Exploring Determinants of Traffic Accident Severity Using Empirical

Data

Danying Wei

SWUFE-UD Institute of Data Science at SWUFE, Southwestern University of Finance and Economics, Chengdu, China

Keywords: Traffic Accident, Data-Driven Approach, Ordinary Least Squares, Random Forest Regression.

Abstract: Traffic accidents have become a major global public safety concern, as they affect individuals and their

families and hinder the development of a country’s economy. This study uses empirical data from Nashville

to investigate the influence of individual characteristics and external environmental factors on the severity of

traffic accidents. A multiple linear regression model and a nonlinear model are employed to examine the

relationships between accident severity and various factors, including periods, weather conditions,

illumination, collision types, and hit-and-run behaviour. The results indicate that accidents occurring during

early morning, evening, and night are more severe; weekend accidents are more serious than weekday

accidents. Interestingly, severe weather such as snow and blowing snow reduces accident severity, while

foggy and cloudy conditions increase it. Poor visibility conditions, such as darkness, dawn, and dusk,

significantly elevate accident severity. Moreover, head-on collisions and hit-and-run behaviour are strongly

associated with more severe outcomes. These findings contribute to improving traffic safety policies and

provide practical implications for accident prevention. Future studies may consider incorporating more

complex models and a broader range of variables to enhance predictive performance and policy relevance.

1 INTRODUCTION

Traffic accidents and safety have long been issues of

great concern to people worldwide. According to data

released by the World Health Organization,

approximately 1.35 million people die in traffic

accidents each year, and tens of millions more suffer

varying degrees of injury (World Health

Organization, 2021). In many countries, road traffic

accidents have become the leading cause of death for

young people aged 15 to 29 (World Health

Organization, 2021). The occurrence of traffic

accidents not only has a significant impact on

individuals and families, but the property losses,

damage to public facilities, and public panic they

cause will, to a certain extent, hinder urbanization and

national economic development. Therefore, how to

effectively identify and control the key influencing

factors that lead to traffic accidents, especially serious

accidents, has become one of the core issues in

current global traffic governance.

In the relevant research on traffic accidents, most

scholars focus on accident prediction, detection, and

the impact of individual variables on accidents.

However, the systematic analysis of "the severity of

traffic accidents" remains relatively limited. Factors

affecting the severity of accidents, as potential causes

of traffic accidents, hold significant reference value

for developing more refined and tiered intervention

policies.

Looking back at the relevant literature, scholars

usually explore the causes of traffic accidents from

three levels: the individual level, external

environmental factors, and macro-social factors. At

the individual level, variables such as gender, age,

and driving experience are considered to have a

significant impact on driving risk. Men are more

likely to be influenced by their emotions and take

irrational driving actions (Khan et al., 2020). Young

drivers are more likely to view driving as a risky

behaviour due to their immature mentality, which can

lead to serious traffic accidents (Khan et al., 2020).

But at the same time, experienced drivers can also

cause accidents due to overconfidence and negligence

(Gebre Meles et al., 2022). It can be seen that

individual differences will affect drivers' psychology

and prompt them to adopt different driving

behaviours. However, existing research does not

capture more dynamic and subjective factors such as

the driver's psychological state and behavioural

672

Wei, D.

Exploring Determinants of Trafﬁc Accident Severity Using Empirical Data.

DOI: 10.5220/0013852100004719

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 2nd International Conference on E-commerce and Modern Logistics (ICEML 2025), pages 672-680

ISBN: 978-989-758-775-7

responses, and is unable to explain the specific role of

individual differences in more detail. In addition, it is

limited by data availability and privacy protection,

resulting in a narrow coverage of variables. At the

level of external environmental factors, weather

conditions (such as rain, snow, and fog), road

structure, and period (such as night or early morning)

all have a certain correlation with the severity of the

accident. For example, rainy days make traffic

accidents more likely but less severe (Edwards,

1998). In the early hours of the morning, drivers are

prone to drowsiness and a lack of concentration, and

accidents may be more serious (Iqbal et al., 2020). It

can be seen that external environmental factors not

only affect the probability of accidents but also have

complex effects on the outcomes of accidents.

However, current research focuses more on the linear

impact of a single variable and rarely reveals the

interactive relationship between environmental

variables. In addition, at a more macro level, stock

market fluctuations, epidemics, and autonomous

driving technology will all affect drivers' emotions or

social behavior, thereby indirectly affecting traffic

safety. For example, volatility in stock market returns

is statistically significantly correlated with traffic

accidents, which may be related to the fact that

investor mood swings (such as excitement or

depression) affect driving behavior (Giulietti et al.,

2020). On the other hand, as a technological hotspot

in recent years, autonomous driving technology has

shown great potential in improving road safety, but it

is also accompanied by problems such as perception

limitations and ethical conflicts. Relevant research

mostly stays at the correlation level. It rarely explores

the mode of influence, making it difficult to

accurately explain how macro factors affect the

outcome of accidents by influencing driver behavior.

In summary, although the above studies have their

emphases, they mostly focus on the impact of a single

influencing variable on accidents, but fail to achieve

integrated modeling, ignore the interaction of

multiple variables, and have certain limitations.

Based on the above background, this study will

systematically analyze the factors that affect the

severity of traffic accidents, comprehensively

consider multiple external environmental conditions

that affect driving behavior, and use actual traffic

accident data in Nashville for empirical modelling

analysis. This paper conducts data cleaning and

variable construction to generate a continuous

variable "Severity Score" to measure the severity of

the accident, and removes missing or outliers. This

paper will use the multivariate linear regression

model (OLS) to identify the direction, significance,

and relative influence of each variable, and at the

same time, attempt to curve fit the possible nonlinear

effects to enhance the explanatory power of the

model.

This study hopes to clarify the specific impact of

external environmental variables on the severity of

traffic accidents, explore the order of influence of

different high-risk conditions on accident severity,

and provide data support for traffic safety policy

making.

2 LITERATURE REVIEW

Traffic accidents can cause serious consequences

around the world, including property damage and

casualties. According to the World Health

Organization, approximately 1.35 million people lose

their lives in traffic accidents each year. (World

Health Organization, 2021). The occurrence of traffic

accidents is caused by the combined effects of

individual characteristics, driving behavior, external

environmental conditions, and macro factors. In

recent years, scholars have been conducting more and

more research on traffic accidents. Studies have

discovered that human factors are the main cause of

accidents, followed by vehicle failures and

environmental factors (Iqbal et al., 2020).

2.1 The Impact of Individual

Characteristics on Traffic

Accidents

Individual characteristics, as an important factor

affecting the severity of traffic accidents, reflect the

differences in drivers' perception, judgment level, and

reaction speed when facing emergencies. Variables

such as the driver's gender, age, and driving

experience determine their driving behavior patterns

and sensitivity to risks to a certain extent, thus

affecting the process and outcome of accidents.

Gender differences can affect how male and

female drivers respond to traffic accidents. A study

on driver-injury severities analyzes Florida crash data

and discovers that men and women have different

crash severity levels when they are under-adjusted for

speed, and that the influencing factors are time-

unstable (Islam & Mannering, 2021). Another study

investigates that male drivers are more susceptible to

negative emotions and adopt risky driving behaviors,

while female drivers are better able to regulate their

emotions and drive rationally (Khan et al., 2020).

Gender differences have a certain impact on drivers'

Exploring Determinants of Trafﬁc Accident Severity Using Empirical Data

673

psychology and cognition, which will affect their

judgment when encountering traffic accidents and

lead to differences in the severity of traffic accidents.

Studies have shown that the age of the driver also

has a significant impact on the severity of traffic

accidents (Haleem & Gan, 2013). The Mixed Logit

Model analysis shows that different age groups show

differences in accident types and impact directions

(Haleem & Gan, 2013). Middle-aged drivers are most

likely to be seriously injured from impacts from the

back, left, and right, while young drivers have a

higher risk of causing serious traffic accidents

(Haleem & Gan, 2013). Young drivers are more

likely to view dangerous driving as a challenge rather

than a risk, and this risk-taking tendency leads to an

increase in the severity of traffic accidents (Khan et

al., 2020).

Conventional wisdom holds that the more

experience a driver has, the lower their accident risk,

but some studies have found that experience does not

protect drivers. A study based on ordered logistic

regression of traffic accidents detects that

experienced drivers are more likely to cause serious

traffic accidents due to their lack of attention to rules

and overconfidence (Gebre Meles et al., 2022).

Variables such as private vehicles and drivers being

vehicle owners can reduce accident severity (Gebre

Meles et al., 2022).

2.2 The Impact of External

Environmental Conditions on

Traffic Accidents

Among the many factors that affect the severity of

traffic accidents, external environmental conditions

are a crucial variable, including weather conditions,

road lighting conditions, time, and more. These

factors will directly affect the driver's field of vision

and driving behavior. Although there are certain

correlations between these environmental variables,

for example, driving at night is often accompanied by

insufficient lighting, and bad weather may lead to

reduced visibility, they also have independent effects

on the severity of accidents through their respective

mechanisms.

Different weather conditions also have different

effects on the severity of traffic accidents. Taking the

UK as an example, an empirical study based on police

accident reports in England and Wales shows that the

frequency of traffic accidents increases on rainy days,

but the severity decreases (Edwards, 1998). There are

geographical differences in the severity of traffic

accidents in foggy weather, and the severity is

reduced in some areas, which may be related to the

"learning adaptation effect" of drivers to reduced

visibility (Edwards, 1998). Strong winds have no

significant impact on traffic accidents (Edwards,

1998). It can be seen that the weather will have

different effects on drivers' psychology, thus

affecting their driving.

A study in Fukuoka, Japan, discovers that road

structure also affects the type and severity of traffic

accidents (Dong et al., 2021). Car-car collision

accidents often occur at intersections, while car-

bicycle accidents are more concentrated at T-

intersections, especially when the stop line is set

improperly, such as moving back (Dong et al., 2021).

Whether there are traffic lights at the intersection and

whether there are vehicle direction restrictions, such

as left/right limits, also affect the probability and

severity of accidents (Dong et al., 2021).

An analysis of accidents on Pakistan’s M-2

highway reveals that the deadliest accidents occurred

at night, in the early morning, in dry weather, and on

straight sections (Iqbal et al., 2020). Especially in the

early morning hours, accidents are frequent and

serious due to dozing problems (Iqbal et al., 2020).

Accident severity also fluctuates by season and time

of week, with more accidents in July and on Sundays

(Iqbal et al., 2020).

2.3 Impact of Macro Factors on Traffic

Accidents

In addition to individual differences and

environmental conditions, some macro factors also

affect traffic accidents. A study based on fatal traffic

accident data in the United States observes that for

every one standard deviation drop in stock market

returns, there is an increase of about 0.6% in fatal car

accidents after the stock market opens (Giulietti et al.,

2020). Drivers' emotional reactions to stock market

changes can affect their driving behavior, leading to

high-risk driving (Giulietti et al., 2020). When the

stock market rises sharply, investors tend to become

excited, leading to distraction while driving; when the

stock market plummets, anxiety and stress may

trigger behavioral reactions such as impulsive driving

and decreased attention, thereby increasing the

probability of serious traffic accidents. Another study

verifies that although the overall traffic volume

(VMT) and total number of accidents decrease

significantly after the "stay-at-home order" is issued,

the single-vehicle accident rate and single-vehicle

fatal accident rate increase by 2.29 times and 4.10

times, respectively. (Doucette et al., 2021)。

With the development of science and technology,

autonomous driving technology has gradually

ICEML 2025 - International Conference on E-commerce and Modern Logistics

674

become popular in the automotive field. Research

shows that autonomous driving technology is still not

fully adaptable to situations such as bad weather,

hacker attacks, or communication interruptions

(Chougule et al., 2024). After an accident, due to

unclear division of responsibilities, drivers often find

it difficult to quickly take over the vehicle or respond

correctly at critical moments, which may aggravate

the severity of the accident (Chougule et al., 2024).

The government should be clearer about the accident

responsibility allocation mechanism, improve the

regulatory framework, and perfect the ethical

algorithm to effectively reduce the consequences of

traffic accidents.

3 RESEARCH HYPOTHESIS

To further explore the factors that affect the severity

of traffic accidents, this article utilizes Nashville

Accident data for research and analysis based on past

research results and logic, and makes the following

assumptions:

Hypothesis 1 (Time). Traffic accidents at night

and in the early morning are more serious than those

during the day. Traffic accidents on weekends are

more serious than those on weekdays.

Hypothesis 2 (Illumination). The worse the

lighting conditions, the more serious the traffic

accidents.

Hypothesis 3 (Weather). The worse the weather

conditions, the less severe the traffic accidents.

Hypothesis 4 (Collision Type). Different types of

collisions have an impact on the severity of traffic

accidents. Hit-and-run behavior increases the severity

of accidents.

4 METHODOLOGY

4.1 Data Description

The data contains information on many dimensions,

including the time and location of the accident,

weather conditions, collision type, number of

casualties, accident vehicles, lighting conditions, etc.

After splitting the "Date and Time" column and

encoding the categorical variables such as "Weather"

and "Collision Type", this article selected the

following columns for analysis:

Severity_Score:

This variable is a comprehensive indicator

constructed to quantify the severity of traffic

accidents. Its definition is as follows:

𝑆𝑒𝑣𝑒𝑟𝑖𝑡𝑦_𝑆𝑐𝑜𝑟𝑒 = 2 ×

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐹𝑎𝑡𝑎𝑙𝑖𝑡𝑖𝑒𝑠 + 1 ×

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝐼𝑛𝑗𝑢𝑟𝑖𝑒𝑠 + 0.5 ×

𝑃𝑟𝑜𝑝𝑒𝑟𝑡𝑦 𝐷𝑎𝑚𝑎𝑔𝑒

(1)

Since death is the most serious consequence of a

traffic accident, it is given the highest weight. Injury

is the second most serious consequence and is

weighted 1. Although property loss is important, it

has a lower weight than personal safety and is

weighted at 0.5.

Explanatory Variables: All explanatory variables

are shown in Table 1. In this study, missing values are

handled by filling in or deleting them. After the above

processing steps, 216,103 records are retained for

analysis in this study. The data covers the period from

January 2018 to April 2025, and is consistent and

very timely.

Table 1: Variable description

Name Definition

Hour hour of the day at which the accident occurred

Is_Weekend binary variable indicating whether the accident occurred on a weekend

Weekday the day of the week on which the accident occurred

Time_Period

categorical variable indicating the time-of-day segment in which the accident occurred

(morning, daytime, evening, late night)

Zip code ZIP code of the accident location

Weather weather condition at the time of the accident

IlluACCIDEmination lighting condition at the time of the accident

Collision Type type of collision involved in the accident

Number of Motor

Vehicles.

number of motor vehicles involved in the accident

Hit and Run binary variable indicating whether the accident was a hit-and-run

Exploring Determinants of Trafﬁc Accident Severity Using Empirical Data

675

4.2 Model Specification

4.2.1 Ordinary Least Squares (OLS)

To explore the impact of numerous variables on the

severity of traffic accidents, this study first utilizes

OLS as the basic framework to analyse the

approximate linear relationship between variables.

The model has a clear inference mechanism and

strong interpretative results, and is widely used in

empirical research fields such as social sciences and

traffic safety. The OLS model can directly measure

the impact of each explanatory variable on

Severity_Score, thereby drawing preliminary

conclusions. Before modelling, all variables were

processed for missing values to enhance the reliability

of the model.

To ensure the validity and robustness of the

estimation results of the regression model, this paper

conducts a Multicollinearity test on the selected

explanatory variables. Multicollinearity is a common

problem in regression analysis. When there is a high

correlation between independent variables, it will

lead to instability of the regression coefficient and

even distort statistical significance judgment, thus

affecting the interpretation of the actual effect of the

variable. Therefore, in the variable processing stage,

this paper uses the Variance Inflation Factor (VIF) to

systematically test all explanatory variables. The VIF

value generally reflects the degree of linear

correlation between a variable and other variable, and

a VIF greater than 10 is usually considered a warning

sign of severe multicollinearity. In this study, except

for the constant term, the VIF values of the variables

IlluACCIDEmination_1.0 (VIF=10.91),

IlluACCIDEmination_3.0 (VIF=8.56), and

Weather_21.0 (VIF=6.02) are relatively high, and

there is a risk of collinearity, so they are eliminated in

this paper. The VIF values of the remaining variables

are all within a reasonable range (VIF < 5) and could

be included in the regression analysis.

The model follows the formula:

𝑆𝑒𝑣𝑒𝑟𝑖𝑡𝑦_𝑆𝑐𝑜𝑟𝑒



=𝛼+𝛽





+𝛽





+⋯+

𝛽





+𝜀



(2)

Among them:



~Χ



: Respectively represent Hour,

Is_Weekend, Weekday, Time_Period, Zip code,

Weather, IlluACCIDEmination, Collision Type,

Number of Motor Vehicles, Hit and Run.

𝜀



: the error term

4.2.2 Nonlinear Model

Considering that there may be a nonlinear

relationship between the severity of the accident and

some variables, for example, the time of the accident,

lighting conditions, and casualties are often

asymmetric, the linear model has limitations in this

regard. Therefore, this paper further utilizes Random

Forest Regression for supplementary analysis. The

model can automatically fit complex nonlinear

structures and can reveal the implicit relationship

between variables through feature importance

analysis, thus making up for the shortcomings of

traditional linear methods. To reveal the marginal

effect and relative importance of each variable, this

paper uses the SHapley Additive exPlanations

(SHAP) method to interpret the model. As a powerful

visualization tool, the SHAP method can help

understand the output of the model and show the

impact of each feature on the prediction results.

5 RESULT ANALYSIS

5.1 Descriptive Statistical Analysis

Grouped descriptive statistics are performed on

several core variables under the three categorical

variables of Is_Weekend, Hit and Run, and

Time_Period. The results are shown in Table 2. In

terms of whether it is the weekend, the average

Severity Score on weekends is 0.52, which is higher

than the 0.44 on non-weekends, and the standard

deviation is larger, indicating that the severity of

weekend accidents fluctuates greatly. Hour,

Weekday, and Number of Motor Vehicles also have

different performances on weekends and non-

weekends. On the dimension of hit and run, the

average severity of hit-and-run cases was only 0.28,

which was significantly lower than the 0.52 of non-

hit-and-run cases, which may reflect that hit-and-run

is more common in minor accidents. However, its

standard deviation is relatively small, indicating that

the severity of this type of accident is more

concentrated. In terms of the time of the accident, the

severity is relatively higher at night and early

morning, which are 0.49 and 0.48, respectively, while

the severity is relatively lowest at daytime, which is

only 0.42. This may reflect that accidents are more

likely to cause more serious consequences at night

and in the early morning due to factors such as poor

visibility or driver fatigue. The average value of the

accident time is consistent with the time division

logic. For example, the average time of "Evening" is

18.61. In terms of the number of vehicles, the average

ICEML 2025 - International Conference on E-commerce and Modern Logistics

676

number is the highest in the evening (1.78 vehicles)

and the lowest in the early morning (1.50 vehicles),

indicating that early morning accidents may be more

concentrated in single-vehicle accidents.

Table 2: Summary statistics

Is_Weekend Hit and Run Time_Period

Yes No Yes No Daytime

Early

Morning

Evening

Night

Severity_Score

Mean 0.52 0.44 0.28 0.52 0.42 0.48 0.45

0.49

SD 0.87 0.77 0.58 0.86 0.76 0.79 0.79

0.83

Hour

Mean 11.44 14.02 11.76 13.93 13.30 5.78 18.61

12.91

SD 7.70 7.16 7.72 7.18 1.93 0.96 1.13

10.16

Weekday

Mean 5.48 2.06 3.12 2.86 2.62 3.52 2.88

2.96

SD 0.50 1.41 2.01 1.91 1.84 2.05 1.91

1.94

Number of

Motor

Vehicles

Mean 1.65 1.74 1.72 1.71 1.73 1.50 1.78

1.74

SD 0.85 0.82 0.84 0.82 0.83 0.85 0.79

0.82

5.2 Regression and Machine Learning

Model Analysis

5.2.1 Time

According to the results in Table 3, Time has a

significant impact on the severity of traffic accidents.

This article uses Daytime as the benchmark group.

The data in the table shows that compared with

daytime, early morning, evening, and night all have a

significant positive impact on Severity_Score, and the

corresponding t values are all over 6.8, and the p

values are all less than 0.001, which are highly

significant. This shows that traffic accidents

occurring in the early morning, evening, and night are

more serious, and traffic accidents occurring at night

are the most serious. This may be caused by fatigue,

and may also be related to night lighting and

visibility.

The results in Table 3 also verify that whether it is

a weekend or not also has an impact on the severity

of the accident. The Severity_Score for accidents

occurring on weekends increased by an average of

0.07 (t = 12.14, p < 0.001). This may be related to

behavioral factors such as the main purpose of driving

on weekends being leisure and entertainment, and the

driver's reduced risk perception. According to the

results in Figure 1, whether it is the weekend or not

has a greater impact on the severity than the specific

time of day. Therefore, Hypothesis 1 is established.

5.2.2 Illumination

The results in Table 3 show that lighting conditions

can also significantly affect the severity of traffic

accidents. The data show that the coefficients for the

three lighting conditions of dark (not lighted), dawn,

and dusk are all positive and statistically significant

(p < 0.001). The coefficients of Dark (not lighted),

dawn, and dusk are 0.63, 0.66, and 0.62, respectively,

and the t values are 49.08, 33.6, and 37.27. This may

be due to the delay in the driver's visual field

adaptation during light transition periods

(dawn/dusk) and darkness, coupled with increased

errors in environmental judgment, leading to higher

accident severity. According to Figure 1, the impact

of darkness (not lighted) on accidents is greater than

that of dawn and dusk. Therefore, Hypothesis 2 is

established.

5.2.3 Weather

The impact of weather on accident severity in the

model is relatively complex, and some variables are

significant. Snow, blowing snow, sleet, and hail all

harm accident severity, which means that the accident

severity is lower in this weather. The accident

severity is higher in foggy and cloudy weather. These

results indicate that when visibility is poor or the

weather changes suddenly, drivers may be slow to

react or fail to brake in time, which can aggravate the

consequences of an accident. However, when weather

conditions are more severe, such as snow, the severity

of the accident may be reduced due to the driver's

increased concentration. However, according to the

coefficients in Figure 1 and Table 3, compared with

other factors, weather has a smaller impact on

accidents.

Therefore, Hypothesis 3 is not true. The severity

of the accident shows different reactions under

different weather conditions.

Exploring Determinants of Trafﬁc Accident Severity Using Empirical Data

677

Table 3: Linear relationships between severity score and different variables

Coef. T-value P-value

Is_Weekend 0.07 12.14 0.000***

Weekday 0.00 -1.26 0.207

Hit and Run -0.25 -63.30 0.000***

Weather

Sleet, Hail -0.13 -2.35 0.019*

Snow -0.10 -4.12 0.000***

Fog 0.09 2.27 0.023*

Cloudy 0.06 6.85 0.000***

Blowing Snow -0.21 -2.48 0.013*

Illumination

Dark (Not Lighted) 0.63 49.08 0.000***

Dawn 0.66 33.60 0.000***

Dusk 0.62 37.27 0.000***

Collision Type

Rear End -1.02 -1.14 0.253

Head-on 0.37 32.21 0.000***

Rear to Rear -0.45 -17.96 0.000***

Angle -0.09 -16.29 0.000***

Sidewipe (Same Direction) -0.43 -65.88 0.000***

Sidewipe (Opposite

Direction

)

-0.26 -20.93 0.000***

Front to Rear -0.20 -34.99 0.000***

Rear to Side -0.50 -21.68 0.000***

Time Period

Early Morning 0.04 6.83 0.000***

Evening 0.05 9.29 0.000***

Night 0.06 9.29 0.000***

5.2.4 Collision Type

The regression results show in detail the impact of

different collision types on traffic accidents. Except

for head-on collision, which has a positive impact on

accident severity, other collision types, such as rear-

to-rear, angle, sideswipe, front-to-rear, and rear-to-

side, all harm accident severity. In addition, hit and

run also hurts accident severity, with a coefficient of

-0.25, a t value as high as -63.30, and a significance

level of 0.001. This result shows that after controlling

factors such as collision type, time, and lighting

conditions, the severity of a traffic accident will be

significantly reduced if there is a hit-and-run incident.

This conclusion may seem counterintuitive at first

glance, as it is generally believed that escaping

behavior means that the driver is more responsible

and may lead to more serious consequences.

However, combined with the actual background

analysis, this result is likely because hit-and-run often

occurs after a minor collision, with the perpetrator

attempting to evade financial compensation or legal

liability, rather than fleeing after causing major

casualties. Therefore, fleeing behavior is statistically

more strongly associated with less serious accidents.

The data in Figure 1 shows that hit and run, sideswipe

(same direction), and angle are the three variables that

have the greatest impact on Severity_Score.

Therefore, Hypothesis 4 does not hold. When a

hit-and-run occurs, the severity of the traffic accident

will decrease.

ICEML 2025 - International Conference on E-commerce and Modern Logistics

678

Figure 1: Nonlinear relationships between severity score and different variables (Picture credit : Original).

In the regression results of this study, the

coefficient values of multiple independent variables

are relatively small, but this does not mean that these

variables have no practical significance for the

severity of the accident. On the one hand, as a

dependent variable, the Severity Score has a limited

distribution range of its values. Therefore, even if the

regression coefficient is not large, as long as it is

statistically significant, it still shows that the marginal

impact of the variable is real. On the other hand, the

severity of an accident is often affected by multiple

factors. The marginal effect of a single variable is

limited, but its role in the overall model cannot be

ignored. This is also an important value of

multivariate regression analysis, which can integrate

multiple influences and reveal complex models.

6 CONCLUSION

This study systematically analyzed the impact of time

factors, weather conditions, lighting conditions, and

collision types on the severity of traffic accidents.

Through empirical analysis, this paper identifies that

early morning, evening, and night are the periods that

lead to more serious traffic accidents, while weekends

are more likely to have serious accidents than

weekdays. Weather conditions such as snow, blowing

snow and sleet, and hail will significantly reduce the

risk of accidents, while fog and cloudy weather will

aggravate the consequences of accidents. Lighting

conditions such as dark (not lighted), dawn, and dusk

significantly increase the severity of accidents.

Exploring Determinants of Trafﬁc Accident Severity Using Empirical Data

679

Moreover, head-on collisions will lead to increased

severity of accidents, while hit-and-run behaviour

will reduce accident severity. If all environmental

variables are considered together, the impact of

behavioural factors such as hit-and-run behaviour and

collision type on the severity of traffic accidents is

usually greater than that of environmental factors

such as weather and lighting. This is because the type

of collision directly determines the degree of physical

damage, while hit-and-run is related to the

psychology of the perpetrator, which often occurs

when the perpetrator wants to evade responsibility for

a minor accident. This has a more direct impact on the

consequences of the accident.

This study verified the significant relationship

between multiple environmental and behavioural

variables and accident severity through quantitative

analysis of actual traffic data, filling the gap of

insufficient comprehensiveness in previous studies.

This not only provides empirical support for traffic

safety researchers but also provides data-based

reference for urban traffic managers when

formulating precise strategies. Managers should

strengthen road inspections during specific weather

conditions or times and optimize infrastructure

construction, such as lighting equipment. However,

behavioural guidance, law enforcement supervision,

and traffic safety education are key strategic

directions to reduce the severity of traffic accidents.

The government can strengthen the publicity of traffic

rules and enhance legal publicity and education for

high-risk collision types.

Although this study conducts a comprehensive

quantitative analysis of environmental and individual

behavioural variables, it still fails to consider some

individual variables and macro factors more

comprehensively, and does not conduct a detailed

discussion and verification of the improvement

methods. In the future, studies can further consider

how traffic accidents will change under the combined

effects of macro variables (such as the epidemic) and

individual factors (such as gender and age), and verify

the specific feasibility of improvement measures such

as legal education.

REFERENCES

World Health Organization, 2021. Road traffic injuries.

World Health Organization.

https://www.who.int/news-room/fact-

sheets/detail/roadtraffic-injuries.

Khan, K., Zaidi, S. B., & Ali, A., 2020. Evaluating the

nature of distractive driving factors towards road traffic

accident. Civil Engineering Journal, 6(8), 1555-1580.

Gebre Meles, H., Brhanu Gebrehiwot, D., Gebrearegay, F.,

Gebru Wubet, G., & Gebregergis, T., 2022.

Identification of determinant factors for car accident

levels occurred in mekelle city, tigray, ethiopia:

Ordered logistic regression model approach. Momona

Ethiopian Journal of Science, 13(2), 225-239.

Edwards, J. B., 1998. The relationship between road

accident severity and recorded weather. Journal of

Safety Research, 29(4), 249-262.

Iqbal, A., Rehman, Z. U., Ali, S., Ullah, K., & Ghani, U.,

2020. Road traffic accident analysis and identification

of black spot locations on highway. Civil Engineering

Journal, 6(12), 2448-2456.

Giulietti, C., Tonin, M., & Vlassopoulos, M., 2020. When

the market drives you crazy: Stock market returns and

fatal car accidents. Journal of Health Economics, 70,

102245.

Islam, M., & Mannering, F., 2021. The role of gender and

temporal instability in driver-injury severities in

crashes caused by speeds too fast for conditions.

Accident Analysis and Prevention, 153, 106039.

Haleem, K., & Gan, A., 2013. Effect of driver's age and side

of impact on crash severity along urban freeways: A

mixed logit approach. Journal of Safety Research, 46,

67-76.

Dong, J., Hirota, M., Nomura, T., & Sato, J., 2021. Analysis

of crossing collision accident characteristics by

accident party. International Journal of Intelligent

Transportation Systems Research, 19(1), 214-229.

Doucette, M. L., Tucker, A., Auguste, M. E., Watkins, A.,

Green, C., Pereira, F. E., Borrup, K. T., Shapiro, D., &

Lapidus, G., 2021. Initial impact of COVID-19's stay-

at-home order on motor vehicle traffic and crash

patterns in connecticut: An interrupted time series

analysis. Injury Prevention, 27(1), 3-9.

Chougule, A., Chamola, V., Sam, A., Yu, F. R., & Sikdar,

B., 2024. A Comprehensive Review on Limitations of

Autonomous Driving and Its Impact on Accidents and

Collisions. IEEE Open Journal of Vehicular

Technology, 5, 142-161.

ICEML 2025 - International Conference on E-commerce and Modern Logistics

680