Accident Analysis and Prevention 59 (2013) 1825

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Risk factors associated with trafc violations and accident

severity in China
Guangnan Zhang a , Kelvin K.W. Yau b, , Guanghan Chen a
a
b

Center for Studies of Hong Kong, Macao and Pearl River Delta, Sun Yat-Sen University, Xingang Xi Road, Guangzhou, China
Department of Management Sciences, City University of Hong Kong, Tat Chee Avenue, Kowloon, Hong Kong

a r t i c l e

i n f o

Article history:
Received 21 August 2012
Received in revised form 18 January 2013
Accepted 8 May 2013
Keywords:
Trafc violations
Accident severity
Risk factors
Road safety
Transport facility improvement

a b s t r a c t
With the recent economic boom in China, vehicle volume and the number of trafc accident fatalities have
become the highest in the world. Meanwhile, trafc accidents have become the leading cause of death in
China. Systematically analyzing road safety data from different perspectives and applying empirical methods/implementing proper measures to reduce the fatality rate will be an urgent and challenging task for
China in the coming years. In this study, we analyze the trafc accident data for the period 20062010 in
Guangdong Province, China. These data, extracted from the Trafc Management Sector-Specic Incident
Case Data Report, are the only ofcially available and reliable source of trafc accident data (with a sample size >7000 per year). In particular, we focus on two outcome measures: trafc violations and accident
severity. Human, vehicle, road and environmental risk factors are considered. First, the results establish
the role of trafc violations as one of the major risks threatening road safety. An immediate implication
is: if the trafc violation rate could be reduced or controlled successfully, then the rate of serious injuries
and fatalities would be reduced accordingly. Second, specic risk factors associated with trafc violations
and accident severity are determined. Accordingly, to reduce trafc accident incidence and fatality rates,
measures such as trafc regulations and legislationtargeting different vehicle types/driver groups with
respect to the various human, vehicle and environment risk factorsare needed. Such measures could
include road safety programs for targeted driver groups, focused enforcement of trafc regulations and
road/transport facility improvements. Data analysis results arising from this study will shed lights on
the development of similar (adjusted) measures to reduce trafc violations and/or accident fatalities and
injuries, and to promote road safety in other regions.
2013 Elsevier Ltd. All rights reserved.

1. Introduction
Deaths due to road trafc accidents have become one of the
top three causes of death for populations aged 5 to 44 years, and
the direct economic loss has accounted for approximately 13%
of Chinas annual gross domestic product (GDP) (WHO, 2009). In
China, given the surging economy, vehicle volume and the number of trafc accident fatalities have become the highest in the
world. In 2010, road trafc accidents resulted in 65,225 deaths and
over 900 million RMB (equivalent to USD 143 million, exchange
rate: 1 USD = 6.3 RMB) in direct property losses (Trafc Management Research Institute, Ministry of Public Security, China, 2011
Road Accidents White Paper, 2011). Trafc accidents as a cause of
death are now the leading killer in China (Liu Jun, Trafc accidents
have become the No. 1 killer, Peoples Daily 2010-8-5). Given these

Corresponding author. Tel.: +852 3442 8585; fax: +852 3442 0189.
E-mail addresses: zsuzgn@hotmail.com (G. Zhang),
mskyau@cityu.edu.hk (K.K.W. Yau).
0001-4575/$ see front matter 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.aap.2013.05.004

alarming gures, and with the launching of the United Nations Road
Safety Action for the Decade, the State Council of China established
a goal of decreasing the trafc fatality rate by 36% in the recent
12th Five-Year national strategic plan. With such a severe situation in Chinas road trafc safety, there is an urgent need to study
the risk factors that determine trafc violations and trafc accident
severity.
Since Haddon (1968) introduced his famous Haddon matrix
model, in which he considered three sets of interactive factors,
namely human, vehicle and environment, before, during and after
a collision, a large number of studies have shown that some of the
most important factors associated with trafc accident incidence
and injuries/fatalities are driving behaviors such as speeding, drunk
driving, fatigue driving and not using protective devices. Moreover,
because driving behaviors are habits that form in accordance with
living standards and social culture, they naturally are related to
personal characteristics such as the drivers gender, age, education
level and income. Specically, Sabey and Taylor (1980) suggested
that drivers personal characteristics account for 95% of the factors that cause accidents. Miaou and Lum (1993) further suggested

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

that, among these personal characteristics, the most important factors are age and gender. Personal characteristics not only relate to
trafc accident incidence, but they are also associated with trafc violations. For example, Shinar et al. (2001) demonstrated that
seat belt usage is positively correlated with age and education
level. Drivers are more willing to limit their speed as age increases,
while such willingness diminishes as drivers education level and
income increase. Fosgerau (2005) found that driving speed is negatively correlated with drivers age and is positively correlated with
drivers income. Male drivers tend to drive faster than their female
counterparts. Drivers with single marital status have an average
driving speed that is slightly faster than that of married drivers.
In addition to the effect of personal characteristics, it has also been
found that as urbanization rates increase, the average driving speed
drops. Other studies have focused on more specic perspectives.
Factor et al. (2008) applied a logistic regression procedure to analyze Israeli data and concluded that the varied trafc accident risks
were due to different social habits and technological advancements.
Social and cultural characteristics are important determining factors affecting trafc safety no matter which social sector a driver
belongs to. They found that in Israel, Muslims, separated, widowed, male, young, low-skilled workers and low-educated drivers
tended to have higher accident rates. Yamamura (2008) used panel
data from Japans 46 counties to assess the impact of social capital, social structure and social norms on drivers driving attitudes.
He concluded that formal legal restrictions have nearly no impact
on dangerous driving behaviors as mandatory security checks make
drivers even less careful. However, informal restrictions can reduce
the occurrence of dangerous driving behaviors.
In sum, the literature generally concluded that male drivers usually exhibit a higher probability of committing trafc violations and
consequently have a higher risk of being involved in serious/fatal
trafc accidents (Holubowycz et al., 1994; Massie et al., 1995; Zhang
et al., 2000; Hayakawa et al., 2000; Valent et al., 2002; Yau, 2004;
Yau et al., 2006; Kim et al., 2008). Median-aged drivers have a
lower probability of becoming involved in serious injury/fatal accidents, in contrast with younger and older drivers (Cooper, 1990;
Holubowycz et al., 1994; Massie et al., 1995; Hayakawa et al., 2000;
Hijar et al., 2000; Norris et al., 2000; Zhang et al., 2000; Derrig et al.,
2002; Valent et al., 2002). Furthermore, Abdel-Aty and Radwan
(2000) applied descriptive statistics, conditional probability and
linear regression to analyze drunk-driving accidents in Florida, USA.
It was found that 2534 year-old drivers exhibited the highest accident rate. As drivers age decreased, male, white and local residents
demonstrated higher rates of accident occurrence. Hasselberg et al.
(2005) investigated the impact of socio-economic status on trafc
accidents for young drivers. Considering the importance of drivers
age and gender on trafc accidents, Bedard et al. (2002) indicated
that the implementation of specic safety measures is necessary.
Research studies have also suggested that road type, street light
condition and weather conditions are important factors that affect
the severity of vehicle trafc crashes (Hijar et al., 2000). However,
weather conditions, though they are important, are not a factor that
policymakers can control when attempting to reduce the risk of
severe trafc accidents. Nevertheless, an assessment of the impact
of weather conditions on trafc safety is valuable to road infrastructure designers and the motor insurance companies that set
premium rates. Yakovlev and Inden (2010) examined trafc fatalities in 48 American states from 1982 to 2006 and concluded that
weather indicators, such as temperature and hail, are the most
important factors that determine trafc safety.
Moreover, other factors including time of day and whether the
accident occurred on a weekend are found to be signicant in determining accident severity (Massie et al., 1995; Hijar et al., 2000;
Valent et al., 2002; Lam et al., 2003). For example, out-of-control
accidents often occur during the evening and nighttime (Laapotti

19

and Keskinen, 1998). Trafc accidents are more likely to happen

during heavy trafc volume (Abdel-Aty and Radwan, 2000). In addition, vehicle type and accident location are found to be important
factors that are signicantly associated with accident severity (AlGhamdi, 2002).
A review of statistical methodologies for analyzing crash-injury
data can be found in Savolainen et al. (2011). Characteristics
of crash-injury severity data and related issues such as underreporting of crashes, ordinal nature of injury severity data and
within-crash correlation are discussed. Traditional and emerging
statistical methods to analyze such data, ranging from the common binary outcome models to advanced hierarchical, mixed and
order logit/probit models are also described.
The road trafc accident database adopted in this study covers trafc accidents that occurred between 2006 and 2010 in
Guangdong Province, China. Located in the Southern part of Mainland China, the annual GDP growth (>10%) in Guangdong Province
was consecutively ranked rst in the country, since the economic
reform and opening-up policy in 1978. As at 2010, the Province has
a resident population of over 104 million, being the only province
having residents more than 100 million. Due to Guangdongs rapid
economic development and consequently the vehicle numbers
growth, trafc accident incidences are the highest among all 31
provinces in China. In 2010, the recorded trafc accidents, deaths
and injuries were 13.84%, 9.51% and 14.37% (relative to nationwide total) respectively. Having the accurate and reliable data
obtained from the Guangdong Provincial Security Department, this
study aims to determine the specic risk factors associated with
(i) trafc violations and (ii) trafc accident severity. Corresponding
measures such as trafc regulations and legislationrespectively
targeting different vehicle types/driver groups with respect to the
various human, vehicle and environment risk factorscan subsequently be devised. By assessing the impact of various risk
factors on trafc violations and accident severity using Guangdong
Province data, results arising from this study will shed lights on
the development of similar (adjusted) measures to reduce trafc
accident fatalities and injuries, and to promote road safety in other
provinces.
2. Materials and methods
2.1. Data
We analyze the trafc accident data for the period 20062010
in Guangdong Province, China. These data, obtained from Guangdong Provincial Security Department, are extracted from the
Trafc Management Sector-Specic Incident Case Data Report.
They are the only ofcially available source of trafc accident
data in China. Data were recorded and reported by the trafc
police on-scene who conducted assessments and provided feedback immediately to the headquarters of the Trafc Management
Department. The data contain 7722, 7511, 7459, 7261 and 7387
annual samples for each year from 2006 to 2010, respectively. Each
sample includes demographic information, injury severity, vehicle characteristics, road conditions, crash time and environmental
conditions.
Recent studies on road safety in China mainly rely on observational or survey data, to assess pedestrian safety (Liu et al., 2011;
Zhuang and Wu, 2011), helmet and seatbelt wearing usage (Routley
et al., 2008; Huang et al., 2011; Yu et al., 2011), and drivers attitude and behavior on road safety (Zhang et al., 2006; Huang et al.,
2008). Other studies have adopted regional trafc accident management data (Huang et al., 2008; Kong and Yang, 2010), yet not
on a large scale. To the best of our knowledge, current study is the
rst of its kind using the most comprehensive and ofcial database
(N = 37,340) to conduct trafc accident analysis in China.

20

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

To determine the risk factors associated with trafc violations

and injury severity, the dependent variables are set as follows: (a)
whether there was any trafc violation1 = yes and 0 = no; (b)
whether the accident resulted in a serious injury/fatality1 = yes
and 0 = no. As the outcome measure is either 0 or 1, a logistic
regression model can be used to estimate the effect of the risk factors on the probability of (i) trafc violation; and (ii) fatal/serious
injury in an accident. Although binary logit model is the most common method to analyze injury-severity outcomes, ordinal discrete
outcome model is a viable alternative when injury data are of ordinal type (Savolainen et al., 2011). In current study, for the ease of
interpretation of results and the sake of devising corresponding
local road safety programs, injury severity is classied into two
groups according to fatal/serious injury or not. The risk factors that
were considered include the following: human, vehicle, road and
environment.
2.2. Risk factors
2.2.1. Human factor
Drivers age and gender are considered to be potential risk factors. With reference to the typical driving habits in different age
groups and level of driving skills (Trafc Management Bureau, Ministry of Public Security, PRC, China Road Trafc Accidents Annual
Statistical Report 2011), driver age is divided into ve groups: 25,
2635, 3645, 4655, and 56 and driving experience is categorized into six groups: 2, 35, 610, 1115, 1620, and 21 years.
Although drivers education, income and social status are
expected to be potential factors that are associated with trafc
violations and accident severity, this information is generally not
recorded in the trafc accident database. Instead, related information indicated by the drivers hukou origin and occupation are
available. Two groups of the hukou household registration system
are examined: rural hukou and urban hukou. Occupations are classied as: military and police, general staff members, workers, civil
servants, the self-employed, farmers, migrant workers, the unemployed and other occupations.
2.2.2. Vehicle factor
Three major vehicle types, namely passenger vehicles, goods
vehicles and motorcycles, are included in the data. In addition,
indicator variables are generated according to the vehicles safety
status, vehicles overload condition, whether the vehicle had any
compulsory third party insurance and whether the vehicle was a
commercial vehicle.
2.2.3. Road factor
Types of trafc lanes can be divided into vehicle lanes,
shared-lanes and other lanes in accordance with their functional
classication. When considering types and grades, roads can be
divided into expressways, ordinary highways and urban highways.
Specically, ordinary highways include the rst-class highways,
second-class highways, third-class highways, fourth-class highways and fth-class highways; urban highways include urban
expressways, urban ordinary highways and urban other highways.
2.2.4. Environment factor
There are eight environmental factors: street-light condition,
weather conditions, visibility level, whether the accident occurred
on a weekend, whether it was a public holiday, time of day, season and year of the accident. Street-light condition is classied
into daylight, good street lighting at night and no street lighting at night. Weather condition (good = 0, bad = 1) and visibility
level (good = 0, bad = 1) are generated as indicator variables accordingly. Weekends are dened as: 17:00 Friday to 23:59 Sunday,
as it is anticipated that the occurrence of Friday trafc accidents

after 17:00 is similar to those occurring on Saturday and Sunday (MacLeod et al., 2011). Public holidays refer to the holidays
stipulated by the State Council of China (before 2007 public holidays in China included New Year, Chinese New Year, International
Labor Day and National Day. From 2008 onwards, three holidays
were added following the holiday reform in 2007; they are Qing
Ming Festival, Dragon Boat Festival and Mid-Autumn Festival.).
Time has been classied into six groups according to working
time pattern and peoples lifestyle in China: 00:0006:59 (midnight to dawn), 07:0008:59 (morning rush hours), 09:0011:59
(morning working hours), 12:0016:59 (afternoon working
hours), 17:0019:59 (afternoon rushing hours), and 20:0023:59
(nighttime). Seasons are dened, according to the Bureau of
Meteorology Category, China, as spring (MarchMay), summer (JuneSeptember), autumn (OctoberNovember) and winter
(DecemberFebruary).
2.3. Statistical data analysis
Contingency tables are constructed to assess the association
between risk factors and (i) trafc violation and (ii) injury severity.
Chi-square tests of independence are conducted with the level of
signicance at 5%.
To estimate the effect of different predictor variables on the
likelihood of the occurrence of (i) trafc violation and (ii) injury
severity, logistic regression analyses are conducted. We consider
the use of multivariate stepwise logistic regression to identify
signicant factors determining (i) trafc violation and (ii) injury
severity and to estimate the magnitude of adjusted odds ratios
(ORs) for each signicant factor while controlling for other confounding factors (Hosmer and Lemeshow, 1989). The adjusted ORs
of signicant factors and their 95% condence intervals (CIs) are
computed using a stepwise logistic regression model in which
all factors were initially included, and insignicant factors are
subsequently removed by the stepwise procedure. Entry and
removal probabilities for the stepwise procedure are both set
at 0.05.
3. Empirical results
Table 1 displays the chi-square statistics for assessing the
dependence between the predictor variables and (i) trafc violations and (ii) fatal/serious injury. To further assess the casual
relationship, via the stepwise logistic regression analysis, adjusted
odds ratios and the corresponding 95% CIs for each of the signicant
factors are displayed in Table 2.
Before going into the detailed data analysis results, preliminary ndings point to an important conclusion. Overall, among all
trafc accidents (N = 37,340) recorded, 74.34% of those involved
trafc violations. Separating the data according to whether trafc violation is involved (N = 27,759) or not (N = 9581), it is found
that the fatal/serious injury proportion occurring with trafc violations (41.38%) is signicantly higher than (p-value < 0.001) the
proportion of fatal/serious injuries occurring in the absence of
trafc violations (35.16%). The immediate implication is that: trafc violations associate positively with accident severity. Thus, for
xed trafc accident number, increased trafc violation events
will imply more severe trafc accident events. In this context,
it is even more alarming to note that the trafc violation proportion in on an increasing trend (2006 = 70.15%, 2007 = 71.84%,
2008 = 73.52%, 2009 = 77.76%, 2010 = 78.73%). Trafc violation
events are becoming the major latent threat to road safety in China.
If the trafc violation rate can be reduced or controlled successfully, the number of fatal/serious injury victims would be reduced
accordingly.

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

21

Table 1
Independence test and measure of association for trafc violations and accident severity in China.
Factors

Human factors
Drivers gender
Drivers age
Drivers driving experience
Drivers hukou origin
Drivers occupation
Vehicle factors
Vehicle type
Vehicle safety status
Vehicle overload condition
Whether the vehicle had any
compulsory third party insurance
Whether the vehicle was a
commercial vehicle
Road factors
Types of trafc lanes
Road type
Environmental factors
Street-light condition
Weather condition
Visibility level
Whether it was a weekend
Whether it was a public holiday
Time
Seasons
Years
Trafc violation
*

Trafc violations
(N = 37,340)
2 (P-value), degree of
freedom

Severe injury/fatality (trafc violations

group) (N = 27,759)
2 (P-value), degree of freedom

Severe injury/fatality (non-trafc

violations group) (N = 9581)
2 (P-value), degree of freedom

79.287* (<0.001), 1
23.163* (<0.001), 4
18.999* (0.002), 5
10.481* (0.001), 1
123.370* (<0.001), 8

142.362* (<0.001), 1
7.811 (0.099), 4
38.928* (<0.001), 5
350.314* (<0.001), 1
266.85* (<0.001), 8

80.350* (<0.001), 1
7.301(0.121), 4
23.412* (<0.001), 5
131.333* (<0.001), 1
89.228* (<0.001), 8

497.959* (<0.001), 2
228.656* (<0.001), 1
3.151 (0.076), 1
0.060(0.807), 1

644.248* (<0.001), 2
301.659* (<0.001), 1
260.959* (<0.001), 1
36.542* (<0.001), 1

224.911* (<0.001), 2
18.939* (<0.001), 1
139.716* (<0.001), 1
0.123(0.726), 1

41.403* (<0.001), 1

693.446* (<0.001), 1

267.816* (<0.001), 1

136.353* (<0.001), 1
378.502* (<0.001), 8

8.309* (0.004), 1
1300* (<0.001), 8

13.724* (<0.001), 1
532.426* (<0.001), 8

6.093* (0.048), 2
30.705* (<0.001), 1
19.42* (<0.001), 1
6.13* (0.013), 1
2.497(0.114), 1
4.820(0.306), 5
7.913* (0.048), 3
217.465* (<0.001), 4
1300.000* (<0.001), 5

577.599* (<0.001), 2
0.329(0.566), 1
63.118* (<0.001), 1
15.973* (<0.001), 1
13.429* (<0.001), 1
116.186* (<0.001), 5
37.248* (<0.001), 3
18.896* (0.001), 4
343.741* (<0.001), 5

311.027* (<0.001), 2
5.265* (0.022), 1
5.127* (0.024), 1
0.043(0.835), 1
0.314(0.575), 1
51.059* (<0.001), 5
16.535* (0.001), 3
8.002 (0.092), 4
75.721* (<0.001), 5

Signicant at 5% level.

3.1. Risk factors affecting trafc violations

As displayed in the second column of Table 1, many factors
are signicantly related to trafc violations. Among all personal
characteristics, a drivers gender is one of the most important factors. Benchmarking with the overall trafc violations proportion
(74.34%), female drivers trafc violations proportion (65.98%) is
much lower than male drivers (74.83%). Drivers age (in years)
and driving experience (in years) also demonstrate certain diversities. Referring to the specic groups trafc violation proportions,
young drivers, novice drivers and experienced drivers tend to have
a higher trafc violation probability. Also, rural hukou drivers are
less likely to commit trafc violations than urban hukou drivers.
For vehicle factors, vehicle type, vehicle safety status and
whether the vehicle was a commercial vehicle are signicantly
associated with trafc violations. Specically, relative to motorcycles (66.59%), both goods vehicles (77.56%) and passenger vehicles
(77.70%) exhibit a higher proportion of trafc violations. Moreover, unt vehicle safety status (87.61%), overload (75.81%) and
commercial vehicle (76.54%) exhibit a higher-than-average trafc
violation proportion. Environment and other factors such as street
light condition, weather condition, visibility level and whether it
was a weekend also signicantly relate to trafc violations. In particular, no street lighting at night (75.53%), good weather (75.02%),
bad visibility (77.39%) and weekends (75.17%) are found to exhibit
a higher trafc violation proportion.
Applying stepwise logistic regression analysis, factors having a
signicantly higher likelihood of trafc violations are identied.
Detailed results can be found in the second column of Table 2.
Results generally are consistent with above ndings using chisquare test and reviewing the corresponding group-specic trafc
violation proportions. Some highlights are as follows. Male drivers
and drivers with less than 6 years of driving experience tend to
have a higher risk of trafc violations. Rural hukou drivers have a

lower risk of trafc violations. Goods vehicles and passenger vehicle drivers, compared to motorcycle drivers, exhibit a higher risk
of trafc violation. Unt safety status and overload are also found
to be signicant factors relating to trafc violations. For environmental factors, street light condition, weather condition, visibility
level, whether it was a weekend, time of day and the year of the
accident are identied as signicant factors associated with trafc
violations. Specically, no street lighting, bad visibility, weekends
and the hours 7:008:59 and 17:0019:59 present a higher risk of
trafc violation.
Factors including the drivers age, whether the vehicle had
compulsory third party insurance, whether the vehicle was a commercial vehicle, whether it was a public holiday and the season
are found to be not signicant in the stepwise logistic regression
analysis when other confounding factors are being controlled.
3.2. Risk factors affecting injury severity
Further analysis has also been conducted to identify the impact
of the risk factors on accident severity, conditional on the presence
or absence of a trafc violation. The last two columns in Table 1 display the chi-square statistics for assessing the dependence between
the predictor variables and accident severity by trafc violation
and non-trafc violation groups respectively. Referring to the trafc violation group data (N = 27,759), the overall proportion of
severe (fatal/serious injury) accidents of 41.38% serves as a base
for comparison. Among all factors relating to the driver, gender
difference is clearly seen. In particular, female drivers involving in
severe accident proportion is 25.78%, which is much lower than
the male drivers group (42.18%). Higher severe accident proportion
for rural hukou offenders (51.12%) is also evident. For vehicle factors, goods vehicles (52.48%), unt safety status (59.63%), overload
(58.65%), no compulsory third party insurance (46.24%) and noncommercial vehicles (52.90%) demonstrate a higher proportion of

22

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

Table 2
Adjusted odds ratios (95% condence intervals) in stepwise logistic regression analysis of risk factors associating with trafc violations and accident severity in China.
Factors*

Trafc Violations

Drivers gender (base: female)

1.332(1.206, 1.470)
Male
Drivers age (base: 56)

25

2635
3645

4655

Drivers driving experience (base: 21)

2
1.207(1.130, 1.289)
35
1.105(1.039, 1.175)
610

1115

1620

Drivers hukou origin (base: urban hukou)

Rural hukou
0.795(0.749, 0.844)
Vehicle type (base: motor cycles)
Passenger vehicles
1.992(1.877, 2.114)
Goods vehicles
2.023(1.895, 2.160)
Vehicle safety status (base: t safety statue)
Unt safety statue
2.586(2.274, 2.940)
Vehicle overload condition (base: not overload)
Overload
1.248(1.128, 1.380)
Whether the vehicle had any compulsory third party insurance (base: no)
Yes

Whether the vehicle was a commercial vehicle (base: no)

Yes
Street-light condition (base: daylight)

Good street-lighting
1.197(1.110,1.291)
No street-lighting
Weather condition (base: good)
0.850(0.803, 0.899)
Bad weather
Visibility level (base: good)
1.097(1.006, 1.196)
Bad visibility
Whether it was a weekend (base: no)
Yes
1.065(1.011, 1.122)
Whether it was a public holiday (base: no)
Yes

Time (base: 9:0011:59)

00:006:59

07:008:59
1.186(1.087, 1.295)
12:0016:59

17:0019:59
1.108(1.035, 1.186)
20:0023:59

Seasons (base: autumn)

Spring

Summer

Winter
Years (base: 2006)

2007
1.111(1.042, 1.184)
2008
1.359(1.270, 1.453)
2009
1.471(1.375, 1.575)
2010
Trafc violations (base: other trafc violations behaviors)
NA
Speeding
NA
Drunk-driving
NA
Illegal meeting
NA
Not giving way according to stipulations
Model prediction measures
Percent correct
74.37
0.644
ROC
37,340
N
*

Serious Injury/Fatality
(trafc violations group)

Serious Injury/Fatality
(non-trafc violations group)

1.382(1.213, 1.574)

1.653(1.353, 2.019)

0.911(0.852, 0.973)

1.154(1.038, 1.283)
1.146(1.022, 1.285)

1.345(1.264, 1.432)

1.496(1.350, 1.657)

1.371(1.284, 1.464)

0.808(0.733, 0.891)

1.774(1.609, 1.956)

1.443(1.124, 1.851)

1.289(1.164, 1.427)

1.472(1.228, 1.765)

0.774(0.715, 0.837)

1.347(1.260,1.440)

1.480(1.334,1.641)

1.173(1.096,1.255)
1.536(1.414,1.667)

1.212(1.085,1.354)
1.806(1.581,2.063)

1.130(1.036, 1.233)

1.107(1.048, 1.169)

1.155(1.049, 1.272)

1.299(1.200, 1.406)

0.816(0.757, 0.879)

0.852(0.748, 0.971)

0.882(0.828, 0.940)
0.827(0.779, 0.877)

0.849(0.773, 0.933)

1.598(1.422, 1.795)
1.232(1.009, 1.505)
1.255(1.110, 1.419)
0.932(0.869, 0.999)

NA
NA
NA
NA

65.12
0.687
27,759

68.01
0.688
9581

For brevity, results for drivers occupation and road factors are omitted.

severe accidents. For environmental factors, no street lighting at

night (57.72%), bad visibility (48.42%), weekends (43.10%) and public holidays (45.22%) have a higher proportion of severe accidents.
Weather conditions, however, are found to be not signicantly
associated with accident severity.
Applying stepwise logistic regression analysis, in the third column of Table 2, the results broadly agree with above ndings using
chi-square test and reviewing the corresponding group-specic

severe accident proportions. Referring to the odds ratios and 95%

CI in Table 2 (third column), male trafc offenders are more likely
to be involved in fatal/serious accidents. Regarding driving experience, novice drivers (2 years of driving experience) exhibit a
lower risk of being involved in severe accidents. Relative to urban
hukou trafc offenders, rural hukou offenders have an increased
likelihood of being involved in fatal/serious injury accidents.
For vehicle factors, goods vehicles drivers who violated trafc

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

regulations have a higher risk of severe injury/fatal accidents compared to motorcycle drivers. Unt safety status, overload and
commercial vehicles demonstrate an increased risk of severe accidents. Compulsory third party insurance is not found to be a
signicant factor. For environmental factors, no street lighting at
night has the highest risk of leading to a severe accident. Good street
lighting at night exhibits median risk, while daylight presents the
lowest risk. Bad visibility, weekends and public holidays demonstrate a signicantly higher risk of severe accidents. Moreover,
midnight to dawn (00:006:59) is found to have a higher risk
of fatal/serious injury accidents, compared to 9:0016:00. However, accidents happening during evening rush hour 17:0019:59
have a decreased risk of accident severity. The autumn season,
compared with spring and summer, presents an increased risk of
fatal/serious injury accidents. In addition, it is worth emphasizing
that, among all trafc violations, speeding, drunk driving and illegal
meeting all have a signicantly higher risk of leading to fatal/serious
injury accidents. Drivers age, weather condition and the year are
not found to be signicant factors associated with accident injury
severity when other factors are controlled in the stepwise logistic
regression procedure.
In addition, risk factors determining injury severity are also
studied for accidents not involving trafc violations. The last column in Table 2 shows that the risk factors that impact injury
severity differ when trafc regulations are not violated. Interestingly, whether or not a trafc violation occurs makes a difference
in terms of the signicant risk factors determining accident severity and their corresponding effects on fatal/serious injury accident
occurrence. For example, in cases where trafc violations are not
present, 610 years of driving experience and 1115 years of
driving experience are more signicantly associated with injury
severity, while novice driver status (2 years of driving experience)
exhibits an insignicant association; passenger vehicles become
a signicant factor for injury severity, while goods vehicles are
an insignicant factor. Moreover, some factors that have previously been identied as signicant factors, including compulsory
third party insurance, weekends, public holidays, the midnight to
dawn hours and the spring season, are found to be not signicantly
associated with injury severity. Such ndings have important
implications for organizing road safety promotion activities for targeted road user groups as well as establishing relevant regulations
and policies to reduce trafc accident severity at a macro level.

4. Discussion
The results indicate that some risk factors are found to be signicant in the logistic regression analysis for both trafc violations
and injury severity. As a summary, males, goods vehicles, an unt
safety status, overload, no street lighting at night, bad visibility
and weekends not only exhibit an increased likelihood of being
associated with a trafc violations, but they also imply a higher
risk of accident severity. Novice drivers (2 years of driving experience) display a signicantly higher risk of trafc violations, yet
their risk on accident severity is signicantly lower than the other
categories of drivers. Drivers with 35 years of driving experience, passenger vehicles, bad weather and the morning rush hour
(7:008:59) display an increased risk of trafc violations, while
their association with a fatal/severe injury is not signicant. Furthermore, since 2007, trafc violation rates increased signicantly;
in fact, an increasing trend was observed, while the fatal/serious
injury accident rate remained rather stable. Commercial vehicles,
good street lighting at night, public holidays and mid-night to dawn
(00:0006:59) do not have a signicant effect on trafc violations,
but when trafc violations do occur, these factors confer a higher
risk of a fatal/serious injury. Nevertheless, compulsory third party

23

insurance and the spring and winter seasons do not demonstrate

signicant impacts on injury severity.
4.1. Human factor
Concurring to the ndings in the literature, male drivers are
clearly demonstrating a signicantly higher risk of trafc violations as well as accident severity (Holubowycz et al., 1994; Massie
et al., 1995; Hayakawa et al., 2000; Zhang et al., 2000; Valent et al.,
2002; Yau, 2004; Yau et al., 2006; Kim et al., 2008). Education level,
income and social status have been shown to be factors associating
with road safety (Hasselberg et al., 2005; Factor et al., 2008). However, information on these factors are not available in our data,
such effect may partly be transmitted to the hukou and occupation variables. Effect of drivers characteristic on trafc accident
severity is also veried (Sabey and Taylor, 1980). Most research
studies have identied the effect of age on various aspects of trafc
accident event and road safety (Cooper, 1990; Holubowycz et al.,
1994; Massie et al., 1995; Abdel-Aty and Radwan, 2000; Hayakawa
et al., 2000; Hijar et al., 2000; Norris et al., 2000; Zhang et al., 2000;
Bedard et al., 2002; Derrig et al., 2002; Valent et al., 2002; Fosgerau,
2005; Factor et al., 2008; Zhuang and Wu, 2011). In our study, the
effect of drivers age is not found to be signicant on both trafc
violations and accident severity. Instead, its impact is shifted to the
closely related driving experience variable in the stepwise logistic regression analysis, which may illuminate why Chinese drivers
concentrate more on driving skills and capabilities (Zhang et al.,
2006).
4.2. Vehicle and environment factors
It should be noted that goods vehicles exhibit signicantly
higher risk of both trafc violations and accident severity. In fact,
this nding supports the importance of vehicle type variable on
trafc violations in the literature (Hijar et al., 2000; Al-Ghamdi,
2002). Tightened vehicle safety and overload condition checking,
particularly for goods vehicles, is also crucial in reducing both trafc violations and accident severity rates. Consistent with the major
ndings in the literature, accidents occurring during weekend also
deserve particular attention (Massie et al., 1995; Hijar et al., 2000;
Valent et al., 2002; Lam et al., 2003). Effect of time and season on
trafc accidents appear to be region-specic. In some studies, trafc
accidents mostly occur at night time (Laapotti and Keskinen, 1998)
or during rush hours (Abdel-Aty and Radwan, 2000). Another study
nds that the occurrence of trafc accidents is quite evenly distributed over time of day, day of the week and season (Wang et al.,
2008). Our data reveal that morning rush hours (07:0008:59) only
increases the likelihood of trafc violations but has no signicant
effect on accident severity, while mid-night to dawn (00:000659)
increases the risk of fatal/severe injury accidents. Weather conditions, such as temperature and hail indicators, are found to be
important factors determining road accident fatality in the United
States (Yakovlev and Inden, 2010). Current study shows that bad
weather reduces the risk of trafc violations and has no signicant
effect on accident severity. On the other hand, bad visibility exerts
pressure on the risk on both trafc violations and accident severity.
4.3. Improving road safety
Trafc accidents are mostly regarded as unexpected, unfortunate events occurring at random. However, scientic analysis of
accident data and the implementation of relevant safety measures
could prevent trafc accident occurrence and reduce injury severity
(WHO, 2004). Current research has established that trafc violations in China are one of the major threats to road safety. It is
even more alarming to note that the trafc violation proportion

24

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

in on an increasing trend. The important implication is: if the

trafc violation rate could be reduced or controlled successfully,
then the rate of serious injuries and fatalities would be reduced
accordingly. Thus, given the current road trafc accident incidence
and fatality rates in China, trafc regulations and laws should be
enacted that target different vehicle types/driver groups and consider the various human, vehicle and environment risk factors. For
example, male drivers and drivers with certain occupation categories (such as farmers, general staff members, the self-employed,
migrant workers and other occupations) are considered to have a
higher risk of being involved in trafc violations and fatal/serious
injury accidents. Therefore, trafc safety propaganda and targeted
follow-up management should be strengthened. Goods vehicles
that are found with unt safety and overload status should face
more serious legal consequences. As no street lighting at night
and bad visibility are signicantly associated with trafc violations
and injury severity, the improvement of transport facilities would
have a clear benet to road safety. Additional measures to reduce
accident incidence and injury severity during weekends are also
needed.
Moreover, trafc safety depends on road design, road layout,
vehicle performance, trafc regulations and their implementation
effectiveness. However, trafc safety management measures in
developed countries may not be entirely applicable to developing
countries because in low-income countries the transport modes
primarily include walking, bicycling, riding motorcycles and taking buses, whereas the major road users in developed countries are
motor vehicle owners. Therefore, road usage patterns that are suitable for mixed trafc and local conditions need to be introduced and
improved, and the effectiveness of these methods in other developing countries should also be comprehensively studied. As a country
with the highest number of trafc accident fatalities in the world,
systematically analyzing road safety data from different perspectives and applying empirical methods/implement proper measures
to reduce the fatality rate will be urgent and challenging tasks for
China in the coming years.
4.4. Limitations and future research
As data are extracted from the Trafc Management SectorSpecic Incident Case Data Report from the Guangdong Provincial
Security Department, which is the only ofcial source of crosssectional trafc accident data available in the Province, data quality
can be assured. However, data are of the crash-injury severity type,
it inherits the general problem of under-reporting minor injury
and property damage only (PDO) accidents, and hence the possible
bias in the parameter estimates due to such under-reporting might
exist. Furthermore, issues such as omitted variable bias, endogeneity and within-crash correlation should be aware of when looking
into the results (Savolainen et al., 2011).
Trafc accident data from the Guangdong Province are adopted
in current study. Given that there are similarities in drivers driving
attitude and habits across different Provinces in China, Guangdong
Provinces experience, to an extent, can be considered as representative and typical, such as the effects of gender, driving experience,
hukou, overload, safety status and insurance on trafc violations
and accident severity. Generalization of the research ndings to
other regions is not totally direct, but requires some ltering and
minor modications. In order to have a comprehensive and comparative review on trafc accidents among other Provinces in China,
efforts are being made to collect the nationwide trafc accidents
data. Second, in current study, trafc violations type has not be
separated for further analysis. For example, speeding, drunk driving or fatigue driving each has its own signicant risk factors.
Future research work will look into type-specic trafc violations
events and assess their corresponding risk factors. Third, for some

accidents, the occurrence of the event may relate to the passenger

or some unexpected road incidences, but not the driver, our models are not able to take them into account as such information are
not recorded in the Trafc Management Incident Case Data Report.
Forth, our database does not include seatbelt/helmet wearing and
mobile phone usage information, though these factors are deemed
important, their effects cannot be assessed. Last, the response variable for accident severity is binary, with either fatal/severe injury
or not. However, some multiple vehicle accidents involving many
fatalities should be considered as super severe events, the associated risk factors might be rather unique and in such case an ordinal
regression model is needed. In our current data format, these events
cannot be distinguished.

Acknowledgements
The authors are grateful to an Associate Editor and two reviewers for very helpful comments on earlier version of the paper. This
research was supported in part by the National Natural Science
Foundation of China grant no. 71173244 We would also like to
thank Dr. Han Zhang of Sun Yat-Sen University and Mr. Zeyi Chen
of GuangDong Provincial Public Security Department for their constructive suggestions and advice.

References
Abdel-Aty, M.A., Radwan, A.E., 2000. Modeling trafc accident occurrence and
involvement. Accident Analysis and Prevention 32, 633642.
Al-Ghamdi, A.S., 2002. Using logistic regression to estimate the inuence of accident
factors on accident severity. Accident Analysis and Prevention 34, 729741.
Bedard, M., Guyatt, G.H., Stones, M.J., Hirdes, J.P., 2002. The independent contribution of driver, crash, and vehicle characteristics to driver fatalities. Accident
Analysis and Prevention 34, 717727.
Cooper, P.J., 1990. Differences in accident characteristics among elderly drivers and
between elderly and middle-aged drivers. Accident Analysis and Prevention 22,
499508.
Derrig, R.A., Segui-Gomez, M., Abtahi, A., Liu, L.L., 2002. The effect of population
safety belt usage rates on motor vehicle-related fatalities. Accident Analysis and
Prevention 34, 101110.
Factor, R., Mahalel, D., Yair, G., 2008. Inter-group differences in road-trafc crash
involvement. Accident Analysis and Prevention 40, 20002007.
Fosgerau, M., 2005. Speed and income. Journal of Transport Economics and Policy
39, 225240.
Haddon Jr., W., 1968. The Changing approach to the epidemiology, prevention, and
amelioration of trauma: the transition to approaches etiologically rather than
descriptively based. American Journal of Public Health 58, 14311438.
Hasselberg, M., Vaeza, M., Laamme, L., 2005. Socioeconomic aspects of the circumstances and consequences of car crashes among young adults. Social Science &
Medicine 60, 287295.
Hayakawa, H., Fischbeck, P.S., Fischhoff, B., 2000. Trafc accident statistics and risk
perceptions in Japan and the United States. Accident Analysis and Prevention
32, 827835.
Hijar, M., Carrillo, C., Flores, M., Anaya, R., Lopez, V., 2000. Risk factors in highway trafc accidents: a case control study. Accident Analysis and Prevention 32,
703709.
Holubowycz, O.T., Kloeden, C.N., McLean, A.J., 1994. Age, sex, and blood alcohol concentration of killed and injured drivers, riders, and passengers. Accident Analysis
and Prevention 26, 483492.
Hosmer, D.W., Lemeshow, S., 1989. Applied Logistic Regression. Wiley, New York.
Huang, Y.H., Rau, P.L.P., Zhang, B., Roetting, M., 2008. Chinese truck drivers attitudes
toward feedback by technology: a quantitative approach. Accident Analysis and
Prevention 40, 15531562.
Huang, Y.H., Zhang, W., Morphy, L., Shi, G., Lin, Y., 2011. Attitudes and behavior
of Chinese drivers regarding seatbelt use. Accident Analysis and Prevention 40,
889897.
Kim, K., Brunner, I.M., Yamashita, E., 2008. Modeling fault among accident-involved
pedestrians and motorists in Hawaii. Accident Analysis and Prevention 40,
20432049.
Kong, C., Yang, J., 2010. Logistic regression analysis of pedestrian casualty risk
in passenger vehicle collisions in China. Accident Analysis and Prevention 42,
987993.
Laapotti, S., Keskinen, E., 1998. Differences in fatal loss-of-control accidents between
young male and female drivers. Accident Analysis and Prevention 30, 435442.
Lam, L.T., Norton, R., Woodward, M., Connor, J., Ameratunga, S., 2003. Passenger
carriage and car crash injury: a comparison between younger and older drivers.
Accident Analysis and Prevention 35, 861867.

G. Zhang et al. / Accident Analysis and Prevention 59 (2013) 1825

Liu, P., Huang, J., Wang, W., Xu, C., 2011. Effects of transverse rumble strips on
safety of pedestrian crosswalks on rural roads in China. Accident Analysis and
Prevention 43, 19471954.
MacLeod, K.E., Griswold, J.B., Arnold, L.S., Ragland, D.R., 2011. Factors associated with
hit-and-run pedestrian fatalities and driver identication. Accident Analysis and
Prevention 45, 366372.
Massie, D.L., Campbell, K.L., Williams, A.F., 1995. Trafc accident involvement rates
by driver age and gender. Accident Analysis and Prevention 27, 7387.
Miaou, S.P., Lum, H., 1993. Modeling vehicle accidents and highway geometric
design relationships. Accident Analysis and Prevention 25, 689709.
Norris, F.H., Matthews, B.A., Riad, J.K., 2000. Characterological, situational, and
behavioral risk factors for motor vehicle accidents: a prospective examination.
Accident Analysis and Prevention 32, 505515.
Routley, V., Ozanne-Smith, J., Li, D., Yu, M., Wang, J., Zhang, J., Tong, Z., Wu, M., Wang,
P., Qin, Y., 2008. China belting up or down? Seat belt wearing trends in Nanjing
and Zhoushan. Accident Analysis and Prevention 40, 18501858.
Sabey, B.E., Taylor, H., 1980. The known risks we run: the highway. In: Report SR567.
Transport Research Laboratory, Crowthorn, Berks.
Savolainen, P.T., Mannering, F.L., Lord, D., Quddus, M.A., 2011. The statistical analysis
of highway crash-injury severities: a review and assessment of methodological
alternatives. Accident Analysis and Prevention 43, 16661676.
Shinar, D., Schechtman, E., Compton, R., 2001. Self-reports of safe driving behaviors in relationship to sex, age, education and income in the US adult driving
population. Accident Analysis and Prevention 33, 111116.
Valent, F., Schiava, F., Savonitto, C., Gallo, T., Brusaferro, S., Barbone, F., 2002. Risk
factors for fatal road trafc accidents in Udine, Italy. Accident Analysis and Prevention 34, 7184.

25

Wang, Y., Hasseberg, M., Wu, Z., Laamme, L., 2008. Distribution and characteristics of road trafc crashes in the Chaoyang District of Beijing, China. Accident
Analysis and Prevention 40, 334340.
WHO, 2004. World Report on Road Trafc Injury Prevention. World Health Organization.
World Health Organization, 2009. Global Status Report On Road Safety: Time for
Action. World Health Organization.
Yakovlev, P.A., Inden, M., 2010.
Mind the weather: a panel data analysis of time-invariant factors and trafc fatalities. Economics Bulletin 30,
26852696.
Yamamura, E., 2008. Impact of formal and informal deterrents on driving behavior.
Journal of Socio-Economics 37, 25052512.
Yau, K.K.W., 2004. Risk factors affecting the severity of single vehicle trafc accidents
in Hong Kong. Accident Analysis and Prevention 36, 333340.
Yau, K.K.W., Lo, H.P., Fung, S.H.H., 2006. Multiple-vehicle trafc accidents in Hong
Kong. Accident Analysis and Prevention 38, 11571161.
Yu, X., Ke, L., Ivers, R., Du, W., Senserrick, T., 2011. Prevalence rates of helmet use
among motorcycle riders in a developed region in China. Accident Analysis and
Prevention 43, 214219.
Zhang, J., Lindsay, J., Clarke, K., Robbins, G., Mao, Y., 2000. Factors affecting the
severity of motor vehicle trafc crashes involving elderly drivers in Ontario.
Accident Analysis and Prevention 32, 117125.
Zhang, W., Huang, Y.H., Roetting, M., Wang, Y., Wei, H., 2006. Drivers views and
behaviors about safety in Chinawhat do they NOT know about driving? Accident Analysis and Prevention 38, 2227.
Zhuang, X., Wu, C., 2011. Pedestrians crossing behaviors and safety at unmarked
roadway in China. Accident Analysis and Prevention 43, 19271936.