Exploration of Crash Features of Electric Vehicles with Traffic Crash Data in Changshu, China

Abstract

The rapid development of electric vehicles (EVs) around the world has resulted in new challenges for road safety. Identifying the features of EV crashes is a precondition for developing effective countermeasures. However, due to the short history of EV development, existing studies on EV crashes are quite limited. China, which has the largest EV market in the world, has witnessed a substantial increase in EV crashes in recent years. Therefore, this study comprehensively investigated the characteristics of EV crashes by analyzing the 2023 traffic crash data from Changshu. This is a pioneering study that discusses EV safety by comparing real EV crashes and ICEV crashes from a city in China, the largest EV market in the world. It was found that EV crashes had a higher fatality rate compared to internal combustion engine vehicle (ICEV) crashes. Compared to ICEV crashes, EV crashes are more likely to hit pedestrians and occur during the starting phase. Among the vehicles involved in crashes, the proportion of EVs used for passenger and freight transport was higher than that of ICEVs. In addition, for EV crashes, the proportion of female drivers was much higher, but the proportion of elderly drivers was much lower. Thus, to identify the significant factors influencing crash severity, a logistic regression model was built. The results confirm that EV crashes are more likely to be more fatal than ICEV crashes. In addition, hitting pedestrians and light trucks and crashes occurring in rural areas, at intersections, during winter, and on weekdays could significantly increase the risk of fatalities. These findings are expected to provide new perspectives for improving EV safety within the wave of automotive electrification.

1. Introduction

With the fast development of the global socio-economic situation, the energy crisis and pollution are becoming increasingly prominent. The transportation industry, as a major fuel consumer and greenhouse gas emitter, is being forced to transform to be more energy-efficient and environmentally friendly. The electrification of automobiles is an important measure to reduce fuel consumption and emissions in transportation. Many countries have introduced numerous policies to support the development of electric vehicles (EVs), such as price subsidies, tax exemptions, and the construction of charging infrastructure [1,2]. As a result, EVs have been fast increasing on a global scale in the past decade. According to the International Energy Agency (IEA), the number of EVs in the world increased to 40.5 million in 2023 from 0.4 million in 2013 [3]. However, at the same time, the numbers of traffic accidents involving EVs have also been rising in many countries [4,5,6]. For example, China has the biggest EV fleet and market in the world. From 2018 to 2020, the numbers of traffic crashes and fatalities involving EVs in China grew by 315% and 278%, respectively [6], whereas the number of EVs only increased by 90% at the same time. In addition, both the crash rate and fatality rate of EVs were found to be higher than those of traditional internal combustion engine vehicles (ICEVs) [6]. Therefore, the traffic safety of EVs is becoming a new threat to society and is attracting more attention.

Compared to ICEVs, EVs exhibit many technical differences in their power systems, braking, etc., which lead to many unique challenges. Firstly, EVs usually adopt a regenerative braking system based on the one-pedal driving mode, which allows drivers to accelerate by pressing down on the accelerator pedal and decelerate by lifting off it. When the accelerator pedal is released, the system could reclaim some kinetic energy in braking to enhance the range of EVs. However, due to the different driving mechanisms, this also greatly increases the risk of mistakenly stepping on the accelerator when braking is actually needed, especially when drivers are unfamiliar with the mechanism or in an emergency [7,8]. In addition, EVs are characterized by strong acceleration capabilities due to high torques [9], so it is easy to cause the vehicle to accelerate quickly, which increases the risk of rear-end collisions when an EV is mis-operated. Norway has the highest penetration rate of EVs globally. Hou et al. [8] analyzed EV crashes in Norway from 2020 to 2021 and found that rear-end collision was the main collision type. Another study on Tesla crashes in Pennsylvania, U.S., also identified that Tesla vehicles had a higher proportion of rear-end collisions than ICEVs [10]. Secondly, compared to ICEVs, EVs operate with significantly lower noise due to the absence of internal combustion engines [11,12]. Although the quietness makes occupants feel more comfortable, it is also thought to be responsible for the high proportion of pedestrian/cyclist collisions in EV crashes. After analyzing traffic crashes in Norway from 2011 to 2018, Liu et al. [13] found that EV crashes were much more likely to involve pedestrians or cyclists compared to ICEV ones. Similarly, a study on EV crashes in Changsha, China, from January 2020 to June 2022, indicated that collisions with pedestrians represented 24.3% of total EV crashes [14]. Wen et al. [15] analyzed EV crashes in Guangdong Province, China, and identified several factors affecting the likelihood of severe injuries, including environmental characteristics, temporal characteristics, vehicle characteristics, road characteristics, and traffic flow characteristics. Zhang et al. [16] analyzed collisions between EVs and pedestrians in the U.K. and found that EV crashes were increasing in number year by year, with more crashes occurring at night. However, as vulnerable road users, pedestrians/cyclists are easily injured in collisions. Su et al. found that 46.7% of vehicle–pedestrian collisions resulted in fatalities [14]. Based on U.K. road safety data from 2009 to 2019, Zhang et al. [17] found that in new energy vehicle (NEV) crashes, vulnerable groups such as bicycle/motorcycle riders and pedestrians were prone to serious injuries and deaths. Additionally, at present, lithium-ion-powered batteries are widely used as energy storage components in EVs; the chemical properties of these batteries also present new challenges to traffic safety [18], as they catch fire easily in overcharging and short-circuiting conditions, as well as during collisions, posing a significant threat [19].

Crash analysis is the most important method for exploring traffic safety. In recent years, with the increase in EV traffic crashes, more researchers have been conducting studies on EV traffic safety by analyzing crashes. Using the EV crash data for Norway, Liu et al. [13] found that EV crashes mainly occurred during peak hours on weekdays, in urban areas, at road junctions, on low-speed roads, and in good-visibility scenarios. Su et al. [14] analyzed EV crash data from Changsha, China, and discussed the factors influencing their severity. However, most EVs involved were being used for commercial purposes, such as taxis and buses, in Su et al.’s study. Based on insurance data from the Netherlands, McDonnell et al. [20] found that EVs had more at-fault claims than traditional ICEVs, which implies the high risk of EVs. Tesla is the best-selling EV brand in the U.S. Based on Tesla crash data from Pennsylvania, USA, Liu et al. [10] found that Tesla crashes had a higher proportion of angle and rear-end collision crashes than ICEV crashes. Stolle et al. [21] compared the differences between BEV, HEV, and ICEV crashes using data from eight U.S. states from 2017 to 2021, and their comparison found that BEVs had a lower percentage of single-vehicle crashes and winter crashes, compared to ICEV crashes.

Although researchers have discussed the traffic safety features of EVs from different perspectives, given the short history of large-scale adoption of EVs, the related research remains very limited. Firstly, except for a few countries, such as China, Norway, and the United States, most countries have been slow to promote EVs, and there has been limited research on EV crashes. Secondly, many studies discuss features of EV crashes without comparing them with ICEV ones [8,14], making it unclear whether these features are unique to EVs or common to all vehicles. Thirdly, due to the initially high prices of EVs, they were primarily promoted for adoption among specific groups in the early stage. For example, in China, EVs were initially adopted primarily as taxis, ride-hailing vehicles, buses, urban freight trucks, etc., supported by government policies. Therefore, EV crashes in many studies primarily involve vehicles used for commercial purposes [14], which are expected to exhibit significantly different driving features compared to private vehicles. As EVs are increasingly being adopted as private vehicles nowadays, it is necessary to explore the characteristics of crashes involving private EVs. Finally, existing studies have primarily focused on exploring whether EVs and ICEVs exhibit significant differences in crash severity rather than identifying the key factors influencing their crash occurrences.

In this regard, this study aims to reveal the characteristics and influencing factors of EV crashes through an in-depth analysis of the 2023 EV crash data in Changshu, China. The organization of this paper is as follows: We introduce the materials used in this study in Section 2. Then, we describe the models used in this study in Section 3. Section 4 conducts a statistical analysis to identify the factors significantly influencing the severity of EV crashes. In Section 5, we showcase the conclusions and discussion.

2. Materials

Changshu is located in the Yangtze River Delta, the most developed metropolitan area in China, with a population of 1.68 million. Amid the wave of automotive electrification, the adoption of EVs in Changshu has grown rapidly. The number of registered EVs increased from 21,127 in 2022 to 35,884 in 2023, with a growth rate of 69.8%, whereas the total number of registered vehicles only grew by 4.0% during the same time.

This study collected traffic crash data from Changshu between 3 January 2023 and 1 January 2024 from the local traffic police department. A total of 1507 crashes were recorded, of which 112 involved EVs. Although EV crashes accounted for only 7.4% of total crashes, they resulted in 12.4% of total fatalities, demonstrating their high risks. A summary of the characteristics of EV crashes is presented in

Table 1. Summary of characteristics of EV and ICEV crashes.

Category	Variable	Definition	EV	ICEV
Collision	Severity	Property damage only (PDO)	6.2%	8.7%
		Injury	78.6%	82.4%
		Fatal	15.2%	8.9%
	Collision Type	Hit moving vehicles	77.7%	82.3%
		Hit fixed object	1.8%	2.8%
		Hit stationary vehicles	5.4%	5.5%
		Hit pedestrians	15.2%	9.5%
	Unit Count	1	0.9%	2.4%
		2	78.6%	75.8%
		3	16.1%	17.7%
		≥4	4.4%	4.1%
Vehicle	Vehicle Usage	Private	74.1%	92.6%
		Passenger (taxi, ridesharing, etc.)	10.7%	0.4%
		Freight	6.2%	0.4%
		Special (police, sprinkler, etc.)	8.9%	6.5%
	Vehicle Type	Passenger car	91.1%	96.7%
		Light truck	8.9%	3.3%
	Movement	Straight	73.2%	68.6%
		Turning	13.4%	19.1%
		Stationary	7.1%	9.0%
		Starting	2.7%	0.1%
		Other	3.6%	3.2%
Driver	Driver Gender	Male	63.4%	81.2%
		Female	36.6%	18.8%
	Driver Age (Years)	≤30	27.7%	25.4%
		(30, 60]	71.4%	71.0%
		>60	0.9%	3.5%
Roadway	Location	Segment	83.0%	75.2%
		Intersection	17.0%	24.8%
	Area	Urban	71.4%	60.1%
		Rural	28.6%	39.9%
Environment	Time of Day	Nighttime (19:00–7:00)	33.0%	32.6%
		Daytime (7:00–19:00)	67.0%	67.4%
	Day of Week	Weekday	72.3%	72.1%
		Weekend	27.7%	27.9%
	Weather	Good	85.7%	89.3%
		Adverse	14.3%	10.7%

, where ICEV crashes are also analyzed for comparison. It should be noted that since all the EVs involved were small-sized, only the small-sized ICEVs were retained for comparison.

2.1. Collision Features

The severity of crashes is a core issue in traffic safety research. As shown in Table 1, the proportion of fatal EV crashes is 15.2%, significantly higher than that of ICEV crashes (8.9%). Meanwhile, the proportion of PDO crashes among EV crashes is 6.2%, which is also lower than that among ICEV crashes (8.7%). That is, the consequences of EV crashes appear to be more severe than those of ICEV crashes.

In terms of collision type, EV crashes primarily involve collisions with moving vehicles (77.7%), pedestrians (15.2%), and stationary vehicles (5.4%). It should be noted that the proportion of hit-pedestrian collisions among EV crashes is obviously higher than that among ICEV crashes (9.5%). Due to the absence of internal combustion engines, EVs are expected to be much quieter, especially in low-speed scenarios, and, thus, may often go unnoticed by pedestrians. However, since pedestrians are vulnerable to collisions, hit-pedestrian crashes often result in severe consequences.

Table 2. Distribution of crashes by severity and collision type.

	Collision Type	Hit Moving Vehicles	Hit Fixed Objects	Hit Stationary Vehicles	Hit Pedestrians
EV	PDO	1.1%	50.0%	83.3%	0.0%
	Injury	87.4%	50.0%	16.7%	58.8%
	Fatal	11.5%	0.0%	0.0%	41.2%
ICEV	PDO	4.3%	88.1%	60.0%	0.0%
	Injury	88.1%	20.0%	38.0%	76.7%
	Fatal	7.6%	12.0%	2.0%	23.3%

indicates that the fatal crash proportion is 41.2% for EV hit-pedestrian crashes, significantly higher than that for ICEV ones (23.3%), highlighting the massive threat of EVs to pedestrians. However, the proportion of hit-pedestrian collisions among EV crashes in Changshu is significantly lower than that in another study analyzing EV crashes in Changsha (24.3%) between 1 January 2020 and 27 June 2022 [14]. This is thought to be partly attributable to the mandatory installment of warning sound systems in EVs to alert pedestrians since 1 July 2019 [22].

Regarding the unit count involved in crashes, the proportion of single-unit (which also means single-vehicle) EV crashes is significantly lower than that among ICEV crashes, which is consistent with the study by Stolle et al. [21]. Their study also reported that the proportion of single-vehicle crashes was lower for BEVs than for ICEVs.

2.2. Vehicle Features

Vehicles are another factor that can greatly influence crash consequences. Regarding vehicle usage, 74.1% of the EVs involved are used for private purposes, which is nearly twice the percentage reported by Su et al. [14] for their analysis of EV crashes (37.5%). Initially, to promote the development of EVs, agencies implemented various measures to encourage the adoption of EVs as buses, taxis, and other public service vehicles in China [23,24,25]. After multiple years of efforts, it seems EVs have become an important choice for many residents, rather than being confined to public service sectors. However, considering the huge differences in usage patterns between public service vehicles and private vehicles, this transition might somehow significantly impact traffic safety.

In terms of vehicle type, the proportion of light trucks involved in EV crashes (8.9%) is more than double that of light trucks involved in ICEV crashes (3.3%). In Changshu, electric light trucks are primarily used in urban logistics. Urban logistics transportation is characterized by short distances, low tonnages, and frequent stop-and-go traffic congestion, which makes it especially suitable for EVs. Therefore, urban logistics is one of the major sectors being promoted for EV adoption in China [26,27]. As a result, electric light trucks have become increasingly popular.

In terms of vehicle movement, the proportion of vehicles in the starting phase was 2.7% for EV crashes, compared to only 0.1% for ICEV crashes. That is, EVs appear to be more likely to lose control during the starting phase. This could be attributed to the fact that EVs generally have greater acceleration capabilities than ICEVs due to their electricity-driven power systems. However, this feature can be particularly challenging for drivers who are unfamiliar with it.

2.3. Driver Features

From the gender perspective, female drivers account for 36.6% of EV crashes, a proportion significantly higher than that for ICEV crashes (18.8%). Li et al. found that the proportion of females choosing EVs is 1.22 times higher than that of those choosing ICEVs [28]. A possible explanation for this is that females are more concerned about environmental protection compared to males, which leads them to be more inclined to purchase EVs [29,30]. In addition, the low noise, ease of use, and smooth acceleration of EVs have been identified as important factors in attracting female consumers [31].

2.4. Roadway Features

Regarding the area, most crashes occur on urban roads. The distribution of crash severity across areas is presented in

Table 3. Distribution of crashes by severity and area.

	Area	Urban	Rural
	PDO	6.3%	6.3%
EV	Injury	83.7%	65.6%
	Fatal	10.0%	28.1%
	PDO	9.7%	7.2%
ICEV	Injury	84.6%	79.0%
	Fatal	5.7%	13.8%

. It is evident that rural crashes have a higher proportion of fatal crashes than urban crashes. In the case of EV crashes and ICEV crashes, the proportion of fatal crashes on rural roads reaches 28.1% and 13.8%, respectively. This may be attributed to the fact that rural roads are often inadequately equipped with safety facilities and lacking measures to mitigate crash severity. Additionally, rural areas are typically located in less accessible regions, where effective rescue may take a longer time to arrive. This could also be one reason for the higher proportion of fatal crashes on rural roads. However, China is currently promoting the adoption of EVs in rural areas. As more EVs operate in rural areas, new traffic safety challenges are also expected to emerge.

Intersections are characterized by frequent traffic conflicts, accelerations, and decelerations, making them more hazardous [32,33]. In terms of location, both EV and ICEV crashes predominantly occur on road segments, with intersections accounting for 17.0% and 24.8% of crashes, respectively. However, as shown in

Table 4. Distribution of crashes by severity and location.

	Location	Segment	Intersection
	PDO	7.5%	0.0%
EV	Injury	79.6%	73.7%
	Fatal	12.9%	26.3%
	PDO	10.5%	3.1%
ICEV	Injury	81.6%	84.9%
	Fatal	7.9%	12.0%

, among the EV crashes at intersections, 26.3% result in death, a proportion higher than that for ICEVs (12.0%). Although the proportion of EV crashes occurring at intersections is lower than that of ICEV crashes, the fatality rate of EV crashes at intersections is twice that of ICEV crashes. This may be attributed to the EVs’ rapid acceleration capabilities and low noise levels, which often make them difficult to notice when approaching vulnerable road users at intersections, leading to severe collisions. Therefore, the safety of EVs at intersections warrants greater attention.

2.5. Environment Features

Regarding the time and day of the week, EV crashes and ICEV crashes exhibit no significant differences. However, by comparing the severity of crashes on weekdays and weekends, as shown in

Table 5. Distribution of crashes by severity and day of week.

	Day of Week	Weekday	Weekend
	PDO	7.4%	3.2%
EV	Injury	75.3%	87.1%
	Fatal	17.3%	9.7%
	PDO	7.8%	11.1%
ICEV	Injury	82.3%	82.6%
	Fatal	9.9%	6.3%

, it can be found that the proportion of fatal crashes on weekends is lower than that on weekdays.

Figure 1 illustrates the distribution of crash severity across the seasons, with December, January, and February classified as winter; March, April, and May as spring; June, July, and August as summer; and September, October, and November as autumn. It is evident that the proportion of fatal crashes involving EVs is significantly higher in winter compared to spring, summer, and autumn. Notably, the proportion of fatal crashes among EV crashes in winter reaches up to 45.5%. This can be attributed to the adverse weather conditions in winter, such as dense fog, snow, and rain, which severely reduce visibility. Reduced visibility can lead to insufficient reaction times during crashes, often leading to more severe consequences. According to research by Lee et al. [34], the number of crash casualties noticeably increases when winter temperatures drop below zero degrees. Additionally, the low temperatures in winter can affect the performance of tires and the effectiveness of lubricants, which may also indirectly contribute to the higher proportion of fatal crashes in winter.

3. Methods

Since the primary goal for traffic police departments in China is to reduce the number of deaths caused by crashes, it is essential to identify the critical factors influencing the fatality of crashes. Therefore, all crashes were divided into two categories by severity: fatal and non-fatal (PDO and Injury). The logistic regression model is a widely used statistical learning method for binary classification problems and has been used extensively in traffic safety research [10]. Accordingly, a logistic regression model is built here and shown as follows:

$$y_i ~ \mathrm{Binomial}(P_i)$$

(1)

$$p_i={\displaystyle \frac{1}{1+\exp \left[-\left({\beta }_0+{\displaystyle \sum _{j=1}^m{\beta }_j}{x}_{ij}\right)\right]}}$$

(2)

where $i$ is the number assigned to the crash;

• $y_i$ is the severity of the $i\mathrm{th}$ crash, with 0 for non-fatal and 1 for fatal;
• $p_i$ is the probability of the $i\mathrm{th}$ crash being fatal;
• $m$ is the number of independent variables;
• $x_i=(x_{i1},x_{i2}\dots ,x_{im})$ is the independent variable vector of the $i\mathrm{th}$ crash;
• ${\beta }_0,{\beta }_1\dots ,{\beta }_m$ are the regression coefficients.

In addition, the marginal effects of independent variables were calculated. Since all the independent variables are categorical data, they are converted into dummy variables in the regression analysis. For a dummy variable, the marginal effects show the fatal probability change when it changes from 0 to 1, while other independent variables are the same. It is calculated as follows:

$$ME(z)={\displaystyle \frac{{\displaystyle \sum _{i=1}^n\left[P_i(z=1)-P_i(z=0)\right]}}{n}}$$

(3)

where $ME(z)$ is the average marginal effect of dummy variable $z$.

Table 6. Summary of variables used for regression analysis of crash fatality.

Variable	Definition	Proportion
Dependent
Crash Severity	0 for Non-fatal	84.8%
	1 for Fatal	15.2%
Independent
Collision Type	Hit moving vehicles *	81.8%
	Hit fixed objects	8.1%
	Hit pedestrians	10.1%
Driver Gender	0 for Male *	79.2%
	1 for Female	20.8%
Driver Age	0 for Old (>30) *	74.3%
	1 for Young (≤30)	25.7%
EV	0 for No *	89.0%
	1 for Yes	11.0%
Vehicle Type	0 for Passenger car *	96.1%
	1 for Light truck	3.9%
Area	0 for Urban *	61.4%
	1 for Rural	38.6%
Location	0 for Segment *	76.1%
	1 for Intersection	23.9%
Season	Spring (March to May) *	22.9%
	Summer (June to August)	33.7%
	Autumn (September to November)	33.1%
	Winter (December to February)	10.2%
Day of Week	0 for Weekday *	72.2%
	1 for Weekend	27.8%
Time of Day	0 for Daytime *	67.4%
	1 for Nighttime	32.6%
Weather	0 for Good *	88.9%
	1 for Adverse	11.1%

Note: * indicates the baseline.

shows a summary of the variables adopted in the regression analysis. Many variables were reorganized to identify the effects of various factors more accurately. For example, in terms of collision type, collisions hitting stationary vehicles were merged with collisions hitting fixed objects as one type. Since the number of older drivers was limited, they were merged with middle-aged ones. The days of the week were divided into weekdays and weekends. The weather was divided into good and adverse (rain, snow, etc.). Finally, 112 EV crashes and 908 ICEV crashes were kept for the regression analysis.

4. Results

The estimated results of the logistic regression model for the severity of crashes are shown in

Table 7. Estimated results of logistic regression analysis of crash severity.

Variable	Coefficient	Standard Error	Z Value	p Value	ME
(Intercept)	−2.922	0.328	−8.904	<0.001 *	NA
Collision type—Hit fixed objects	−0.468	0.557	−0.841	0.401	NA
Collision type—Hit pedestrians	1.675	0.293	5.719	<0.001 *	0.195
Female	−0.381	0.307	−1.242	0.214	NA
Young	−0.089	0.269	−0.333	0.739	NA
EV	0.696	0.328	2.121	0.034 *	0.063
Light truck	0.969	0.448	2.163	0.031 *	0.098
Rural	0.779	0.237	3.285	0.001 *	0.061
Intersection	0.694	0.257	2.703	0.007 *	0.059
Season—Summer	−0.333	0.329	−1.013	0.311	NA
Season—Autumn	−0.188	0.323	−0.583	0.560	NA
Season—Winter	1.050	0.343	3.067	0.002 *	0.105
Weekend	−0.586	0.284	−2.063	0.039 *	−0.041
Nighttime	−0.099	0.258	−0.384	0.701	NA
Weather—Adverse	−0.425	0.402	−1.056	0.291	NA

Note: *, significant at 95% confidence interval.

.

The Hosmer–Lemeshow test was used to evaluate the model’s goodness of fit. The p-value of the H-L test is 0.306, which is greater than the commonly used significance level, so we cannot reject the original hypothesis. This means there is no significant difference between the model’s predicted and actual observed values, suggesting that the model is well fitted.

Bootstrapping provides a resampling simulation approach to estimate standard errors and other measures of statistical precision by repeatedly and randomly sampling subsets of data from the original dataset. To estimate the accuracy of the logistic model estimates, we conducted bootstrap sampling with 5000 replications. The results are presented in

Table 8. Bootstrapping results.

Variable	Original Coefficient	Bias	95% Confidence Interval
(Intercept)	−2.922	−0.057	(−3.597, −2.236)
Collision type—Hit fixed objects	−0.468	−0.387	(−1.874, 0.492)
Collision type—Hit pedestrians	1.675	0.021	(1.036, 2.260)
Female	−0.381	−0.023	(−1.037, 0.226)
Young	−0.089	−0.014	(−0.644, 0.455)
EV	0.696	−0.002	(0.046, 1.252)
Light truck	0.969	−0.009	(−0.038, 1.932)
Rural	0.779	0.015	(0.278, 1.247)
Intersection	0.694	0.006	(0.140, 1.231)
Season—Summer	−0.333	−0.001	(−1.013, 0.351)
Season—Autumn	−0.188	0.002	(−0.825, 0.495)
Season—Winter	1.050	0.029	(0.335, 1.734)
Weekend	−0.586	−0.032	(−1.152, −0.006)
Nighttime	−0.099	−0.015	(−0.627, 0.421)
Weather—Adverse	−0.425	−0.076	(−1.248, 0.319)

. It can be seen that the bias of the variables is very small, except that for “hit fixed objects”. The small biases for most variables suggest that the model estimates are stable and reliable.

4.1. Collision Features Analysis

Collision type is one of the most important factors influencing crash outcomes. Here, the coefficient of the hitting-pedestrian collisions is 1.668, with marginal effects of 0.195. That is, hit-pedestrian collisions are 19.5% more likely to result in death than collisions with a moving vehicle. As mentioned above, pedestrians lack protection and are highly vulnerable to severe injuries in collisions. Given that hit-pedestrian collisions account for the second largest share of crashes, especially for EV crashes, it is crucial to take adequate measures to prevent their occurrence. A further exploration indicates that most hit-pedestrian collisions can be attributed to driver faults, such as not yielding to pedestrians. Therefore, agencies should focus on strengthening driver education and law enforcement to reduce such misbehaviors.

Meanwhile, hit-fixed-object collisions do not show significant differences. Theoretically, those with fixed objects often produce a significant shock compared to collisions with moving vehicles. Many studies have highlighted that hit-fixed-object collisions often result in severe outcomes [35,36], but it should be noted that these crashes mainly occur on high-speed roadways, such as freeways. However, since Changshu is a small city, most crashes occur on low-speed urban roadways, dramatically reducing their severity.

4.2. Vehicle Features Analysis

EVs show significantly positive effects on the crash severity, with a coefficient of 0.696 and marginal effects of 0.063, consistent with the finding in Section 2. That is, compared to ICEV crashes, EV crashes are 6.3% more likely to result in death. It should be noted that although many studies have discussed the severity of ICEV and EV crashes, this is one of the pioneering studies quantifying the risk of EV crashes by regression analysis.

Light trucks also show a significantly positive effect, with a coefficient of 0.969 and marginal effects of 0.098. That is, compared to passenger cars, light trucks are 9.8% more likely to result in death in crashes. In the context of this study, these light trucks are mainly used for urban logistics and have gross vehicle weights of less than 4.5 tons [37]. Although many studies have also discussed the dangers of trucks, they primarily targeted heavy trucks used for long-distance travel. Our study further verifies the risk associated with light trucks in collisions.

4.3. Driver Features Analysis

Drivers are also important factors influencing crash features. Here, driver gender does not show significant effects, which is thought to be reasonable. Theoretically, there is no basis to think that female drivers are riskier, as they pass the same driver’s license exam as males. Additionally, driver age does not have significant effects. These results are consistent with the findings of Su et al. [14].

4.4. Roadway Features Analysis

In terms of area, rural crashes show significantly positive effects, with a coefficient of 0.779 and marginal effects of 0.061. That is, crashes occurring on rural roadways are 6.1% more likely to be fatal than those on urban roadways, which might be attributed to multiple factors. Firstly, in Changshu, rural roadways generally feature higher speed limits, more trucks, and fewer traffic control devices, all of which could lead to severe consequences in collisions. Secondly, emergency response times are generally shorter in urban areas than in rural areas after crashes occur, as major hospitals are located around urban areas in Changshu. According to Adeyemi et al., an increase in response time has an effect on the fatality rate, and the fatality rate is higher in rural than urban areas [38]. Also, the likelihood of high-speed vehicle collisions is relatively low in urban areas, possibly contributing to lower fatality rates than in rural areas.

The location also shows significant effects, with a coefficient of 0.694. That is, crashes at intersections are 5.9% more likely to result in death than those on road segments, which can also be attributed to multiple factors. Firstly, as is well known, angle crashes are one of the most dangerous crash types [39], and they primarily occur at intersections. When drivers run red lights or stop signs, it easily results in high-speed angle collisions with the crossing traffic, leading to severe outcomes. Besides this, the many turning movements at intersections increase the risk of angle collisions and rollovers, such as conflicts involving straight traffic and left-turning traffic. Secondly, pedestrians, cyclists, and other non-motorized road users mainly cross roadways at intersections, increasing the likelihood of vehicle collisions with them and resulting in casualties.

4.5. Environment Features Analysis

The season also shows significant effects in this study. The results indicate that crashes occurring in winter are 10.5% more likely to be fatal than those in other seasons. According to Yu et al. [40], low temperatures increase the probability of severe crashes. It is thought that due to the low temperatures in the winter, roadway surfaces easily become slippery in the presence of rain and water, which might cause vehicles to lose control and result in severe collisions. Further investigations are suggested to identify the underlying causes. However, this finding highlights the importance of reducing winter crashes. In addition, the weekend shows significantly negative effects, with a coefficient of −0.586 and marginal effects of −0.041. That is, crashes occurring on weekends are 4.1% less likely to be fatal than those on weekdays, which might be attributed to differences in travel patterns between weekdays and weekends. People primarily travel for work on weekdays but for leisure and shopping on weekends. Compared to weekdays, when commuters are concentrated during peak hours, weekend travel is more dispersed, reducing the likelihood of death.

Nighttime does not show significant effects. Although crashes are expected to occur more at nighttime (19:00–7:00) due to low visibility, it does not appear to influence crash severity. Weather does not show significant effects either. It should be noted that in the survey period, some crashes occurred in rainy conditions, but none occurred in snowy conditions.

5. Conclusions and Discussion

With the rapid increase in the number of EVs, EV safety has emerged as a new traffic safety challenge. Understanding the characteristics of EV crashes is the precondition for developing effective countermeasures. Therefore, this study investigated features of EV crashes by analyzing traffic crash data for Changshu, China, from 3 January 2023 to 1 January 2024. Firstly, this study provided an in-depth comparative analysis of EV and ICEV crashes to identify their differences and similarities. Subsequently, a logistic regression model was developed to identify the key factors influencing the fatality of crashes. The specific findings are as follows.

Firstly, the fatality of EV crashes is notably higher than that of ICEVs. Secondly, the collision features of EVs differ from those of ICEVs. EV crashes more frequently involve hit-pedestrian collisions, with a proportion of 15.2% compared to 9.5% for ICEV crashes. This is particularly concerning as pedestrians are vulnerable road users. In terms of vehicle features, EV crashes more often occur during the starting phase, with a proportion of 2.7% compared to 0.1% for ICEV crashes. This may be due to the high torque of EVs, which can lead to a loss of control or excessive acceleration if not operated properly. The model results indicate that the severity of EV crashes is higher than that of ICEV crashes. At the same time, it was found that crashes were more likely to result in death when they involved light trucks. Crashes in rural areas are more likely to result in death, possibly because rural roads often lack secure devices and are farther from hospitals, resulting in longer emergency response times after a crash. Crashes occurring at intersections are more likely to result in death due to the complexity of traffic flows and the potential for serious collisions. Furthermore, crashes occurring in winter are more likely to be fatal, attributed to adverse weather conditions such as rain, fog, and cold temperatures. Crashes on weekends are less likely to result in death, which may be related to people’s travel habits.

The findings of this study have several implications for enhancing the safety of EVs. Firstly, the higher fatality rate of EV crashes highlights the need for targeted safety measures. The higher proportion of EV crashes during the starting phase suggests that driver training and education programs may be necessary to ensure safe operation, particularly during the starting phase. The impact of environmental factors on crash severity underscores the importance of infrastructure improvements, such as better lighting and visibility at intersections, and the provision of weather-appropriate driving guidelines. Furthermore, the seasonal variation in crash severity suggests that additional safety measures may be required during winter.

In response to these findings, automakers could improve the traffic safety of EVs in several ways. Firstly, considering the significantly high proportion of hit-pedestrian collisions in EV crashes, developing advanced pedestrian protection devices, such as more prominent acoustic vehicle alerting systems and more sensitive pedestrian monitoring systems to automatically slow down or brake when a pedestrian is detected, is important. Secondly, the poor performance of EV batteries in low temperatures might exacerbate crash severity in winter. Therefore, manufacturers need to develop more advanced Battery Management Systems (BMSs) to enhance the durability of batteries in winter, such as boosting the battery temperature through the preheat function and increasing insulation measures in vehicle designs to mitigate the impact of low temperatures. Thirdly, light truck crashes were found to be more severe, whereas electric light trucks are currently being promoted for urban logistics in China. Therefore, it is necessary to equip these trucks with advanced driver assistance systems, such as lane departure warnings, blind zone monitoring, and automatic emergency braking, to reduce the risk of crashes.

In conclusion, this study comprehensively analyzed the characteristics and influencing factors of EV crashes. The findings highlight the need for a multi-faceted approach to enhance EV safety, involving technological advancements, driver education, and infrastructure improvements. Future research should continue to monitor the evolving nature of EV crashes as the adoption of EVs increases and new safety technologies are developed. However, due to the limited access to data, the data in this study covered only one year in Changshu City. In fact, studies in different regions have found inconsistent magnitudes of the relationship between the fatality rates of EV and ICEV crashes [6,13,14,21]. Therefore, whether the conclusion that EV crashes are more severe than ICEV crashes can be generalized to other regions remains to be further discussed. We aim to collect EV crash data from a broader area in future research. Meanwhile, with the accumulation of EV crash data, EV crashes are expected to be analyzed exclusively to identify the factors affecting their severity in the future. With more data support, we will explore the factors influencing EV crash severity using more complex modeling methods, such as ordered logit models [8] and random parameter binary probit models [15].