Available via license: CC BY 3.0
Content may be subject to copyright.
IOP Conference Series: Materials Science and Engineering
PAPER • OPEN ACCESS
An Empirical Research on the Effect of Driver Training Industry on Road
Traffic Safety
To cite this article: Ni Sun et al 2019 IOP Conf. Ser.: Mater. Sci. Eng. 688 055010
View the article online for updates and enhancements.
This content was downloaded from IP address 158.46.129.190 on 06/12/2019 at 17:18
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
1
An Empirical Research on the Effect of Driver Training
Industry on Road Traffic Safety
Ni Sun1*, Lei Feng1, Ruqing Wei1
1Government Management Innovation Standardization Research Center, China
National Institute of Standardization, Haidian District, Beijing, 100191, China
*Corresponding author’s e-mail: sunni2006@163.com
Abstract. The 12 th Five-Year Plan of Road Traffic Safety and the 13th Five-Year Plan of Road
Traffic Safety have been issued successively, demonstrating the CPC and the Chinese
government has attached great importance to the driver training industry and road traffic safety.
However, relevant studies in China are still very limited. Based on the cross-sectional data of
100 cities in 2015, this paper analyzes the impact of driver training quality in this industry on
road traffic safety. The conclusion shows that the correlation between the passing rate of
subject 2 and road safety is stable, as high passing rate can lead to the effective reduction of the
occurrence of traffic accidents. The passing rate of Subject 1 and road traffic safety shows a
positive correlation in the period when the section is relatively static, and a negative correlation
in the panel analysis with timing characteristics.
1. Introduction
According to a report of the World Health Organization (WHO) in May 2017, about 1.25 million
people die from road traffic crashes every year, half of whom are "vulnerable road users", namely
pedestrians, cyclists and motorcyclists. About 20 to 50 million people suffer non-fatal injuries, many
of them disabled. And the costs of road traffic crashes account for 3 percent of the gross domestic
product of most countries. Without sustained action, road traffic accidents are expected to become the
seventh leading cause of death in the world by 2030.
With the development of China's economy, road safety issues are increasingly attracting people's
attention. In the road traffic system consisting of "people, vehicles, roads and environment", the
human factor is the most critical, serving as the bridge connecting other elements. In the modern road
traffic, the drivers’ skills and awareness of road safety are of great importance to road traffic, since
they are the ones who comprehend the system, give the instructions and conduct the operation. And
the driver training industry, which cultivates and trains a driver’s series of skills and habits, is deemed
as a "source" of road safety management. It can be said that as the driver training industry is directly
involved in the efforts to promote road traffic safety, and the industry is closely related to the road
safety issues, the development of the industry comes with more social responsibilities.
At present, the number of motor vehicle driving license holders has exceeded 300 million in China,
and the number will grow by over 20 million annually in the next 10 years1. With the surge in the
number, a problem which can’t be overlooked is the various levels of individual driving skills and
road safety awareness, attributed to the various levels in the quality of driver training. Good training
1Source of the data: The First International Forum on Driver Training and Road Traffic Safety held in June 2016
in Beijing
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
2
can help beginners exercise their driving skills with dexterity in complex situations and bear in mind
the idea of road traffic safety whenever they are driving, while bad training would lead to the increase
of the probability of road traffic accidents by producing more “road killers”. When it comes to driving
schools, the quality of training is of great importance. Therefore, it can be said that the development of
driver training industry is closely related to road traffic safety. At present, most studies on the
correlation between driver training industry and road traffic safety in China are statistical research, and
there are few researches that go deep into the quantitative and empirical level, leaving much blank in
this field. In view of this, this paper intends to take the driver training quality as the focal point to
analyze the impact of this industry on road traffic safety from the quantitative and empirical
perspectives. It is an important supplement to the theory of driver training industry, and can also lay a
foundation for the future policy making concerning driver training and the development of this
industry, providing huge theoretical and practical significance.
2. Current Research Status at Home and Abroad
There are abundant studies on the correlation between the driving industry and road traffic safety
abroad. O'Malley P M and Johnston LD (1999), through the survey data of 1984-1997 of the US by
using Logistic regression to study the demographic factors (gender, region, average grade, population
density, family education and race) and lifestyle factors (religious commitment, high school
graduation, truancy, illegal drug abuse, night travel mileage per week, and weekly travel mileage)
effects on drunk driving behavior of American high school students. It found a significant decline in
teen drunk driving rates from the mid-1980s to the early 1990s or mid-1990s, but the decline did not
continue. White male high school graduates living in the western and northeastern United States and
the rural areas have a higher proportion of drunk driving. Truancy, the number of night trips, illegal
drug abuse, and drunk driving behavior had significant positive correlations with the number of miles
traveled per week, and the average grade and religious commitment were negatively correlated [1].
Eensoo D, Paaver M and Harro J (2011) studied the impact of interventions in driving school training
on road traffic safety by dividing 1889 novice students into a control group and an intervention group.
After three years of tracking, it was found that the drunk driving behavior of the control group was
significantly higher than that of the intervention group, which indicated that psychological
intervention can effectively reduce road traffic accidents caused by drunk driving behavior, but for
some drivers with high impulsive tendency and long-term drinking behavior, short-term psychological
intervention didn’t work[2]. Chakrabarty, Neelima Shukla, Anuradha Singh, H. Shokeen and Nancy
(2012) argued that inappropriate driving behavior is one of the main causes of road traffic accidents in
India. Traffic accidents increased as a result of various insecure behaviors of Indian drivers such as
lack of lane discipline, disregard of traffic regulations, frequent traffic violations, and self-centered
driving. Therefore, improving driver behavior can be an effective measure to reduce road traffic
accidents [3]. Živković S, Nikolić V, and Markič M (2015) studied the impact of drivers’ personal
characteristics on road traffic safety based on a questionnaire of 59 professional drivers. It was found
that the individual characteristics of professional drivers are very important for road traffic safety [4].
Kureckova V, Gabrhel V, Zamecnik P, et al (2017) designed a 16-hour course based on experienced
first-aid and compared its effects to standard 4-hour training. The results show that there are
significant differences in the effects of the two sets of training, which was not only shown in the first-
aid knowledge and skills, but also in the ability to cope under simulated conditions. Therefore
strengthening the acquiring of first-aid knowledge and skills as well as the improvement of
psychological quality should be an important part of the driving course [5].
There are relatively few studies on the driver training industry and road traffic safety in China.
Most of the existing researches are targeted at single object, and are qualitative analysis. Liu Junli
(2013) believes that the current training industry is ill-positioned, and driving schools should ensure
the quality of training and take on their due social responsibility while pursuing business functions [6].
Zhang Luo (2013) believes that the current drivers are not skilled or even qualified, which is a major
reason for the serious road safety challenges in China. Based on the current status of driving training
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
3
industry in Changsha, Zhang analyzed the five major aspects that make it difficult to improve training
quality: laws and regulations, coaching quality, supply and demand, driving school system, and cost
management mode, and gave relevant suggestions [7]. Liu Guangping, Ding Limin and Wang Jinfeng
(2015) clarified the necessity of traffic safety publicity and education in a society with high level of
automobile prevalence from the perspective of traffic safety publicity and education and believes that
traffic safety education is not only conducive to road traffic safety and smoothness, but also important
to the development of the national safety culture, the building of the rule consciousness among
nationals, and the improvement of legal awareness[8]. Cai Yang and He Jinjiang (2016) constructed a
road traffic accident probability scorecard using logistic regression analysis method based on big data
analysis, analyzing the driver's gender, age, driving age, vehicle status, driving school and other
characteristic variables, so as to predict the traffic accidents.
Comparing domestic and foreign research, it can be found that foreign research has been relatively
mature with diverse methods and rich results, while domestic research is still in its infancy with few
results and mostly focusing on qualitative analysis. It can be said that the research on driver training
industry and road traffic safety in the new stage in China is still to be fully rolled out. Especially with
the issuing of the 13th Five-Year Plan on Road Traffic Safety relevant research is in urgent need.
Next, the paper studies the impact of driver training industry on road traffic safety from the
perspective of the training quality. Abstractly, the industry can be regarded as the carrier of the
training quality, which is the direct embodiment of the driver training industry, and connects the
industry and road traffic safety as a medium. This paper follows the above ideas to study the impact of
the industry on road traffic safety, as shown in Figure 1.
Figure 1 Chart of Relations between Driving Training Industry and Road Traffic Safety
3. Empirical Analysis of the Correlation between Driving Training Quality and Road Traffic
Safety
Before the analysis, an issue that needs to be explained is that the records of road traffic accidents in
various regions of China have not been standardized for a long period of time, which leads to certain
difficulty in obtaining data for theoretical research. Therefore, the analysis mainly focuses on the
cross-sectional data and takes into account the introduction of control variables, striving to
comprehensively explore the impact of driver training quality on traffic safety.
3.1. Model Introduction and Variable Selection
3.1.1. Sectional Model
Based on the availability of data and the purpose of simplified analysis, this paper selects the number
of traffic accidents in 100 cities in 2015 as the research object in the cross-sectional analysis, and
constructs the cross-sectional model as follows:
( 1, , )
i i i i
y x z i n
= + + =
(1)
influences
carrier
reflect
Driving
training
industry
Quality
of
driving
training
Traffic
safety
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
4
Among them, ‘y’ represents road traffic safety, ‘i’ represents individual cities, ‘n’ represents
number of cities, ‘
i
x
’ represents main explanatory variables , and ’
i
z
’ represents introduced control
variables.
3.1.2. Quantile Regression
The paper adds quantile regression on the basis of the above regression. The quantile regression can
describe the whole situation of the conditional distribution of the explained variables in more detail,
not just the expectation of the explained variables in the conventional analysis. The regression
coefficient estimates are often different under different quantiles. Assume the general quantile
()
q
yx
of the conditional distribution
|yx
is a linear function of
x
:
()
q i i q
y x x
=
(2)
In the formula, ‘x’ represents the explanatory variable containing the control variable,
q
is called
the ‘quantile regression coefficient’, and its estimated value can be defined by the following
minimization problem:
::
min | | (1 )| |
i i q i i q
q
nn
i i q i i q
i y x i y x
q y x q y x
− + − −
(3)
3.1.3. Variable Selection
(1) Explained Variables: Road Traffic Safety
Road traffic safety means when people carry out activities on the road, they should drive and walk
safely in accordance with the provisions of traffic regulations to avoid personal casualties or property
losses. This paper selects the number of traffic accidents (tranum) in 100 cities in 2015 as a proxy
variable to measure the traffic safety situation in these regions. Most of the data can be found in the
statistical yearbooks of different provinces in 2016, but a small amount of data still needs to be found
in the statistical yearbooks of different regions.
(2) Core Explanatory Variable: Driver Training Quality
At present, there are no specific indicators to measure the quality of driving schools in China. In
general, the quality of driving training is closely related to the infrastructure and teaching staff of
driving schools. It measures how easy students could pass the driving test after the training and their
ability to deal with emergencies on the road later on. This paper selects the average s1 and s2 passing
rates of all driving schools in the region as the proxy variable to measure the training quality of the
region, reflecting the training competence of the city in that year. The general logic is as follows: in
the case of a certain level of difficulty of the test, the higher the passing rates of all driving school
students in a certain region, the higher the average training quality of driving schools in that region is,
i.e., the higher the quality of driver training in that region is. The data of the passing rates of subject 1
and subject 2 was obtained from the Traffic Safety Integrated Service and Management Platform of
the Ministry of Public Security, which includes 87,768 data from 3,657 driving schools in 100 cities.
The mean value was taken to obtain the regional driving quality data. Table 1 shows the number and
data of driving schools in different regions used in the cross-sectional analysis in this paper.
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
5
Table 1. Source of Driving School Data
Region
Main Cities
Number of
Cities
Number of
Driving Schools
Number
of Data
East China
Shanghai
1
356
4272
Ningbo
1
72
864
Anhui Province: Hefei, Wuhu,
Bengbu, Huainan, Ma’anshan,
Huaibei, Tongling, Anqing,
Huangshan, Chuzhou, Fuyang, Lu’an,
Bozhou, Chizhou, and Xuancheng
15
367
4404
Jiangxi Province: Nanchang,
Jingdezhen, Pingxiang, Jiujiang,
Xinyu, Yingtan, Ganzhou, Ji’an,
Yichun, Fuzhou, and Shangrao
11
282
3384
South China
Guangzhou
1
88
1056
Hainan Province: Haikou and Sanya
2
39
468
North China
Beijing
1
111
1332
Tianjin
1
135
1620
Taiyang
1
48
576
Shandong Province: Jinan, Qingdao,
Zibo, Zaozhuang, Dongying, Yantai,
Weifang, Jining, Tai’an, Weihai,
Rizhao, Laiwu, Linyi, Dezhou,
Liaocheng, Binzhou, and Heze
17
451
5412
Inner Mongolia Autonomous Region:
Hohhot, Baotou, Wuhai, Chifeng,
Tongliao, Erdos, Hulun Buir,
Bayannur, Ulanqab, Hinggan League,
Xilin Gol League, and Alxa League
12
485
5820
Central
China
Wuhan
1
93
1116
Southwest
China
Chongqing
1
366
4392
Guiyang
1
11
132
Sichuan Province: Chengdu, Zigong,
Panzhihua, Luzhou, Deyang,
Guangyuan, Suining, Neijiang,
Leshan, Nanchong, Meishan, Yibin,
Guang’an, Dazhou, Ya’an, Bazhong,
Ziyang, Aba Tibetan and Qiang
Autonomous Prefecture, Ganzi
Tibetan Autonomous Prefecture, and
Liangshan Yi Autonomous Prefecture
20
320
3840
Northeast
China
Shenyang
1
75
900
Harbin
1
91
1092
Northwest
China
Yinchuan
1
12
144
Shaanxi Province: Xi’an, Tongchuan,
Baoji, Xianyang, Weinan, Yan’an,
Hanzhong, Yulin, Ankang, Shangluo,
and Yangling Demonstration Zone
11
255
3060
Total
100
3657
43884
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
6
(3) Control Variables
The population factors, road factors and economic factors are considered in the selection of control
variables.
The population factors include urban population density (citymidu) and number of city travellers
(citytraveller). In theory, when the urban population density and number of city travellers grow, it is
prone to traffic jams and the probability of traffic accidents increases, if the road area and mileage
remains unchanged.
Road factors include the per capita area of city road (cityroad) and the number of raod
infrastructure streetlights (light). The per capita area of city road can measure the convenience of the
road traffic in an area to a certain extent. The larger the per capita area of city road is, the better the
traffic will be, and the lower the possibility of road traffic accidents is. Meanwhile, this paper
considers the impact of the road infrastructure streetlights on traffic accidents from the perspective of
the driver's sight. Obviously, when a driver is driving at night or is affected by extreme weather, and is
therefore prone to traffic accidents due to the blocked sight, the streetlights will improve the visibility
of the driver so that the probability of traffic accidents will be reduced. Of course, the above variables
also have the possibility leading to the increase of traffic accidents, because smooth traffic and clear
sight may increase the number of passengers and vehicles, leading to more traffic accidents.
When it comes to the economic factors, the per capita GDP (agdp) is selected as a proxy variable.
Generally speaking, the higher a region’s GDP, the more frequent its traffic will be, which may lead to
more traffic accidents.
In terms of data sources, except for the city traveller data (citytraveller) from the statistical
yearbooks of 2016 of different regions, all other data are from the China City Statistical Yearbook
2016. Table 2 shows the descriptive statistical characteristics of each variable.
Table 2. Descriptive Statistics
Variable
Observed
value
Unit
Mean
Value
Standard
Deviation
Minimum
Maximum
Number of Traffic Accidents
(tranum)
100
/Year
706.7200
947.6662
12
5358
Passing Rate of Subject 1
(S1)
100
%/Year
78.5179
6.6040
62.71
93.65
Passing Rate of Subject 2
(S2)
100
%/Year
45.7727
7.4146
26.47
68.99
Urban Population Density
(citymidu)
100
persons/
km2
3072.443
0
2162.350
0
8
1136
Number of City Travelers
(citytraveller)
100
10,000
persons/
Year
8426.665
0
19719.76
00
15.2247
156957
Per capita Area of City Road
(cityroad)
100
m2/Year
19.3153
9.1555
2.8
56.34
Number of Streetlights
(light)
100
Number/
Year
81049.14
00
96981.70
00
4280
532032
Per capita GDP (agdp)
100
Yuan/
Year
56471.87
00
34070.49
00
15076
207163
3.2. Cross-sectional Analysis
Before the model estimation, this paper first draws the linear trend graphs showing the correlation
between the number of traffic accidents (tranum) and the passing rates of subject 1 (s1) and subject 2
(s2) according to the distribution of cross-sectional data.
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
7
Figure 2: tranum and S1, S2
As shown in the above figure, the tranum is positively correlated with s1 and negatively correlated
with s2. The specific regression results are shown in Table 2.
Table 3. Cross-sectional Analysis Results of 100 Cities in China
Variable
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Model 7
Model 8
Constant
Terms
-1993.0200
(1202.6710)
-1993.0200
(1204.7590)
-2112.4960*
(1264.3450)
-2112.4960
(1276.7830)
-2807.8770**
(1180.0590)
-2807.877**
(1095.0120)
-2114.5420**
(975.9632)
-2114.5420*
(1107.226)
s1
44.5708***
(15.1695)
44.5708***
(13.5480)
48.9392***
(16.2694)
48.9392***
(14.8202)
50.3909***
(15.0243)
50.3909***
(12.8840)
31.3122**
(12.6953)
31.3122***
(11.3591)
s2
-24.9975*
(12.8524)
-24.9975
(13.7520)
-26.2851*
(13.3317)
-26.2851*
(13.9195)
-20.9512*
(12.3795)
-20.9512*
(12.6106)
-8.9063
(10.3444)
-8.9063
(10.0495)
citytraveller
0.0128***
(0.0046)
0.0128
(0.0108)
0.0124**
(0.0047)
0.0124
(0.0105)
0.0105**
(0.0044)
0.0105
(0.0087)
0.0032
(0.0038)
0.0032
(0.0051)
citymidu
0.0829*
(0.0432)
0.0829
(0.0506)
0.0838*
(0.0468)
0.0838
(0.0545)
0.0662
(0.0434)
0.0662
(0.0482)
0.0406
(0.0359)
0.0406
(0.0458)
cityroad
-8.9026
(10.6280)
-8.9026
(7.8923)
-19.4875*
(10.1607)
-19.4875*
(10.2407)
0.2787
(8.8894)
0.2787
(6.6635)
agdp
0.0108***
(0.0027)
0.0108***
(0.0034)
0.0024
(0.0026)
0.0024
(0.0026)
light
0.0062***
(0.0009)
0.0062***
(0.0022)
White-sta
32.4800***
32.4800***
32.6400**
32.6400**
45.5300**
45.5300**
77.2400***
77.2400***
F-sta
4.8600***
4.8600***
3.7900***
3.7900***
6.3800***
6.3800***
14.1300***
14.1300***
Note: "***", "**", and "*" correspond respectively to the significance level of p<0.01, <0.05, and <0.1, and
the value in the brackets are standard error.
Models 1, 3, 5, and 7 in the table show the variation of the coefficients of each variable in the
equation as the control variables are introduced one by one. Considering the heteroscedasticity
problem that may exist in the cross-sectional data, the model adds the heteroscedastic White test. It
can be seen that the above models all reject the null hypothesis of the homoscedasticity, which means
the heteroscedasticity exists. Models 2, 4, 6, and 8 show results after using heteroscedastic robust
standard error estimation. It can be seen that the coefficients are unchanged, the standard errors are
changed, and the significance of the relevant variables has changed to various degrees. Meanwhile, it
can be seen from the F values in the table that all the models are overall significant. Observing models
1-8, one can find that the passing rate of s1 has significant positive correlation in all equations with the
tranum, while the passing rate of s2 is significantly weaker, as it is negatively correlated with the
tranum in models 4 and 6, and is not significant in other models. Taking Model 4 as an example, under
the precondition that the control variables do not change, a 1% increase in the passing rate of s1 will
lead to an increase of road traffic accidents by 48.9392 per year, and a 1% in the passing rate of s2 will
reduce road traffic accidents by 26.2851 per year.
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
8
The control variables of per capita area of city road (cityroad), per capita GDP (agdp) and road
infrastructure streetlights (light) are significantly positively correlated with the number of traffic
accidents (tranum). The empirical results of the per capita area of city road (cityroad) are different
from the theoretical expectations, probably due to the fact that the current speed of road construction
in China lags behind that of vehicles’ launch in the market, which means the construction of road
infrastructure lags behind the progress of motorization. The correlation between the per capita GDP
(agdp) and the number of traffic accidents (tranums) meets theoretical expectations, indicating that as
people's living standards improve, the increase in their travel will lead to an increase in traffic
accidents to a certain extent. The number of streetlights (light) reflects the factors leading to the
increase in traffic accidents as mentioned in the theory part of this paper.
The population factors including the urban population density (citymidu) and number of city
travellers (citytraveller) are significant in the models with heteroscedasticity and are not significant
after being adjusted. Therefore, it can be said that the population factors are not the main factors
causing traffic accidents during this period.
3.3. Quantile Regression
The data of traffic accidents cited in the paper are those of different cities in China, including large
cities such as Beijing, Shanghai and Guangzhou, medium-sized cities such as Ningbo, Jinan, Hefei and
Wuhan, as well as a larger number of small-sized cities. Therefore, the distributions of number of
traffic accidents vary. Figure 3 shows the distribution of tranum of the 100 cities in the sampled data.
Figure 3 Number of Traffic Accidents (tranum)
The horizontal axis of the above figure represents the number of traffic accidents (tranums), and the
vertical axis represents the frequency, i.e. the number of cases falling within the range of number of
traffic accidents. It can be seen from the figure that the distribution of number of traffic accidents is
extremely uneven. There are about 60 cities whose tranum is no more than 500, about 20 cities whose
tranum is within the range of 500-1000, and a few cities 1000-3000. Outliers exist around 5000.
Meanwhile, combined with the linear trend graphs of s1, s2 and tranum in Figure 2, the existence of
outliers can also be found. Therefore, it is necessary to use quantile regression, which can avoid the
impact of outliers on the regression results and bring the more robust conclusion. Considering the fact
that it is found in the previous cross-sectional analysis of the paper that the population factors are not
significant during this period, the control variables of quantile regression here do not include the urban
population density (citymidu) and the number of city travellers (citytraveller). The specific results are
shown in Table 4.
Table 4. Results of tranum Quantile Regression
Variables
Model 1
1/10 Quantile
Model 2
3/10 Quantile
Model 3
1/2 Quantile
Model 4
7/10 Quantile
Model 5
9/10 Quantile
Constant
Terms
-184.3458
(512.7731)
-835.5702
(924.9161)
-1147.6740
(720.6238)
-1686.6100**
(831.8178)
-3148.5030*
(1612.2360)
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
9
s1
4.4579
(6.5429)
15.2396*
(8.8867)
21.6746**
(8.7983)
24.2232**
(9.5092)
22.8664
(18.4232)
s2
-3.3111
(5.7787)
-3.5510
(6.6738)
-8.9157
(7.1846)
-5.9884
(8.3244)
32.1766
(22.8293)
cityroad
0.4239
(5.5411)
-7.4568
(6.9461)
-6.5742
(6.6380)
-5.9884
(8.3244)
6.4156
(11.5395)
agdp
-0.0002
(0.0014)
0.0021
(0.0023)
0.0004
(0.0023)
0.0009
(0.0037)
0.0025
(0.0065)
light
0.0020*
(0.0010)
0.0027
(0.0024)
0.0086***
(0.0024)
0.0090***
(0.0019)
0.0145***
(0.0032)
Note: "***", "**", and "*" correspond respectively to the significance level of p<0.01,
<0.05, and <0.1, and the value in the brackets are standard error of bootstrap method.
Observing the passing rate of subject 1 in Table 4, one can see that with the increase of quantiles,
the quantile regression coefficient of s1 develops from insignificant, to significantly positive, then to
insignificant, and is gradually increasing at 3/10, 1/2, and 7/10 quantiles. It shows that the passing rate
of s1 is not a major factor in the regions with low or high number of traffic accidents, but is a major
factor in the regions with moderate number of traffic accidents. The quantile regression coefficient of
the passing rate of s2 is not significant in all models, similar to the weak significance of s2 passing rate
in the cross-sectional analysis.
The control variables per capita area of city road (cityroad) and per capita GDP (agtp) are not
significant. The number of road infrastructure streetlights (light) has a significant positive correlation
with the number of traffic accidents, and it shows a trend of regression coefficient increase with the
increase of the quantile. The conclusion is consistent with the previous cross-sectional analysis.
In summary, the quantile regression results are basically consistent with the conclusions of the
cross-sectional analysis. However, what is contrary to common sense in its conclusion is the
correlation between the passing rate of s1 and the number of traffic accidents (tranum). In theory,
higher passing rate of subject 1 indicates the higher quality of driver training provided by driving
schools, which is supposed to lead to lower number of traffic accidents, but the conclusion is the
opposite. Given the fact that the above analysis is carried out from a static perspective, this paper then
explores the correlation between the passing rate of s1, the passing rate of s2 and the number of traffic
accidents (tranum) from the dynamic perspective based on the panel data, striving to shed some light
on the the impact of the driver training quality on the road traffic safety.
3.4. Panel Analysis on the Impact of Driver Training Quality on the Road Traffic Safety
Panel analysis should take into account the availability of data in specific time spans. First of all, the
standard records of traffic accidents in China started around 2013. Secondly, the earliest records of
passing rates of s1 and s2 that are available for query over the Traffic Safety Integrated Service and
Management Platform of the Ministry of Public Security also date back to 2013. Therefore,
considering the above objective conditions, this paper selects the panel data of 18 cities from 2013 to
2016 for analysis, namely, Beijing, Tianjin, Taiyuan, Hohhot, Baotou, Shenyang, Harbin, Shanghai,
Hefei, Nanchang, Jinan, Wuhan, Guangzhou, Haikou, Chongqing, Chengdu, Xi'an and Yinchuan,
including provincial capitals, municipalities directly under the Central Government, and first-tier cities.
The data is representative to some extent.
The variable section is consistent with the previous ones cited in this paper to make sure the
conclusions are comparable. Since tranum data of 2016 has not been published, this paper uses linear
interpolation
2
to seek the data. Meanwhile, the data of the passing rates of s1 and s2 of the above-
mentioned cities are reached by calculating the mean values of all the driving schools in the targeted
regions in different years. For cities with no records of the passing rate of 2013, the linear
2
[Assuming the time series a1, a2, a3 represents from the past to the present, as a1 and a3 are known, then the
estimated value of a2 can be expressed as (a1+a3)/2.]
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
10
interpolation method is also used. The following table shows the regression results of the panel fixed
effect model for the impact of driver training quality on the road traffic safety.
Table 5. Results of the Panel Fixed Effect Model
Variables
Model 1
Model 2
Model 3
Model 4
Constant
Terms
5481.2930***
(771.6312)
7403.1670***
(1593.2910)
7384.5590***
(1612.2250)
7468.5780***
(1583.0880)
s1
-29.5081***
(8.7274)
-35.6933***
(9.7527)
-35.4123***
(9.9788)
-34.2640***
(9.8168)
s2
-15.6556
(9.3857)
-17.8641*
(9.4438)
-17.6464*
(9.6170)
-11.0542
(10.1997)
agdp
-0.0135
(0.0098)
-0.0128
(0.0106)
-0.0128
(0.0104)
light
-0.0003
(0.0019)
-0.0007
(0.0019)
cityroad
-47.3078*
(27.7433)
city-
effect
33.0700***
33.7500***
26.6700***
24.7700***
F-sta
30.2000***
29.2800***
27.3500***
27.2400***
Note: "***", "**", and "*" correspond respectively to the significance level of p<0.01,
<0.05, and <0.1, and the value in the brackets are standard error.
It can be seen from Table 5 that in the period from 2013 to 2016, the passing rate of s1 had
significantly negative correlation with the number traffic accidents (tranum). It indicates that from the
perspective of dynamic characteristics of the data, as the s1 increases, the occurrence of road traffic
accidents will decrease. The passing rate of s2 was also negatively correlated with tranum, but the
overall significance was weak. The model tests the individual city effect (city-effect) and the overall
significance of the equation (F-sta), which turn out to be good.
The control variables per capita GDP (agdp) and the number of streetlights (light) were not
significant during this period, probably due to the short time span. Considering it it not the focus of the
research, the paper will not give much analysis of it. The per capita road area (cityroad) was
negatively correlated with the tranum, indicating that the increase of the per capita road area can help
improve road traffic, thereby reducing traffic accidents.
However, why is the conclusion from s1 passing rate of Table 5 opposite to that of the cross-
sectional analysis? From a common point of view, the conclusions of the panel analysis are more
reasonable, i.e., the higher the s1 passing rate (indicating better quality of the driving schools and the
drivers’ better knowledge of road traffic) would indicate less probability of traffic accidents. The
conclusions of the cross-sectional analysis show that the higher s1 (i.e. the higher training quality)
would indicate the growth in traffic accidents. The reason is that as the paper analyzes the variable
characteristics of the panel data and the cross-sectional data, the former accommodates the time span
of the cross-sectional data and explores the changes of the cross-sectional individuals from the
dynamic perspective, while the latter explores the commonalities among different individual behaviors
in the same period of time at the static level.
Specifically, the cross-sectional analysis in the paper examines the impact of driver training quality
on road traffic safety in 100 cities across the country in 2015 alone. It is a relatively static analysis.
The positive correlation between s1 and tranum in the year can be explained by the fact that currently
some students failing to meet the requirements can still pass the s1 test (many just log in and remain
online without actually studying the contents on the required website or do not give much attention to
studying), which leads to the phenomenon that with the increase of s1, the probability of traffic
accidents increases due to drivers’ poor command of traffic rules and lack of safety awareness. When
exploring the correlation between s1 and tranum dynamically based on the panel data, the negative
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
11
correlation between the two shows that with the country's growing attention to the driver training
industry, especially the quality of training, the true level of s1 has risen, thereby reducing the
probability of traffic accidents.
4. Conclusion and Policy Recommendations
The paper first analyzes the impact of the driver training quality of the industry on road traffic safety
based on the panel data of 100 cities across the country in 2015. In the analysis, it is found that the
existence of outliers in the data is likely to affect the analysis results. Therefore, the quantile model
analysis is adopted and then enriches the conclusions of this paper from the perspective of panel data.
It can be said that the difference of samples is a necessary condition for ensuring the results of
significant analysis, and an important guarantee for comprehensively exploring the correlation among
variables. It is concluded that the correlation between the passing rate of s2 and road traffic safety is
relatively robust, as the improvement of s2 can indeed reduce the occurrence of traffic accidents. The
passing rate of s1 and road traffic safety shows a positive correlation in the relatively static cross-
sectional period, and a negative correlation in the panel analysis with time series characteristics,
reflecting that the true level of s1 is steadily increasing, but some mismatch between the s1 and the
true practice still exists.
At present, there are few empirical studies on the correlation between the driver training industry
and road traffic safety. One of the causes is that the related road traffic data is difficult to obtain,
which also indicates the people used to neglect the industry and road traffic safety. The successive
issuing of the 12th Five-Year Plan on Road Traffic Safety and the 13th Five-Year Plan on Road
Traffic Safety demonstrates the determination of the CPC Central Committee and the State Council on
promoting road traffic safety. But we need to be aware of the fact that the improvement of road traffic
takes time and requires unremitting efforts.
Acknowledgments
This paper is supported by the National Key R&D Program Product Quality Process Measurement
Analysis and Improvement Technical Standards Research Project Item 5 “Development of Quality
Basic Capability Evaluation Technical Standards” (Subject No. 2017YFF0206505).
References
[1] O'Malley P M, Johnston L D. Drinking and driving among US high school seniors, 1984-1997.
[J]. American Journal of Public Health, 1999, 89(5):678-684.
[2] Eensoo D, Paaver M, Harro J. Drunk Driving Among Novice Drivers, Possible Prevention
with Additional Psychological Module in Driving School Curriculum[J]. Association for
the Advancement of Automotive Medicine, 2011:283-291.
[3] Chakrabarty, Neelima Shukla, Anuradha Singh, H. Shokeen, Nancy Driver Training: An
Effective Tool for Improving Road Safety in India[J]. Journal of Engineering &
Technology. 2012, 2(2):113-117.
[4] Živković S, Nikolić V, Markič M. Influence of professional drivers' personality traits on road
traffic safety: case study[J]. International Journal of Injury Control & Safety Promotion,
2015, 22(2):100
[5] Kureckova V, Gabrhel V, Zamecnik P, et al. First aid as an important traffic safety factor –,
evaluation of the experience – based training[J]. European Transport Research Review,
2017, 9(1):5.
[6] Liu Junli, Poor Function of Driving Schools and Mismatch between Driving Practice and
Exam[J]. Transport Business China, 2013, (06):63-65.
[7] Zhang Luo, Study on the Efforts of A Driving School of Changsha to Improve Training
Quality[D]. Journal of Changsha University of Science & Technology, 2013
ICTETS 2019
IOP Conf. Series: Materials Science and Engineering 688 (2019) 055010
IOP Publishing
doi:10.1088/1757-899X/688/5/055010
12
[8] Liu Guangping, Ding Limin, Wang Jinfeng, Views on the Campaign to Promote Traffic
Safety Awareness[J]. Journal of People’s Public Security of China Science and
Technology, 2015, 21(04):50-54. [2017-09-11].
[9] Cai Yang, He Jinjiang, Big Data Models and Analysis of Motor Vehicle Road Traffic
Accidents[J]. Statistics and Management, 2016, (10):32-34.
[10] Luo Jiexiong, Suggestions on Enhancing Drivers’ Education on Safety[J]. Road Traffic
Management, 2017, (07):47. [2017-09-11].
[11] Zhang Chunyi, “Letting the Pedestrians Cross the Road First” Should be Incorporated in the
Driving Exam[N]. Changchun Daily, 2017-07-04(006).
[12] Yang Kaixin, Problems and Solutions in the Cost Management of Driving Schools[J]. Assets
and Finances in Administration and Institution, 2015, (30):29-30. [2017-09-11].