive analyses of administrative traffic records currently available in the Chinese context. The analytical framework developed here provides a theoretically grounded basis for evaluating traffic safety policies in other developing countries facing comparable developmental trajectories [21].
Previous studies have shown that the risk factors affecting the risk of being at fault and casualty severity of young people differ from those of non-young people. Therefore, when analyzing the risk factors for traffic accidents involving young drivers, both the the risk of being at fault and casualty severity are analyzed. The dependent variables are set as follows: first, “whether the driver was at fault in the accident”, with 1 indicating that the young driver was found to be entirely or primarily responsible for the accident (at fault) and 0 indicating secondary or no responsibility1; and second, “whether the accident resulted in serious casualties”, with 1 indicating serious consequences such as “death or serious injury” and 0 indicating other outcomes such as “minor injury or no injury”2 [39]. The analysis of casualty severity is conditioned upon the occurrence of a traffic violation where the driver was found to be at fault [21]. This approach allows the study to measure “severity given a violation”, focusing on how specific behavioral and environmental factors exacerbate the physical outcomes of at-fault accidents, rather than assessing general casualty severity across the entire population [18]. This distinction is critical for developing targeted interventions for high-risk driving behaviors.
To ensure the statistical robustness and clarity of the analysis, we excluded accident reports with ambiguous liability determinations or incomplete critical information. Specifically, reports where the primary cause of the accident was not clearly identified or where essential covariates—such as driver demographics, road conditions, or vehicle characteristics—were missing were removed from the final sample. This filtering process resulted in a final analytical sample of 78,668 reports. By focusing on cases with complete and deterministic information, we mitigate potential biases arising from measurement error or missing data, thereby providing a more reliable empirical basis for identifying the risk factors associated with at fault and casualty severity. However, a limitation of this study is the lack of exposure data (e.g., total vehicle miles traveled), which precludes the calculation of true crash rates per mile. Thus, our results represent associations within the subset of occurred accidents rather than population-level risks.
Risk factors
To study the risk factors affecting young drivers’ risk of at fault and casualty, this study investigates the influences of drivers’ personal characteristics, vehicle, road, and environmental factors [18].
Personal factors: The driver’s gender, age, and driving experience are considered potential risk factors for traffic accidents and casualties among young drivers [3, 33, 40]. Therefore, according to the influence of driving habits and driving skills on traffic safety by different age groups and driving experiences and considering the legal adult age in China, drivers are divided by age into three groups, ≤ 18, 18–24, and ≥ 24, and by driving experience into three groups, ≤ 2 years, 2–4 years, and ≥ 4 years. In addition, education, income, and social status influence the formation of young people’s driving behavior [41, 42]. Although these indicators cannot be directly examined via the data used in this study, they can be indirectly studied through the analysis of household registration and occupation as proxy variables [18]. Household registration status is divided into rural and urban types. Occupation is divided into farmers, students, office staff, workers, business owners, and others.
Vehicle: The safety status, safety features, and insurance of the vehicle are important risk factors affecting at fault and accident casualties among young people [2, 30]. Hence, we set dummy variables according to whether the vehicle safety status is normal, whether it is overloaded, whether it has compulsory third-party liability insurance, and whether it is in operation [18]. The risk differences between different vehicle types are also considered. Vehicles are classified into cars, trucks, and motorcycles.
Road types: Different types of roads also affect the risk of traffic accidents for young drivers [26]. Therefore, the functional classification of roads considered in this study includes motor vehicle lanes, mixed lanes, and other lanes. Moreover, according to road types and classes, roads are classified into express, ordinary roads, and urban roads. Ordinary roads include first-class roads, second-class roads and below, and urban roads include urban expressways, urban ordinary highway, and other urban roads [21].
Environmental factors: Driving time period and lighting conditions have a great impact on the accident risk of young drivers [2, 26, 28]. Therefore, this study considers the influence of environmental factors, such as lighting, weather, weekends, seasons, and years. Lighting conditions include daytime, nighttime with streetlights, and nighttime without streetlights, with weather conditions set as a dummy variable. Weekends are defined as Saturday and Sunday. The seasons are classified according to the Meteorological Administration as follows: spring (March to May), summer (June to September), autumn (October to November), and winter (December to the next February). The years included are 2006 to 2014 [39].
Logistic regression model
To identify the risk factors associated with at fault and casualty severity among young drivers in China, we employ a logistic regression framework. This approach is well-suited for modeling the binary outcomes within our dual-risk framework: the probability of at fault and the probability of a serious casualty outcome [18]. The basic structure of the logistic regression model is defined as:
 |
where 
represents the probability of the event (e.g., committing at fault or resulting in a serious casualty) for accident 
. 
denotes the 
-th explanatory variable—including personal, vehicular, road, and environmental characteristics—and 
represents the estimated coefficients. The results are reported as Odds Ratios (OR), calculated as 
which indicate the statistical association between a unit change in the predictor and the odds of the outcome.
To account for potential clustering and higher-level structures, we included year and region (prefecture-level cities) as categorical dummy variables in the regression. This fixed-effects approach effectively captures time-varying trends (e.g., changes in national safety policies) and spatial heterogeneity (e.g., differences in regional economic development or infrastructure levels), thereby reducing omitted variable bias. Within the administrative database provided by the Ministry of Public Security, each record represents a unique, discrete accident event. Since the accidents are stochastically distributed across the road network, the assumption of independence between observations is maintained after controlling for the aforementioned regional and temporal fixed effects. Given our primary research goal is to identify common risk patterns among young drivers across China to inform national-level policy, the pooled model provides robust, population-averaged estimates that are directly interpretable for broad safety interventions.
Descriptive statistical analysis
Table 1 presents the relevant variables and their statistical distributions. Within the sample, 59.0% of all drivers were associated with at fault, a figure that rises to 67.1% among the younger cohort. Regarding outcomes, 22.0% of accidents resulted in serious casualties, with 22.1% of young drivers involved in such consequences.
Table 1.
Descriptive statistics
| Not young (≥ 25) | Young (< 25) | Total | |
|---|---|---|---|
| N = 31,378 | N = 47,290 | N = 78,668 | |
| At fault | |||
| Entire/Primary | 14,688 (46.8%) | 31,738 (67.1%) | 46,426 (59.0%) |
| Secondary/No | 16,690 (53.2%) | 15,912 (32.9%) | 32,242 (41.0%) |
| Severity | |||
| Killed and serious injured | 6844 (21.8%) | 10,458 (22.1%) | 17,302 (22.0%) |
| Minor and no injured | 24,534 (78.2%) | 36,832 (77.9%) | 61,366(78.0%) |
| Human factor | |||
| Gender | |||
| Male | 29,008 (92.4%) | 44,063 (93.2%) | 73,071 (92.9%) |
| Female | 2370 (7.6%) | 3227 (6.8%) | 5597 (7.1%) |
| Driving without license | |||
| Never | 2806 (8.9%) | 14,450 (30.6%) | 17,256 (21.9%) |
| Once | 80 (0.3%) | 32 (0.1%) | 112 (0.1%) |
| Mismatching | 132 (0.4%) | 374 (0.8%) | 506 (0.6%) |
| Holds a license | 28,360 (90.4%) | 32,434 (68.5%) | 60,794 (77.3%) |
| Driving experience | |||
| Driving without license | 3018 (9.6%) | 14,856 (31.4%) | 17,874 (22.7%) |
| 0–2 | 4009 (12.8%) | 18,832 (39.8%) | 22,841 (29.0%) |
| 3–4 | 3447 (11.0%) | 9009 (19.1%) | 12,456 (15.8%) |
| > 4 | 20,904 (66.6%) | 4593 (9.7%) | 25,497 (32.4%) |
| Household registration | |||
| Urban | 24,249 (77.3%) | 35,887 (75.9%) | 60,136 (76.4%) |
| Rural | 7129 (22.7%) | 11,403 (24.1%) | 18,532 (23.6%) |
| Occupation | |||
| Farmer | 4359 (13.9%) | 6680 (14.1%) | 11,039 (14.0%) |
| Student | 17 (0.1%) | 1412 (3.0%) | 1429 (1.8%) |
| Staff | 2010 (6.4%) | 2350 (5.0%) | 4360 (5.5%) |
| Worker | 11,812 (37.6%) | 19,764 (41.8%) | 31,576 (40.1%) |
| Boss | 4004 (12.8%) | 4075 (8.6%) | 8079 (10.3%) |
| Other | 9176 (29.2%) | 13,009 (27.5%) | 22,185 (28.2%) |
| Vehicle factor | |||
| Vehicle safety status | |||
| Fine | 29,970 (95.5%) | 44,957 (95.1%) | 75,197 (95.2%) |
| Bad | 1408 (4.5%) | 2333 (4.9%) | 3741 (4.8%) |
| Insurance | |||
| Valid | 25,805 (82.2%) | 35,380 (74.8%) | 61,185 (77.8%) |
| Invalid | 5573 (17.8%) | 11,910 (25.2%) | 17,483 (22.2%) |
| Ownership of the vehicle | |||
| Commercial | 7028 (22.4%) | 5913 (12.5%) | 12,941 (16.5%) |
| Non-commercial | 24,350 (77.6%) | 41,377 (87.5%) | 65,727 (83.5%) |
| Vehicle overload condition | |||
| Overload | 1550 (4.9%) | 3362 (7.1%) | 4912 (6.2%) |
| Non-overload | 29,828 (95.1%) | 43,928 (92.9%) | 73,756 (93.8%) |
| Type of Vehicle | |||
| Truck | 7668 (24.4%) | 8042 (17.0%) | 15,710 (20.0%) |
| Passenger car | 12,538 (40.0%) | 12,012 (25.4%) | 24,550 (31.2%) |
| Motorcycle | 11,172 (35.6%) | 27,236 (57.6%) | 38,408 (48.8%) |
| Road factor | |||
| Type of traffic lane | |||
| Isolated lane | 13,525 (43.1%) | 19,241 (40.7%) | 32,766 (41.7%) |
| Without isolated lane | 17,853 (56.9%) | 28,049 (59.3%) | 45,902 (58.3%) |
| Type of road | |||
| Express | 806 (2.6%) | 832 (1.8%) | 1638 (2.1%) |
| First-class highway | 4545 (14.5%) | 6299 (13.3%) | 10,844 (13.8%) |
| Second-class or below highway | 11,101 (35.4%) | 16,481 (34.9%) | 27,582 (35.1%) |
| Urban expressway | 1598 (5.1%) | 2315 (4.9%) | 3913 (5.0%) |
| Urban ordinary highway | 9694 (30.9%) | 15,242 (32.2%) | 24,936 (31.7%) |
| Urban other highway | 3634 (11.6%) | 6121 (12.9%) | 9755 (12.4%) |
| Type of traffic control device | |||
| No device | 6399 (20.4%) | 10,609 (22.4%) | 17,008 (21.6%) |
| Traffic light | 2845 (9.1%) | 3503 (7.4%) | 6348 (8.1%) |
| Warning sign | 19,116 (60.9%) | 27,949 (59.1%) | 47,065 (59.8%) |
| Traffic line | 22,645 (72.2%) | 32,978 (69.7%) | 55,623 (70.7%) |
| Other device | 4331 (13.8%) | 6761 (14.3%) | 11,092 (14.1%) |
| Environment factor | |||
| Lighting condition | |||
| Daylight | 16,911 (53.9%) | 23,615 (49.9%) | 40,526 (51.5%) |
| Dark but lighted | 11,207 (35.7%) | 17,791 (37.6%) | 28,998 (36.9%) |
| Dark | 3260 (10.4%) | 5884 (12.4%) | 9144 (11.6%) |
| Day of week | |||
| Weekdays | 22,531 (71.8%) | 33,860 (71.6%) | 56,391 (71.7%) |
| Weekends | 8847 (28.2%) | 13,430 (28.4%) | 22,277 (28.3%) |
| Weather condition | |||
| Fine | 23,913 (76.2%) | 35,774 (75.6%) | 59,687 (75.9%) |
| Bad | 7465 (23.8%) | 11,516 (24.4%) | 18,981 (24.1%) |
| Season | |||
| Spring | 7756 (24.7%) | 11,747 (24.8%) | 19,503 (24.8%) |
| Summer | 10,991 (35.0%) | 16,439 (34.8%) | 27,430 (34.9%) |
| Fall | 5249 (16.7%) | 8065 (17.1%) | 13,314 (16.9%) |
| Winter | 7382 (23.5%) | 11,039 (23.3%) | 18,421 (23.4%) |
| Year | |||
| 2006 | 4641 (14.8%) | 6898 (14.6%) | 11,539 (14.7%) |
| 2007 | 4070 (13.0%) | 6118 (12.9%) | 10,188 (13.0%) |
| 2008 | 3433 (10.9%) | 5148 (10.9%) | 8581 (10.9%) |
| 2009 | 3066 (9.8%) | 4792 (10.1%) | 7858 (10.0%) |
| 2010 | 2982 (9.5%) | 4522 (9.6%) | 7504 (9.5%) |
| 2011 | 2815 (9.0%) | 4277 (9.0%) | 7092 (9.0%) |
| 2012 | 2798 (8.9%) | 4118 (8.7%) | 6916 (8.8%) |
| 2013 | 3448 (11.0%) | 5357 (11.3%) | 8805 (11.2%) |
| 2014 | 4125 (13.1%) | 6060 (12.8%) | 10,185 (12.9%) |
| Traffic violations | |||
| No illegal act | 1170 (3.7%) | 1836 (3.9%) | 3006 (3.8%) |
| Drunk driving | 569 (1.8%) | 1083 (2.3%) | 1652 (2.1%) |
| Overspeed | 283 (0.9%) | 825 (1.7%) | 1108 (1.4%) |
| Aggressive driving | 9787 (31.2%) | 12,649 (26.7%) | 22,436 (28.5%) |
| Violation of traffic signals | 1267 (4.0%) | 1396 (3.0%) | 2663 (3.4%) |
| Other illegal act | 18,288 (58.3%) | 29,515 (62.4%) | 47,803 (60.8%) |
(1) "At fault" is an administrative measure of liability. Traffic violations are not mutually exclusive, as one driver may commit multiple acts. Severe violations (e.g., drunk driving, speeding) are ofen assigned Entire/Primary responsibility, whereas minor infractions often result in Secondary/No responsibility. (2) Driving experience is calculated based on the duration since the initial issuance of a valid driver's license. "Driving without license" is treated as a distinct category in the analysis, as these drivers without Driving experience. (3)Household Registration refers to the Hukou system in China, where "Rural" and "Urban" categories serve as proxies for a driver’s socio-economic background and access to developmental resources. (4) Age is bifurcated at 25 years to isolate high-risk youth cohorts as defined by World Health Organization (WHO) standards. (5)Vehicle safety status is a categorical variable indicating whether the vehicle’s mechanical condition was fine or bad at the time of the accident, as determined by post-crash police assessments
Personal factors
Males dominated the sample (92.9%). A significant disparity exists in licensing: 30.6% of young drivers had never obtained a license, compared to only 8.9% of non-young drivers. Experience also varied significantly; 39.8% of young drivers had 0–2 years of experience, whereas 66.6% of non-young drivers had over four years. Occupational data shows that workers (40.1%) and farmers (14.0%) were the most prevalent groups, with students (3.0%) notably more common among young drivers.
Vehicular factors
While 77.8% of vehicles were insured, the rate was lower for young drivers (74.8%) than for non-young drivers (82.2%). Commercial vehicle use was less frequent among the younger group (12.5% vs. 22.4%). Notably, 57.6% of young drivers operated motorcycles—substantially higher than the 35.6% for non-young drivers. Overloading was also more frequent among young drivers (7.1% vs. 4.9%).
Road and environmental factors
Accidents involving young drivers were most frequent on second-class or lower highways (35.1%) and ordinary urban roads (31.7%), with only 1.8% occurring on expressways. Roughly 21.6% of these incidents occurred at locations lacking indicator signals. Temporally, 51.5% of accidents happened during daylight, 28.3% occurred on weekends, and the summer season saw the highest frequency (34.9%).
Illegal behaviors
Specific risk behaviors showed distinct patterns. Driving under the influence (2.3%) and speeding (1.7%) were more prevalent among young drivers than their older counterparts. Conversely, non-young drivers exhibited higher rates of aggressive driving (31.2% vs. 26.7%) and signal disobedience (4.0% vs. 3.0%).
The association of each potential variable with young drivers’ at-fault risk and casualty severity was initially examined using the Pearson chi-square test (Table 2). The results indicated that the majority of variables exhibited a strong correlation with the two dependent variables (p < 0.001). Regarding at-fault risk, factors such as vehicle ownership, day of the week, and season did not pass the chi-square test. For casualty severity, variables including the absence of traffic control devices, traffic lines, other traffic facilities, and weather conditions were found to be non-significant. However, chi-square tests are limited to preliminary univariate exploration and may either overemphasize statistical significance in large samples or mask true associations by failing to account for confounding effects. Therefore, regardless of their univariate significance, all variables were retained in the subsequent multivariate analysis using a full-model approach based on their theoretical importance in road safety literature rather than a step-wise selection process [21]. By employing binary logistic regression models, we were able to control for confounding variables and reveal suppressed effects, thereby robustly quantifying the independent association of each factor with the outcomes. This methodological transition explains why certain variables (e.g., valid insurance or weekends) shifted from non-significant in the chi-square test to significant in the regression models.
Table 2.
Chi-square test of Independence for accident incidence and severity for young drivers
| Variable | At fault (N = 78,668) | Severity (N = 78,668) |
|---|---|---|
| χ2 (P-value), degree of freedom | χ2 (P-value), degree of freedom | |
| Human factor | ||
| Gender | 86.64*** (< 0.001), 1 | 223.00***(< 0.001), 1 |
| Driving without license | 308.27*** (< 0.001), 3 | 452.18*** (< 0.001), 3 |
| Driving experience | 1542.65*** (< 0.001), 3 | 531.92*** (< 0.001), 3 |
| Household registration | 164.90*** (< 0.001), 1 | 935.82*** (< 0.001), 1 |
| Occupation | 56.96***(< 0.001), 5 | 639.25***(< 0.001), 5 |
| Vehicle factor | ||
| Vehicle safety status | 45.14*** (< 0.001), 1 | 277.97*** (< 0.001), 1 |
| Insurance | 47.11*** (< 0.001), 1 | 397.39*** (< 0.001), 1 |
| Ownership of the vehicle | 0.13 (0.7227), 1 | 874.70*** (< 0.001), 1 |
| Vehicle overload condition | 20.97*** (< 0.001), 1 | 419.43*** (< 0.001), 1 |
| Type of Vehicle | 628.15*** (< 0.001), 2 | 773.62*** (< 0.001), 2 |
| Road factor | ||
| Type of traffic lane | 98.20*** (< 0.001), 1 | 28.46*** (< 0.001), 1 |
| Type of road | 160.29*** (< 0.001), 5 | 1565.51*** (< 0.001), 5 |
| Type of traffic control device: No device | 74.38*** (< 0.001), 1 | 1.46 (0.2277), 1 |
| Type of traffic control device: Traffic light | 103.53*** (< 0.001), 1 | 13.48*** (< 0.001), 1 |
| Type of traffic control device: Warning sign | 52.81*** (< 0.001), 1 | 3.98* (0.0460), 1 |
| Type of traffic control device: Traffic line | 76.19*** (< 0.001), 1 | 2.15 (0.1430), 1 |
| Type of traffic control device: Other device | 4.70* (0.0302), 1 | 0.04 (0.8327), 1 |
| Environmental factor | ||
| Lighting condition | 105.36* (0.048), 2 | 1141.39*** (< 0.001), 2 |
| Weather condition | 30.705*** (< 0.001), 1 | 2.83 (0.0925), 1 |
| Day of week | 0.12 (0.7303), 1 | 14.09*** (< 0.001), 1 |
| Season | 2.59 (0.4596), 3 | 34.52*** (< 0.001), 3 |
| Year | 32.94*** (< 0.001), 8 | 543.92*** (< 0.001), 8 |
| Traffic violations | - | 335.32*** (< 0.001), 5 |
(1) Since a single accident site may feature a combination of multiple types of traffic control devices, each device category was treated as an independent binary variable. (2) *p < 005, **p < 0.01, ***p < 0.001
Results
Analysis of the at fault risk factors: traffic accidents involving young drivers
The estimation results in Column 1 of Table 3 identify several risk factors associated with the probability of at fault among young drivers. Variance Inflation Factor (VIF) tests confirmed that all variables exhibited values well below 10 (Mean VIF = 1.6), indicating that multicollinearity was not a significant issue in the model specification. Regarding personal characteristics, age and licensing status emerge as the most substantial predictors of at fault risk. Drivers under the age of 18 (OR = 2.157) and those aged 18–24 (OR = 2.224) exhibit more than double the odds of committing at fault compared to older groups, reflecting the high risk propensity and overconfidence prevalent among youth [14]. The large effect size of the 18–24 age group (a 122.4% increase in odds) suggests that this cohort is the most critical target for behavioral interventions. Similarly, unlicensed driving (OR = 1.641) shows a strong positive association, with a 64.1% increase in the odds of at fault relative to licensed drivers. In contrast, licensed driving experience demonstrates a significant protective effect; compared to novices, those with over 4 years of experience (OR = 0.764) show a 23.6% reduction in the odds of at fault [16]. Male drivers also show a 17.6% higher likelihood of at fault than female drivers (OR = 1.176; [3, 33]).
Table 3.
Logistic Model for Traffic Accidents Involving Young Drivers
| (1) | (2) | |
|---|---|---|
| At fault | Severity | |
| Human Factor | ||
| Gender(base: Female) | ||
| Male | 1.176*** | 1.409*** |
| [1.110,1.246] | [1.263,1.571] | |
| Age(base: ≥ 25) | ||
| < 18 | 2.157*** | 1.032 |
| [2.008,2.318] | [0.931,1.143] | |
| 18–24 | 2.224*** | 1.079* |
| [2.141,2.310] | [1.013,1.148] | |
| Driving without license(base: Correct license) | ||
| Never | 1.641*** | 1.737*** |
| [1.557,1.729] | [1.604,1.880] | |
| Once | 0.951 | 1.395 |
| [0.645,1.402] | [0.712,2.736] | |
| Mismatching | 1.166 | 1.950*** |
| [0.970,1.403] | [1.488,2.555] | |
| Driving experience(base: 0–2) | ||
| 3–4 | 0.909*** | 1.030 |
| [0.867,0.953] | [0.958,1.107] | |
| > 4 | 0.764*** | 1.038 |
| [0.730,0.800] | [0.964,1.118] | |
| Household registration(base: Urban) | ||
| Rural | 0.753*** | 1.301*** |
| [0.725,0.783] | [1.230,1.377] | |
| Occupation(base: Farmer) | ||
| Student | 0.988 | 1.191* |
| [0.872,1.120] | [1.018,1.394] | |
| Staff | 0.997 | 0.718*** |
| [0.922,1.078] | [0.634,0.812] | |
| Worker | 1.003 | 0.740*** |
| [0.954,1.054] | [0.688,0.796] | |
| Boss | 0.985 | 0.788*** |
| [0.924,1.051] | [0.715,0.868] | |
| Other | 0.961 | 0.844*** |
| [0.912,1.013] | [0.781,0.911] | |
| Vehicle Factor | ||
| Vehicle safety status(base: Good) | ||
| Bad | 1.289*** | 1.402*** |
| [1.199,1.385] | [1.277,1.539] | |
| Insurance(base: Invalid) | ||
| Valid | 1.162*** | 0.774*** |
| [1.116,1.211] | [0.727,0.824] | |
| Ownership of the vehicle(base: Non-commercial) | ||
| Commercial vehicle | 0.877*** | 1.523*** |
| [0.831,0.926] | [1.405,1.650] | |
| Vehicle overload condition(base: Non-overload) | ||
| Overload | 0.865*** | 1.382*** |
| [0.812,0.920] | [1.266,1.508] | |
| Type of Vehicle(base: Passenger car) | ||
| Truck | 0.983 | 1.191*** |
| [0.932,1.036] | [1.100,1.289] | |
| Motorcycle | 0.419*** | 0.854*** |
| [0.402,0.437] | [0.798,0.914] | |
| Road factor | ||
| Type of traffic lane(base: Without isolated lane) | ||
| Isolated lane | 0.937*** | 1.091** |
| [0.905,0.969] | [1.034,1.150] | |
| Type of road(base: First-class highway) | ||
| Express | 0.532*** | 2.045*** |
| [0.475,0.594] | [1.732,2.414] | |
| Second-class or below highway | 1.108*** | 1.207*** |
| [1.054,1.165] | [1.118,1.303] | |
| Urban expressway | 0.980 | 1.168** |
| [0.907,1.059] | [1.038,1.313] | |
| Urban ordinary highway | 1.128*** | 0.751*** |
| [1.073,1.186] | [0.693,0.814] | |
| Urban other highway | 1.092** | 0.965 |
| [1.026,1.162] | [0.876,1.063] | |
| Type of traffic control device(base: Invalid) | ||
| Traffic light | 0.773*** | 0.898* |
| [0.731,0.817] | [0.815,0.989] | |
| Warning sign | 0.998 | 1.143*** |
| [0.955,1.044] | [1.067,1.225] | |
| Traffic line | 0.914*** | 1.036 |
| [0.871,0.959] | [0.963,1.114] | |
| Other device | 1.023 | 1.049 |
| [0.979,1.068] | [0.982,1.122] | |
| Environment factor | ||
| Lighting condition(base: Daytime) | ||
| Dark but lighted | 1.078*** | 1.357*** |
| [1.043,1.114] | [1.287,1.431] | |
| Dark | 1.248*** | 1.750*** |
| [1.187,1.312] | [1.635,1.873] | |
| Weather condition (base: Fine) | ||
| Bad | 1.006 | 0.978 |
| [0.971,1.042] | [0.926,1.033] | |
| Day of week(base: Weekday) | ||
| Weekends | 0.999 | 1.039 |
| [0.967,1.033] | [0.988,1.093] | |
| Season(base: Fall) | ||
| Spring | 0.969 | 1.032 |
| [0.924,1.016] | [0.960,1.110] | |
| Summer | 0.964* | 0.979 |
| [0.922,1.007] | [0.914,1.047] | |
| Winter | 0.976 | 1.105** |
| [0.931,1.023] | [1.028,1.188] | |
| Year(base: 2006) | ||
| 2007 | 0.973 | 0.956 |
| [0.920,1.030] | [0.880,1.039] | |
| 2008 | 1.029 | 0.850*** |
| [0.970,1.092] | [0.778,0.928] | |
| 2009 | 1.028 | 0.657*** |
| [0.967,1.092] | [0.599,0.722] | |
| 2010 | 1.007 | 0.716*** |
| [0.946,1.071] | [0.652,0.787] | |
| 2011 | 1.007 | 0.762*** |
| [0.945,1.073] | [0.692,0.838] | |
| 2012 | 0.970 | 0.739*** |
| [0.910,1.034] | [0.670,0.814] | |
| 2013 | 0.999 | 0.603*** |
| [0.941,1.061] | [0.549,0.662] | |
| 2014 | 0.988 | 0.510*** |
| [0.933,1.047] | [0.465,0.560] | |
| Traffic violations (base: Legal act) | ||
| Drunk driving | 2.371*** | |
| [1.978,2.840] | ||
| Overspeed | 2.017*** | |
| [1.669,2.438] | ||
| Aggressive driving | 1.093 | |
| [0.960,1.244] | ||
| Violation of traffic signals | 1.340** | |
| [1.114,1.612] | ||
| Other illegal act | 1.080 | |
| [0.953,1.224] | ||
| pseudo R2 | 0.059 | 0.063 |
| r2_cu | 0.104 | 0.099 |
| ll | −50,082.251 | −22,853.462 |
| chi2 | 6321.060 | 3086.277 |
| aic | 100,256.501 | 45,808.924 |
| bic | 100,683.059 | 46,254.951 |
| pcorrectly | 63.997 | 78.213 |
| ROC | 0.661 | 0.676 |
| N | 78,668 | 46,426 |
(1) The sample size (N) in Severity model represents the subset of accidents where the driver was determined to be at fault (Entire/Primary responsibility) [18]. The model estimates the odds of serious casualties contingent upon an at-fault violation occurring [21]. (2) Exponentiated coefficients; 95% confidence intervals in brackets; *p < 0.05, **p < 0.01, ***p < 0.001
Vehicular characteristics also delineate distinct risk profiles. Substandard vehicle conditions are positively associated with at-fault risk, showing a 28.9% increase in odds (OR = 1.289; [2]). Conversely, certain factors are associated with a notable decrease in risk. Motorcycle use (OR = 0.419) exhibits one of the strongest negative associations, with a 58.1% reduction in the odds of at fault compared to other vehicle types. Notably, overloading (OR = 0.865) is associated with a 13.5% decrease in the risk of at fault, suggesting more cautious behavioral choices among young drivers under high-load conditions in China. Furthermore, while vehicle ownership was statistically insignificant in the univariate Chi-square test, it emerged as a significant protective factor in the logistic model (OR = 0.877). This reversal suggests a suppressor effect, where the impact of vehicle ownership on at-fault risk—potentially reflecting higher levels of vehicle care or driver responsibility—only becomes apparent after controlling for confounding variables such as vehicle type and driver occupation. Regarding insurance, the presence of a valid policy is associated with a 16.2% increase in at-fault risk (OR = 1.162). This finding aligns with the economic theory of "Moral Hazard," suggesting that insurance coverage may inadvertently reduce the perceived financial consequences of a crash, thereby diminishing the incentive for cautious driving and leading to increased risk-taking behavior [43, 44].
Road and infrastructure factors show significant statistical relationships with at fault probability, often with substantial effect sizes. Expressways (OR = 0.532) are linked to a 46.8% reduction in the odds of at fault compared to other road types. In contrast, urban ordinary highways (OR = 1.128) and second-class or lower highways (OR = 1.108) are associated with higher at fault rates [26]. The presence of traffic signals (OR = 0.773) shows a 22.7% reduction in the odds of at fault, underscoring its importance as a regulatory tool. Finally, environmental conditions remain critical: poor lighting (OR = 1.248) is associated with a 24.8% increase in the probability of at fault, highlighting the risks inherent in low-visibility environments [2, 28]. Regarding weather condition, while it showed high significance in the univariate Chi-square test, its effect became non-significant in the pooled logistic model. This suggests that the impact of weather on at-fault risk may be absorbed by other correlated environmental and infrastructure variables, such as lighting and road surface conditions, or may be subject to regional variations across the diverse geographical landscape of Guangdong.
Analysis of risk factors for casualties: Traffic accidents involving young drivers
The estimation results in Column 2 of Table 3 identify the risk factors associated with the probability of serious casualties in accidents involving young drivers. Variance Inflation Factor (VIF) tests confirmed that all variables exhibited values well below 10 (Mean VIF = 1.8), indicating that multicollinearity was not a significant issue in the model specification Regarding personal characteristics, licensing status and gender show substantial effect sizes. Unlicensed driving (OR = 1.737) and driving with a mismatched license type (OR = 1.950) are linked to a 73.7% and 95.0% increase in the odds of serious casualties, respectively, representing some of the strongest predictors of fatal or severe outcomes [3]. Male drivers are associated with a 40.9% higher likelihood of serious casualties than females (OR = 1.409), echoing gender-based risk patterns in casualty risk [33]. Younger drivers aged 18–24 (OR = 1.079) also show a positive association with increased severity compared to other cohorts. Furthermore, rural household registration (OR = 1.301) is linked to a 30.1% increase in the odds of serious casualties. Occupations also show distinct correlations: students (OR = 1.191) are associated with a higher risk, while office staff (OR = 0.718), workers (OR = 0.740), and self-employed individuals (OR = 0.788) are linked to a reduction in the probability of serious casualties.
Vehicular characteristics demonstrate varying degrees of association with casualty severity. Substandard vehicle conditions (OR = 1.402) are positively correlated with a 40.2% increase in the odds of serious casualties, reinforcing the critical role of vehicle maintenance in safety [2, 23]. Notably, having insurance (OR = 0.774) is associated with a 22.6% reduction in the risk of serious outcomes. In terms of vehicle type and load, commercial vehicles (OR = 1.523) and overloaded trucks (OR = 1.382) are linked to a higher probability of serious casualties. Conversely, motorcycle use (OR = 0.854) is associated with a 14.6% lower risk of serious casualties in this specific sample. This "Motorcycle Paradox" suggests that while motorcycles lack structural protection, their significantly lower mass results in less kinetic energy during impacts compared to heavier passenger car [45], thereby reducing the overall casualty severity in collisions where the motorcycle driver is at fault. This potentially reflects the biological vulnerability differences observed in diverse traffic modes [22].
Road and environmental factors contribute significantly to the casualty profile. Expressways (OR = 2.045) show a major effect size, associated with more than double the odds of serious casualties compared to other roads, while urban ordinary highways (OR = 0.751) show a 24.9% reduction in risk. Infrastructure elements such as isolated lanes (OR = 1.091) and warning signs (OR = 1.143) are linked to a higher probability of casualties, whereas the presence of traffic signals (OR = 0.898) is associated with a 10.2% decrease in the risk of serious outcomes. Environmental visibility remains a dominant factor; dark conditions (OR = 1.750) and dark-but-lighted conditions (OR = 1.357) are positively associated with a 75.0% and 35.7% increase in the odds of serious casualties, highlighting the extreme risk posed by low-visibility environments [26, 28].
Finally, specific illegal behaviors show the most dramatic associations with casualty severity. Driving under the influence (OR = 2.371) and speeding (OR = 2.017) are associated with more than a 100% increase in the odds of serious casualties [24], while disobeying traffic signals (OR = 1.340) is linked to a 34.0% increase in risk. These findings underscore that while multiple factors are at play, behavioral violations and visibility constraints are the primary correlates of fatal accidents in China.
Risk factors for at fault: a comparative analysis of young and non-young drivers
Columns 1 and 2 of Table 4 further examine the factors associated with at fault among young and non-young drivers within the same crash contexts in China. This comparative approach highlights both universal risk factors and age-specific heterogeneities within the dual-risk framework. The results indicate that while certain factors show consistent statistical associations across both groups, the magnitude of these associations (effect sizes) often differs, providing nuanced insights for targeted interventions.
Table 4.
Logistic Model for Traffic Accident Risk Between Young Drivers and Non-young Drivers
| At fault | Severity | |||
|---|---|---|---|---|
| (1) | (2) | (3) | (4) | |
| Young | No Young | Young | No Young | |
| Human Factor | ||||
| Gender(base: Female) | ||||
| Male | 1.224*** | 1.126** | 1.407*** | 1.404** |
| [1.135,1.321] | [1.031,1.231] | [1.236,1.602] | [1.146,1.719] | |
| Age(18–24) | ||||
| < 18 | 1.009 | - | 0.955 | - |
| [0.942,1.081] | [0.871,1.047] | |||
| Driving without license(base: Correct license) | ||||
| Never | 1.573*** | 1.819*** | 1.825*** | 1.530*** |
| [1.480,1.672] | [1.633,2.026] | [1.671,1.993] | [1.261,1.856] | |
| Once | 1.122 | 0.919 | 1.922 | 1.101 |
| [0.541,2.325] | [0.576,1.467] | [0.694,5.324] | [0.440,2.753] | |
| Mismatching | 1.116 | 1.281 | 2.083*** | 1.438 |
| [0.899,1.384] | [0.895,1.834] | [1.549,2.802] | [0.732,2.824] | |
| Driving experience(base: 0–2) | ||||
| 3–4 | 0.899*** | 0.927 | 1.003 | 1.147 |
| [0.851,0.950] | [0.844,1.018] | [0.926,1.087] | [0.964,1.364] | |
| > 4 | 0.833*** | 0.740*** | 1.014 | 1.101 |
| [0.776,0.895] | [0.689,0.794] | [0.916,1.123] | [0.963,1.259] | |
| Household registration(base: Urban) | ||||
| Rural | 0.768*** | 0.728*** | 1.249*** | 1.443*** |
| [0.731,0.808] | [0.686,0.773] | [1.168,1.335] | [1.299,1.604] | |
| Occupation(base: Farmer) | ||||
| Student | 1.051 | 1.722 | 1.166 | 1.642 |
| [0.924,1.195] | [0.625,4.743] | [0.992,1.370] | [0.444,6.072] | |
| Staff | 1.112 | 0.867* | 0.714*** | 0.754** |
| [0.999,1.238] | [0.773,0.972] | [0.613,0.831] | [0.610,0.934] | |
| Worker | 1.156*** | 0.814*** | 0.709*** | 0.826** |
| [1.083,1.234] | [0.752,0.881] | [0.650,0.774] | [0.720,0.948] | |
| Boss | 0.978 | 0.954 | 0.769*** | 0.854 |
| [0.895,1.069] | [0.868,1.049] | [0.681,0.869] | [0.725,1.005] | |
| Other | 1.040 | 0.856*** | 0.823*** | 0.900 |
| [0.972,1.114] | [0.789,0.928] | [0.751,0.902] | [0.780,1.037] | |
| Vehicle Factor | ||||
| Vehicle safety status(base: Good) | ||||
| Bad | 1.342*** | 1.208*** | 1.353*** | 1.506*** |
| [1.219,1.477] | [1.081,1.351] | [1.212,1.512] | [1.266,1.791] | |
| Insurance(base: Invalid) | ||||
| Valid | 1.070** | 1.357*** | 0.765*** | 0.802*** |
| [1.016,1.126] | [1.268,1.452] | [0.712,0.821] | [0.704,0.913] | |
| Ownership of the vehicle(base: Non-commercial) | ||||
| Commercial vehicle | 0.920 | 0.874*** | 1.453*** | 1.644*** |
| [0.843,1.004] | [0.815,0.937] | [1.301,1.622] | [1.460,1.852] | |
| Vehicle overload condition(base: Non-overload) | ||||
| Overload | 0.822*** | 0.948 | 1.398*** | 1.323** |
| [0.762,0.887] | [0.850,1.056] | [1.264,1.545] | [1.106,1.584] | |
| Type of Vehicle(base: Passenger car) | ||||
| Truck | 0.881** | 1.057 | 1.134* | 1.284*** |
| [0.810,0.957] | [0.986,1.133] | [1.019,1.263] | [1.139,1.448] | |
| Motorcycle | 0.397*** | 0.442*** | 0.771*** | 1.073 |
| [0.375,0.420] | [0.415,0.469] | [0.712,0.836] | [0.947,1.217] | |
| Road factor | ||||
| Type of traffic lane(base: Without isolated lane) | ||||
| Isolated lane | 0.948* | 0.922** | 1.092** | 1.077 |
| [0.906,0.992] | [0.875,0.971] | [1.024,1.164] | [0.978,1.186] | |
| Type of road(base: First-class highway) | ||||
| Express | 0.654*** | 0.421*** | 2.054*** | 2.027*** |
| [0.559,0.765] | [0.356,0.498] | [1.681,2.508] | [1.504,2.732] | |
| Second-class or below highway | 1.160*** | 1.049 | 1.142** | 1.362*** |
| [1.085,1.240] | [0.973,1.132] | [1.041,1.253] | [1.187,1.563] | |
| Urban expressway | 0.978 | 0.986 | 1.071 | 1.400** |
| [0.882,1.084] | [0.876,1.109] | [0.929,1.236] | [1.136,1.724] | |
| Urban ordinary highway | 1.155*** | 1.094* | 0.733*** | 0.779*** |
| [1.081,1.235] | [1.013,1.181] | [0.666,0.807] | [0.672,0.903] | |
| Urban other highway | 1.154*** | 1.008 | 0.925 | 1.058 |
| [1.063,1.252] | [0.915,1.109] | [0.824,1.038] | [0.884,1.266] | |
| Type of traffic control device(base: Invalid) | ||||
| Traffic light | 0.842*** | 0.695*** | 0.853** | 0.986 |
| [0.781,0.908] | [0.640,0.755] | [0.758,0.961] | [0.832,1.169] | |
| Warning sign | 1.029 | 0.960 | 1.195*** | 1.051 |
| [0.970,1.091] | [0.897,1.027] | [1.100,1.299] | [0.929,1.188] | |
| Traffic line | 0.845*** | 1.021 | 1.034 | 1.020 |
| [0.793,0.900] | [0.949,1.099] | [0.947,1.128] | [0.893,1.165] | |
| Other device | 1.066* | 0.962 | 1.058 | 1.016 |
| [1.007,1.130] | [0.900,1.029] | [0.978,1.145] | [0.897,1.151] | |
| Environment factor | ||||
| Lighting condition(base: Daytime) | ||||
| Dark but lighted | 1.103*** | 1.048 | 1.373*** | 1.318*** |
| [1.056,1.153] | [0.996,1.103] | [1.288,1.464] | [1.196,1.451] | |
| Dark | 1.411*** | 1.040 | 1.739*** | 1.794*** |
| [1.320,1.507] | [0.960,1.126] | [1.604,1.884] | [1.577,2.042] | |
| Weather condition (base: Fine) | ||||
| Bad | 1.047 | 0.954 | 1.022 | 0.879* |
| [0.999,1.097] | [0.904,1.008] | [0.958,1.090] | [0.793,0.974] | |
| Day of week(base: Weekday) | ||||
| Weekends | 1.016 | 0.979 | 1.039 | 1.033 |
| [0.973,1.062] | [0.931,1.030] | [0.978,1.103] | [0.941,1.133] | |
| Season(base: Fall) | ||||
| Spring | 0.972 | 0.967 | 1.043 | 0.997 |
| [0.913,1.034] | [0.899,1.040] | [0.956,1.137] | [0.872,1.141] | |
| Summer | 0.969 | 0.955 | 0.960 | 1.016 |
| [0.914,1.027] | [0.893,1.022] | [0.885,1.041] | [0.897,1.151] | |
| Winter | 0.972 | 0.981 | 1.109** | 1.090 |
| [0.913,1.034] | [0.912,1.055] | [1.017,1.209] | [0.954,1.244] | |
| Year(base: 2006) | ||||
| 2007 | 1.007 | 0.934 | 0.959 | 0.953 |
| [0.935,1.085] | [0.856,1.019] | [0.867,1.061] | [0.820,1.108] | |
| 2008 | 1.027 | 1.035 | 0.867** | 0.801** |
| [0.949,1.110] | [0.945,1.133] | [0.779,0.964] | [0.683,0.939] | |
| 2009 | 1.042 | 1.012 | 0.677*** | 0.603*** |
| [0.961,1.129] | [0.921,1.112] | [0.606,0.757] | [0.508,0.717] | |
| 2010 | 1.005 | 1.012 | 0.739*** | 0.654*** |
| [0.926,1.091] | [0.920,1.113] | [0.660,0.827] | [0.551,0.778] | |
| 2011 | 1.009 | 1.009 | 0.753*** | 0.767** |
| [0.928,1.098] | [0.916,1.112] | [0.672,0.845] | [0.646,0.910] | |
| 2012 | 0.979 | 0.968 | 0.761*** | 0.675*** |
| [0.899,1.066] | [0.878,1.067] | [0.676,0.855] | [0.565,0.805] | |
| 2013 | 0.953 | 1.074 | 0.619*** | 0.552*** |
| [0.880,1.032] | [0.980,1.178] | [0.553,0.692] | [0.466,0.654] | |
| 2014 | 0.958 | 1.031 | 0.530*** | 0.451*** |
| [0.887,1.035] | [0.944,1.125] | [0.474,0.592] | [0.380,0.534] | |
| Traffic violations (base: Legal act) | ||||
| Drunk driving | 2.480*** | 1.955*** | ||
| [2.015,3.051] | [1.334,2.865] | |||
| Overspeed | 2.033*** | 1.875** | ||
| [1.642,2.516] | [1.216,2.892] | |||
| Aggressive driving | 1.075 | 1.089 | ||
| [0.921,1.255] | [0.857,1.384] | |||
| Violation of traffic signals | 1.489*** | 1.144 | ||
| [1.183,1.874] | [0.831,1.574] | |||
| Other illegal act | 1.079 | 1.061 | ||
| [0.932,1.251] | [0.837,1.346] | |||
| pseudo R2 | 0.031 | 0.035 | 0.063 | 0.066 |
| r2_cu | 0.053 | 0.064 | 0.100 | 0.102 |
| ll | −29,035.113 | −20,921.015 | −15,916.287 | −6893.760 |
| chi2 | 1834.054 | 1529.294 | 2155.721 | 979.700 |
| aic | 58,160.226 | 41,930.030 | 31,932.574 | 13,885.520 |
| bic | 58,554.608 | 42,297.600 | 32,350.838 | 14,257.665 |
| pcorrectly | 67.253 | 59.554 | 77.387 | 79.902 |
| ROC | 0.618 | 0.627 | 0.676 | 0.679 |
| N | 47,290 | 31,378 | 31,738 | 14,688 |
(1) The sample size (N) in Severity model represents the subset of accidents where the driver was determined to be at fault (Entire/Primary responsibility) [18]. The model estimates the odds of serious casualties contingent upon an at-fault violation occurring [21]. (2) Exponentiated coefficients; 95% confidence intervals in brackets; *p < 0.05, **p < 0.01, ***p < 0.001
Several factors exhibit a consistent positive association with at fault probability for both cohorts. Gender remains a prominent correlate, with male drivers exhibiting higher odds of at fault than females among both young (OR = 1.224, a 22.4% increase in odds) and non-young (OR = 1.126, a 12.6% increase) groups [3, 33]. Unlicensed driving shows a substantial effect size for both, particularly for non-young drivers (OR = 1.819) compared to their younger counterparts (OR = 1.573). Substandard vehicle conditions are also linked to higher at fault likelihood (young: OR = 1.342,non-young: OR = 1.208). Interestingly, holding insurance is positively correlated with at fault risk for both groups (young: OR = 1.070; non-young: OR = 1.357), suggesting a potential moral hazard effect in China. Conversely, several factors are consistently associated with a reduced risk of at fault. Long-term licensed driving experience (over four years) is linked to a 16.7% and 26.0% reduction in the odds of at fault for young (OR = 0.833) and non-young (OR = 0.740) drivers, respectively [16]. Rural household registration also shows a protective effect (young: OR = 0.768,non-young: OR = 0.728). Regarding transportation modes, motorcycle use exhibits a strong negative association with at fault risk, with young drivers (OR = 0.397) showing a 60.3% reduction in odds. Infrastructure also provides significant regulatory benefits; the presence of traffic signals is associated with a 15.8% reduction in odds for young drivers (OR = 0.842) and an even larger 30.5% reduction for non-young drivers (OR = 0.695).
The analysis reveals critical differences in how certain factors correlate with at fault risk between the two groups. In China, young drivers with 2–4 years of experience (OR = 0.899) are less likely to at fault than novices, whereas this association is not statistically significant for the non-young group. Occupational differences are also stark: being workers (OR = 1.156) is associated with higher odds of at fault only for young drivers, while the opposite or non-significant associations are observed for non-young drivers. Vehicular and environmental influences also show age-dependent patterns. For young drivers, overloading (OR = 0.822) and driving trucks (OR = 0.881) are linked to a reduction in at fault risk, associations not significantly observed among non-young drivers. This may suggest more cautious behavioral adjustments by younger individuals under specific heavy-load conditions [2]. Furthermore, environmental constraints such as dark environments (OR = 1.103) and lower-class highways (OR = 1.160) are associated with increased at fault risk specifically for the younger cohort [26]. Finally, traffic markings show a significant negative association with at fault probability only among young drivers (OR = 0.845), suggesting that younger road users in China may be more responsive to visual regulatory cues on the road.
Risk factors for casualties: a comparative analysis of young and non-young drivers
Columns 3 and 4 of Table 4 present a detailed analysis of factors associated with the risk of serious casualties in accidents involving young drivers, distinguishing between vehicles operated by young and non-young cohorts. In the context of China, the magnitude of these statistical associations reveals how various risk factors correlate with the physical outcomes of crashes across different age groups.
Several factors show a consistent positive association with higher casualty severity for both groups. Male drivers exhibit elevated risk among both young (OR = 1.407, a 40.7% increase in odds) and non-young (OR = 1.404) cohorts [33]. Unlicensed driving remains a critical correlate, showing a substantial effect size for both young (OR = 1.825) and non-young (OR = 1.530) drivers [3]. Rural household registration, while linked to fewer at fault, is associated with a 24.9% and 44.3% increase in the odds of severe casualties for young (OR = 1.249 and non-young (OR = 1.443 drivers, respectively. Substandard vehicle conditions also show a strong positive correlation with severity (young: OR = 1.353,non-young: OR = 1.506; [2]). In contrast, having insurance is associated with a reduced risk of serious casualties for both young (OR = 0.765) and non-young (OR = 0.802) drivers. Regarding vehicle types and loads, commercial vehicle use (young: OR = 1.453,non-young: OR = 1.644) and overloading (young: OR = 1.398; non-young: OR = 1.323) are both linked to higher odds of severe outcomes. Infrastructure factors also exhibit significant associations: expressways show the largest effect size, being associated with more than double the odds of serious casualties for both groups (OR = 2.054 and 2.027). Environmental visibility is equally critical, as dark conditions are linked to a roughly 74% to 79% increase in the probability of severe casualties [26, 28].
The analysis reveals significant age-dependent differences in how specific factors and behaviors correlate with casualty severity in China. Driving under the influence and speeding exhibit more pronounced effect sizes for young drivers; the positive association for drunk driving (OR = 2.480) and speeding (OR = 2.033) is substantially higher among the younger cohort than the non-young group (OR = 1.955 and 1.875, respectively), echoing findings on the heightened vulnerability of youth to high-risk behaviors [24]. Furthermore, certain factors appear uniquely predictive of severity for young drivers. Operating a vehicle not permitted by one's license class (OR = 2.083, a 108.3% increase in odds) is associated with higher casualty risk only for the younger group. Interestingly, motorcycle use (OR = 0.771) is linked to a 22.9% reduction in the risk of serious casualties for young drivers in this sample, a correlation not observed for non-young riders [22]. Regarding infrastructure, traffic signals significantly correlate with a 14.7% decrease in the risk of serious casualties (OR = 0.853) exclusively for young drivers, while disobeying traffic signs (OR = 1.489) is associated with higher odds only for this cohort. Finally, adverse weather conditions correlate with a lower probability of severity (OR = 0.879) only for non-young drivers, suggesting potential differences in adaptive driving behavior across generations in China.
Discussion
Comparative analysis of at fault risk and casualty risk
The synthesis of empirical results suggests that personal characteristics and the "experience paradox" are associated with the unique risk profile of young drivers in China. Consistent with existing literature, being male, aged 18–24, and unlicensed are the most prominent correlates of both higher at-fault risk and casualty severity [3, 31–33]. However, a "risk-outcome paradox" is identified regarding driving experience and age. While drivers under 18 show significantly higher odds of at-fault risk, they do not do not exhibit a statistically significant association with higher casualty severity compared to older cohorts, potentially due to higher biological resilience or different crash configurations [16]. Notably, in the Chinese context, licensed driving experience is associated with a protective effect against at-fault risk for young drivers, yet it lacks a linear relationship with reduced casualty severity. This suggests that as experience increases, so does the likelihood of secondary risks such as distraction or overconfidence during high-speed maneuvers [16]. From the perspective of Risk Homeostasis Theory, experienced drivers may compensate for their increased skill by adopting riskier behaviors, thereby neutralizing the safety benefits of their experience [46]. Furthermore, the higher casualty risk associated with rural household registration—despite lower at-fault rates—points toward potential systemic disparities in post-crash emergency response and infrastructure safety in rural China.
Vehicular and modal disparities further distinguish the behavioral and physical risks within the dual-risk framework. The analysis identifies vehicle safety and mode as central to the casualty profile. Substandard vehicle conditions are a universal risk factor, correlating with both higher at fault and severity odds [2, 31]. A unique finding in this study, contrary to some Western literature, is that motorcycle use among young drivers in this sample is associated with a lower probability of both at fault and serious casualties compared to heavy-load commercial vehicles. This "Motorcycle Paradox" may reflect the specific regional characteristics of Guangdong, where motorcycles are often utilized for low-speed urban commutes,the lower mass of these vehicles results in significantly less kinetic energy during impacts compared to heavier vehicle types. Conversely, commercial vehicle operation and overloading show a distinct "protective-yet-hazardous" pattern: young drivers appear more cautious (lower at fault odds) when operating these vehicles, yet the physical outcomes are substantially more severe once a crash occurs. This suggests that the heavy mass of these vehicles in China's traffic mix overrides the benefits of cautious behavior in determining injury outcomes.
The role of infrastructure and environment in China exhibits a significant trade-off between behavioral regulation and physical safety. Expressways and roads with isolated lanes show a negative association with at-fault rates due to restricted access and clearer markings, yet they are associated with significantly higher casualty severity, reflecting the increased kinetic energy associated with higher impact speeds. Conversely, urban ordinary highways exhibit a positive association with at-fault risk but a negative association with serious casualty outcomes. Traffic signals emerge as a primary regulatory tool, showing a significant negative association with both at-fault and casualty risks across age groups. Temporally, the heightened severity of accidents in winter and the increased at fault risk during nighttime with poor lighting underscore the persistent challenge of visibility and adverse road conditions for young drivers [26, 28].
Behavioral criticality reinforces that drunk driving and speeding remain the most lethal correlates in China, especially for the younger cohort. The magnitude of these associations is notably stronger for young drivers than for non-young drivers [24]. This cohort-specific vulnerability emphasizes that the same illegal behavior is associated with disproportionately worse outcomes for youth, necessitating targeted policy interventions that account for both behavioral tendencies and physical outcomes.
As visually synthesized in Fig. 1, a defining characteristic of young drivers' risk profile in China is the directional divergence of Odds Ratios (ORs) across the two risk dimensions, revealing a clear "decoupling" between behavioral culpability and physical consequences. Specifically, a critical cluster of variables—including Rural household registration, Commercial vehicle operation, Vehicle overload, Isolated traffic lanes, and Expressways—exhibits a "low at-fault risk, high severity" pattern (Model 1 OR < 1.0; Model 2 OR > 1.0). For expressways and isolated lanes, the negative association with at-fault risk suggests that restricted access and standardized environments effectively regulate driver behavior; however, the high ORs for casualty severity underscore the "lethality of speed," where increased kinetic energy at high velocity overrides the benefits of cautious behavior. This "risk-outcome paradox" [16] is further evidenced by commercial vehicle operation, where the heavy physical mass in China's traffic mix determines injury outcomes regardless of the driver's behavioral caution. Conversely, Valid insurance and Urban ordinary highways display a contrasting "high at-fault risk, low severity" pattern (Model 1 OR > 1.0,Model 2 OR < 1.0). This suggests that while insurance may induce a "moral hazard" leading to riskier maneuvers [46], and urban roads present more complex conflict points, the lower operating speeds in these contexts significantly mitigate physical impact. These divergent patterns reinforce that the same variables can have disproportionately different associations with at-fault risk and casualty severity [24], necessitating a dual-risk perspective for targeted policy interventions.
Fig. 1.
Comparison of Odds Ratios for At-fault Risk and Casualty Severity across Key Correlates. Note: This forest plot provides a visual synthesis of the "Dual-Risk" framework by comparing the coefficients of key variables from Table 3. It specifically highlights variables exhibiting divergent directional associations across the two risk dimensions: those with lower at-fault risk but higher casualty severity (e.g., Expressways, Commercial vehicles), and those with higher at-fault risk but lower casualty severity (e.g., Valid insurance, Urban ordinary highways). The error bars represent 95% confidence intervals
Finally, it is essential to acknowledge several inherent limitations regarding causal inference in this study. While the observed associations between driver characteristics, behaviors, and accident outcomes are statistically robust, they should not be interpreted as definitive causal links [47]. Due to the reliance on administrative police records, our analysis lacks granular exposure data, such as annual vehicle miles traveled (VMT), which is a critical determinant of accident frequency [48]. Furthermore, the absence of direct psychological data regarding drivers' behavioral intent means that the mechanisms proposed in this discussion—while aligned with established safety theories—remain hypothesized based on observed statistical patterns [49]. Future research incorporating naturalistic driving studies or survey-based exposure metrics is required to further validate these potential causal pathways.
Traffic violations by young people: driving without a license, drinking, speeding, aggressive driving, and disobeying traffic signals
This study further investigates the risk profiles of young drivers in China by examining five primary illegal behaviors: unlicensed driving, drunk driving, speeding, aggressive driving, and disobeying traffic signals. The results, presented in Table 5, reveal substantial variations in the statistical associations between these behaviors and the dual dimensions of violation risk and casualty severity.
Table 5.
Risk Analysis of Young Drivers in Various Illegal Accidents
| At fault | Severity | |
|---|---|---|
| Panel A: Driving Without license | ||
| Age (base: ≥ 25) | ||
| < 18 | 3.337*** | 1.215 |
| [2.765,4.028] | [0.993,1.487] | |
| 18–24 | 3.283*** | 1.224* |
| [2.815,3.828] | [1.016,1.474] | |
| Control | Yes | Yes |
| pcorrectly | 87.245 | 70.548 |
| ROC | 0.894 | 0.621 |
| N | 10,114 | 6794 |
| Panel B: Drunk driving | ||
| Age (base: ≥ 25) | ||
| < 18 | 2.053* | 1.160 |
| [1.053,4.003] | [0.531,2.535] | |
| 18–24 | 3.548*** | 1.384 |
| [2.673,4.709] | [0.917,2.089] | |
| Control | Yes | Yes |
| pcorrectly | 73.802 | 68.019 |
| ROC | 0.773 | 0.686 |
| N | 1649 | 1035 |
| Panel C: Overspeed | ||
| Age (base: ≥ 25) | ||
| < 18 | 7.857*** | 1.542 |
| [3.578,17.254] | [0.613,3.882] | |
| 18–24 | 11.082*** | 1.278 |
| [6.957,17.653] | [0.769,2.124] | |
| Control | Yes | Yes |
| pcorrectly | 82.821 | 66.864 |
| ROC | 0.841 | 0.704 |
| N | 1106 | 845 |
| Panel D: Aggressive driving | ||
| Age (base: ≥ 25) | ||
| < 18 | 1.050 | 0.858 |
| [0.904,1.219] | [0.655,1.124] | |
| 18–24 | 1.132*** | 0.989 |
| [1.054,1.215] | [0.883,1.107] | |
| Control | Yes | Yes |
| pcorrectly | 65.078 | 79.905 |
| ROC | 0.685 | 0.667 |
| N | 22,436 | 12,575 |
| Panel E: Violation of traffic signals | ||
| Age (base: ≥ 25) | ||
| < 18 | 1.388 | 1.265 |
| [0.882,2.184] | [0.608,2.633] | |
| 18–24 | 1.211 | 1.414 |
| [0.997,1.471] | [0.995,2.010] | |
| Control | Yes | Yes |
| pcorrectly | 55.576 | 78.300 |
| ROC | 0.586 | 0.699 |
| N | 2663 | 1318 |
(1) The sample size (N) in Severity model represents the subset of accidents where the driver was determined to be at fault (Entire/Primary responsibility) [18]. The model estimates the odds of serious casualties contingent upon an at-fault violation occurring [21]. (2) Exponentiated coefficients; 95% confidence intervals in brackets; *p < 0.05, **p < 0.01, ***p < 0.001
Unlicensed driving and drunk driving exhibit strong positive associations with at-fault risk among the youngest cohorts. In the Chinese context, young drivers under age 18 operating without a license are associated with more than triple the odds of being involved in an at-fault accident (OR = 3.337) compared to drivers over age 24. Similarly, drivers aged 18–24 without a license show significantly higher odds of at-fault risk (OR = 3.283) and casualty severity (OR = 1.224). Regarding drunk driving, the association with at-fault risk is particularly pronounced for the 18–24 age group (OR = 3.548), showing a 254.8% increase in odds. While drunk driving is also a risk factor for those under 18 (OR = 2.053), its statistical association with casualty severity for this specific sub-group did not reach significance in this sample. These patterns are consistent with identified risk profiles in other high-motorization contexts [32, 50].
Speeding stands out as the factor with the largest effect size regarding at-fault risk, despite a non-significant association with casualty severity in certain cohorts. For drivers under 18, speeding is associated with a nearly eight-fold increase in the odds of at-fault risk (OR = 7.857). This association is even more extreme for young drivers aged 18–24, where speeding is associated with an 11-fold increase in odds (OR = 11.082). The critical role of speed choice as a primary behavioral risk for young drivers has been documented in various field settings [51]. Despite these massive effect sizes for at-fault risk, the correlation between speeding and casualty severity was not statistically significant for these young cohorts in the sampled accidents. This "risk-severity decoupling" suggests that while speeding is a primary predictor of at-fault risk for youth in China, other factors may be associated with the ultimate physical outcome, such as vehicle safety features or younger drivers' faster reaction times [24].
Aggressive driving and signal violations reveal age-dependent risk patterns within the Chinese traffic environment. Aggressive driving is positively associated with at-fault risk for the 18–24 cohort (OR = 1.132), though it shows no significant relationship with outcomes for those under 18. Regarding violation of traffic signals, the lack of statistical significance across all cohorts suggests that its specific effect may be absorbed by other highly correlated infrastructure variables in this sample, such as road type or the presence of traffic control facilities, which mediate the observed risk of signal-running behavior.
In summary, the comparative analysis of these illegal behaviors underscores that young drivers in China are disproportionately associated with high-velocity and high-impairment factors leading to at fault accidents. While behaviors like drunk driving and speeding are strong correlates of at fault risk for all age groups, their magnitude is particularly elevated among youth [52]. Furthermore, while unlicensed driving are linked to both at-fault risk and greater casualty severity, the extreme associations observed for speeding emphasize the critical need for targeted behavioral monitoring for young drivers in China. However, while these subgroup analyses provide valuable insights, several statistical constraints must be noted. Specifically, certain high-risk behaviors—such as drunk driving and speeding among the youngest cohorts—involve a relatively small number of cases. Although the reported Odds Ratios (OR) show large magnitudes, these estimates may be subject to imprecision and wider confidence intervals due to limited subgroup sizes.Therefore, these findings should be interpreted as exploratory and conditional on the available sample size within the Chinese administrative data. To ensure transparency, the specific number of cases (N) for each logistic regression model and relevant subgroups are detailed in the notes of Table 5. By framing these results as exploratory, this study establishes a baseline for future research into behavioral criticality while acknowledging the inherent limitations of small-cell frequency in large-scale datasets.
Conclusion
The identification of key risk factors associated with at-fault risk and casualty severity among young drivers in China highlights the critical need for a dual-risk analytical perspective. By utilizing administrative data from the Ministry of Public Security and a logistic regression framework, this study reveals that the correlates of at-fault risk often diverge from those associated with physical outcomes. Specifically, unlicensed driving and younger age (under 24) exhibit the strongest positive associations with at-fault probability, while drunk driving and speeding remain the most lethal behavioral correlates of casualty severity. Furthermore, the analysis identifies a "risk-outcome paradox" where infrastructures like expressways, despite being associated with lower at-fault odds, exhibit a positive association with significantly higher casualty severity (OR > 2.0). These findings suggest that for young drivers in China, the potential benefits of behavioral caution may be outweighed by the physical dynamics of high-speed environments [47].
Evidence-based policy implications derived from the observed effect sizes provide a prioritized roadmap for targeted road safety interventions in China. Given that speeding and drunk driving exhibit the most substantial associations with at-fault risk (OR > 11 and OR > 3, respectively), administrative efforts should first prioritize stringent AI-driven speed enforcement and the implementation of a Graduated Driver Licensing (GDL) system to manage the high-risk probationary period for novice drivers [53, 54]. To address environmental factors, the significant negative association between traffic signals and at-fault risk supports the expansion of signalized control and improved nighttime lighting, which are associated with a 5.2 to 7.6 percentage point reduction in risk according to our marginal effect estimates. Finally, because unlicensed driving is associated with a 73.7% increase in the odds of casualty severity, and rural household registration points toward higher injury risk, there is a critical need for enhancing emergency medical response systems and roadside protection in rural areas. By ranking these interventions according to their statistical effect sizes, policymakers can maximize safety returns by addressing the most potent correlates of risk first.
This study acknowledges several limitations inherent in police-reported data. First, while authoritative, the exclusion of incomplete records may introduce selection bias. Second, the absence of driving exposure data (e.g., annual mileage) limits the interpretation of absolute risk differences between groups [48]. Consequently, these results reflect statistical associations rather than definitive causal links, and the mechanisms proposed—while aligned with established safety theories—remain hypothesized based on observed patterns [49]. These constraints highlight the need for future research in China to transition toward more technologically integrated datasets. Specifically, the adoption of AI-driven video analytics for detecting near-miss events and smartphone-based telematics to monitor real-time braking and acceleration patterns would provide granular insights into behavioral intent. In the context of China’s ‘Smart City’ initiatives, integrating these technologies with longitudinal medical records and insurance datasets will allow for a more dynamic and causally robust validation of the dual-risk mechanisms identified in this study.
Acknowledgements
This research received no external funding.
Clinical trial number
Not applicable.
Authors’ contributions
Mingqin Chen: Methodology, Data curation, Formal analysis, Writing – original draft. Feiyu Yang: Conceptualization, Writing – original draft. Guangnan Zhang: Conceptualization, supervision, writing – review & editing.
Funding
This research received no external funding.
Data availability
The traffic accident data that support the findings of this study are available from Ministry of Public Security (China), but restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. The data are, however, available from the corresponding author upon reasonable request and with the permission of Ministry of Public Security (China).
Declarations
Ethics approval and consent to participate
Not applicable.
Consent for publication
Not applicable.
Competing interests
The authors declare no competing interests.
Footnotes
According to the Road Traffic Management Information Codes – Part 48: Codes for Types of Traffic Accident Responsibility (GA/T 16.48—2012) issued by the Ministry of Public Security of China, accident liability is categorized into four levels: Entire Responsibility (the driver’s violation is the sole cause of the accident); Primary Responsibility (the driver’s violation is the major contributing factor); Secondary Responsibility (the driver’s violation is a minor contributing factor); and No Responsibility (the driver’s behavior has no causal link to the accident).
According to the Standards for Assessment of Human Body Injury Degree (2004) issued by the Ministry of Justice of China, Serious Injury refers to injuries that cause physical disability, facial disfigurement, loss of hearing, vision, or other organ functions, or other significant harm to human health. Minor Injury refers to injuries causing moderate damage to limbs or appearance, partial impairment of hearing, vision, or organ functions, or other moderate harm to health.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.World Health Orgnization. Global Plan for the Decade of Action for Road Safety 2021–2030. World Health Orgnization; 2021. https://www.who.int/publications/m/item/global-plan-for-the-decade-of-action-for-road-safety-2021-2030.
- 2.Gershon P, Ehsani J, Zhu C, O’Brien F, Klauer S, Dingus T, et al. Vehicle ownership and other predictors of teenagers risky driving behavior: evidence from a naturalistic driving study. Accid Anal Prev. 2018;118:96–101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Williams AF, Tison J. Motor vehicle fatal crash profiles of 13-15-year-olds. J Saf Res. 2012;43(2):145–9. [DOI] [PubMed] [Google Scholar]
- 4.Cooper PJ. Differences in accident characteristics among elderly drivers and between elderly and middle-aged drivers. Accid Anal Prev. 1990;22(5):499–508. [DOI] [PubMed] [Google Scholar]
- 5.Doherty ST, Andrey JC, MacGregor C. The situational risks of young drivers: the influence of passengers, time of day and day of week on accident rates. Accid Anal Prev. 1998;30(1):45–52. [DOI] [PubMed] [Google Scholar]
- 6.Hakamies-Blomqvist LE. Fatal accidents of older drivers. Accid Anal Prev. 1993;25(1):19–27. [DOI] [PubMed] [Google Scholar]
- 7.Macdonald WA. Young driver research program – A review of information on young driver performance characteristics and capacities. Australia: Monash University Accident Research Centre; 1994. [Google Scholar]
- 8.Xie CQ, Parker D. A social psychological approach to driving violations in two Chinese cities. Transportation Research Part F: Traffic Psychology and Behaviour. 2002;5(4):293–308.
- 9.Kaye SA, Rodwell D, Lewis I. A theoretically informed investigation of young drivers' intentions to engage in hand-held phone use for video and text messaging in vehicles with and without ADAS. Transportation Research Part F: Traffic Psychology and Behaviour. 2026;118:103536. [Google Scholar]
- 10.Gauld C, Reeves C. Normative influences on young drivers’ illegal smartphone use: applying an extended theory of normative social behaviour. Accid Anal Prev. 2023;180:106904. [DOI] [PubMed] [Google Scholar]
- 11.Ogden J, Brown PM, George AM. Young drivers and smartphone use: the impact of legal and non-legal deterrents. J Saf Res. 2022;83:329–38. [DOI] [PubMed] [Google Scholar]
- 12.Wang H, Su F, Schwebel DC. Mobile phone use while cycling among e-bikers in China: reasoned or social reactive? J Saf Res. 2023;85:8–14. [DOI] [PubMed] [Google Scholar]
- 13.The United Nations. With 1.3 million annual road deaths, UN wants to halve number by 2030. The United Nations; 2021. https://news.un.org/en/story/2021/12/1107152
- 14.Yang Y, Li M, Lin HC. Parental psychological control, social capital, substance use, and driving under the influence among college students: Sex differences. Journal of Child and Family Studies. 2022:31(5):1207–1219.
- 15.National Highway Traffic Safety Administration. Traffic Safety Facts 2000: A Compilation of Motor Vehicle Crash Data from the Fatality Analysis Reporting System and the General Estimates System (DOT HS 809 337). U.S. Department of Transportation; 2001. https://rosap.ntl.bts.gov/view/dot/5192/dot_5192_DS1.pdf.
- 16.Li Q, Guo F, Klauer SG, Simons-Morton BG. Evaluation of risk change-point for novice teenage drivers. Accid Anal Prev. 2017;108:139–46. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Wang Z, Zheng Z, Fleiter JJ. Does family background impact driving attitudes and risky behaviours? An investigation on Chinese young drivers. Accid Anal Prev. 2016;95:67–77. [DOI] [PubMed] [Google Scholar]
- 18.Zhang G, Yau KKW, Chen G. Risk factors associated with traffic violations and accident severity in China. Accid Anal Prev. 2013;59:18–25. [DOI] [PubMed] [Google Scholar]
- 19.Continental AG, Horizon Corporate Volunteer Consultancy. 2018–2020 White Paper on Distraction and Traffic Accidents among Chinese Youth. Continental AG and Horizon Corporate Volunteer Consultancy; 2021. https://file.smarket.net.cn/s/template/filedetail/?fileId=1531290&scope=snsapi_base&weChatId=1407&retCode=0&excode=bc48514cfaf892e4cc2e472a1b6e5382&sessionid=.
- 20.Zhang G, Yau KKW, Zhang X. Analyzing fault and severity in pedestrian–motor vehicle accidents in China. Accid Anal Prev. 2014;73:141–50. [DOI] [PubMed] [Google Scholar]
- 21.Zhang G, Yau KKW, Zhang X, Li Y. Traffic accidents involving fatigue driving and their extent of casualties. Accid Anal Prev. 2016;87:34–42. [DOI] [PubMed] [Google Scholar]
- 22.Williams AF, Wells JK. Deaths of teenagers as motor-vehicle passengers. J Saf Res. 1995;26(3):161–7. [Google Scholar]
- 23.Winston FK, Kallan MJ, Senserrick TM, Elliott MR. Risk factors for death among older child and teenaged motor vehicle passengers. Archives of pediatrics & adolescent medicine. 2008;162(3):253–260. [DOI] [PubMed]
- 24.Ferguson SA. Other high-risk factors for young drivers—how graduated licensing does, doesn’t, or could address them. J Saf Res. 2003;34(1):71–7. [DOI] [PubMed] [Google Scholar]
- 25.Chapman EA, Masten SV, Browning KK. Crash and traffic violation rates before and after licensure for novice California drivers subject to different driver licensing requirements. J Saf Res. 2014;50:125–38. [DOI] [PubMed] [Google Scholar]
- 26.Hamann C, Price M, Peek-Asa C. Characteristics of crashes and injuries among 14 and 15 year old drivers, by rurality. J Saf Res. 2020;73:111–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 27.Kidd DG, Buonarosa ML. Distracting behaviors among teenagers and young, middle-aged, and older adult drivers when driving without and with warnings from an integrated vehicle safety system. J Saf Res. 2017;61:177–85. [DOI] [PubMed] [Google Scholar]
- 28.Nesson E, Shrestha V. The effects of false identification laws on underage alcohol-related traffic fatalities. Health Econ. 2021;30(9):2264–83. [DOI] [PubMed] [Google Scholar]
- 29.Laapotti S, Keskinen E, Hatakka M, Katila A. Novice drivers’ accidents and violations—a failure on higher or lower hierarchical levels of driving behaviour. Accid Anal Prev. 2001;33(6):759–69. [DOI] [PubMed] [Google Scholar]
- 30.Weast RA, Monfort SS. Characteristics of vehicles driven by teens and adults killed in crashes, 2013–2017. J Saf Res. 2021;77:263–7. [DOI] [PubMed] [Google Scholar]
- 31.Bina M, Graziano F, Bonino S. Risky driving and lifestyles in adolescence. Accid Anal Prev. 2006;38(3):472–81. [DOI] [PubMed] [Google Scholar]
- 32.Martín-delosReyes LM, Martínez-Ruiz V, Rivera-Izquierdo M, Jiménez-Mejías E, Lardelli-Claret P. Is driving without a valid license associated with an increased risk of causing a road crash? Accid Anal Prev. 2021;149:105872. [DOI] [PubMed] [Google Scholar]
- 33.Li G, Baker SP, Langlois JA, Kelen GD. Are female drivers safer? An application of the decomposition method. Epidemiology. 1998;9(4):379–84. [PubMed] [Google Scholar]
- 34.Weiss HB, Kaplan S, Prato CG. Analysis of factors associated with injury severity in crashes involving young New Zealand drivers. Accid Anal Prev. 2014;65:142–55. [DOI] [PubMed] [Google Scholar]
- 35.McCartt AT, Northrup VS, Retting RA. Types and characteristics of ramp-related motor vehicle crashes on urban interstate roadways in Northern Virginia. J Saf Res. 2004;35(1):107–14. [DOI] [PubMed] [Google Scholar]
- 36.Williams AF, Shabanova VI. Situational factors in seat belt use by teenage drivers and passengers. Traffic Inj Prev. 2002;3(3):201–4. [Google Scholar]
- 37.Doherty ST, Andrey JC. Young drivers and graduated licensing: the Ontario case. Transportation. 1997;24(3):227–51. [Google Scholar]
- 38.Gilpin G. Teen driver licensure provisions, licensing, and vehicular fatalities. J Health Econ. 2019;66:54–70. [DOI] [PubMed] [Google Scholar]
- 39.Zhang G, Zhong Q, Tan Y, Yang Q. Risky behavior analysis for cross-border drivers: a logit model and qualitative comparative analysis of odds of fault and injury vulnerability in Guangdong, Hong Kong and Macau. J Saf Res. 2022;82:417–29. [DOI] [PubMed] [Google Scholar]
- 40.Kahane CJ. Injury vulnerability and effectiveness of occupant protection technologies for older occupants and women; 2013.
- 41.Hasselberg M, Vaez M, Laflamme L. Socioeconomic aspects of the circumstances and consequences of car crashes among young adults. Social science & medicine. 2005;60(2):287–295. [DOI] [PubMed]
- 42.Factor R, Mahalel D, Yair G. Inter-group differences in road-traffic crash involvement. Accident Analysis & Prevention. 2008;40(6):2000–2007. [DOI] [PubMed]
- 43.Shavell S. On moral hazard and insurance. The quarterly journal of economics. 1979;93(4):541–562.
- 44.Cummins JD, Phillips RD, Weiss MA. The incentive effects of no-fault automobile insurance. The Journal of Law and Economics. 2001;44(2):427–464.
- 45.Anderson ML, Auffhammer M. Pounds that kill: the external costs of vehicle weight. Rev Econ Stud. 2014;81(2):535–71. [Google Scholar]
- 46.Wilde GJ. The theory of risk homeostasis: implications for safety and health. Risk Anal. 1982;2(4):209–25. [Google Scholar]
- 47.Savolainen PT, Mannering FL, Lord D, Quddus MA. The statistical analysis of highway crash-injury severities: a review and assessment of methodological alternatives. Accid Anal Prev. 2011;43(5):1666–76. [DOI] [PubMed] [Google Scholar]
- 48.Jovanis PP, Chang HL. Modeling the relationship of accidents to miles traveled. Transportation Research Record. 1986;1068(42–51):57–58.
- 49.Mannering FL, Bhat CR. Analytic methods in accident research: Methodological frontier and future directions. Analytic methods in accident research. 2014;1:1–22.
- 50.Shinar D, Schechtman E, Compton R. Self-reports of safe driving behaviors in relationship to sex, age, education and income in the US adult driving population. Accid Anal Prev. 2001;33(1):111–6. [DOI] [PubMed] [Google Scholar]
- 51.Bolderdijk JW, Knockaert J, Steg EM, Verhoef ET. Effects of pay-as-you-drive vehicle insurance on young drivers’ speed choice: results of a dutch field experiment. Accid Anal Prev. 2011;43(3):1181–6. [DOI] [PubMed] [Google Scholar]
- 52.Gesser-Edelsburg A, Zemach M, Lotan T, Elias W, Grimberg E. Perceptions, intentions and behavioral norms that affect pre-license driving among Arab youth in Israel. Accid Anal Prev. 2018;111:1–11. [DOI] [PubMed] [Google Scholar]
- 53.Williams AF. Graduated driver licensing (GDL) in the United States in 2016: A literature review and commentary. Journal of Safety Research. 2017;63:29–41. [DOI] [PubMed]
- 54.Foss RD. Improving graduated driver licensing systems: A conceptual approach and its implications. Journal of Safety Research. 2007;38(2):185–192. [DOI] [PubMed]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Data Availability Statement
The traffic accident data that support the findings of this study are available from Ministry of Public Security (China), but restrictions apply to the availability of these data, which were used under licence for the current study and so are not publicly available. The data are, however, available from the corresponding author upon reasonable request and with the permission of Ministry of Public Security (China).